PROMPT   = What you want the model to do (task, role, format, constraints).
CONTEXT  = Information needed to make the task possible (code, data, contract, versions).
TOOL     = The means of finding the truth (Search, Cloud Browser, Python, Data plugin).

Translate naturally into English: [text]

1. ROLE AND GOAL    = The required expertise and the exact desired outcome.
2. CONTEXT          = Technical environment, versions, jurisdiction, market, and data.
3. INPUT MATERIAL   = Code, text, a table, or error log, separated with clear delimiters.
4. CONSTRAINTS      = What the model must NEVER assume or do.
5. OUTPUT FORMAT    = A table, JSON schema, bullet points, or management summary.
6. VERIFICATION / QA = Accuracy criteria, test cases, and checks for counter-evidence.

[BASE PROMPT]       -> Defines the core goal and format.
+ [DOMAIN MODULE]    -> Adds field-specific details (for example, Next.js 16, Serbian labor law, IFRS).
+ [FRESHNESS MODULE] -> Requires search and sets a date filter.
+ [SOURCE MODULE]    -> Defines source hierarchy (official documentation before forums).
+ [QA MODULE]        -> Defines what to check before final delivery.

Analyze the following non-disclosure agreement (NDA).

JURISDICTION: Serbia
RELEVANT DATE: September 2026

DOCUMENT TO ANALYZE:
"""
[Paste the agreement text here]
"""

Find clauses that:
1. impose an unlimited confidentiality obligation;
2. define disproportionate contractual penalties;
3. fail to clearly define what is NOT considered confidential information.

LEVEL 1: PRIMARY OFFICIAL SOURCES
         (Official institutional websites, laws in official gazettes,
         official software documentation, SEC financial filings, manufacturer specifications)
         ↓
LEVEL 2: PROFESSIONAL AND PEER-REVIEWED ACADEMIC SOURCES
         (Peer-reviewed papers, IEEE/ACM libraries, official technical standards such as ISO/RFC,
         medical best-practice guidelines)
         ↓
LEVEL 3: REPUTABLE SPECIALIST MEDIA
         (Reuters, Bloomberg, specialist technology and industry publications)
         ↓
LEVEL 4: SECONDARY AND AGGREGATOR SOURCES
         (Blogs, third-party tutorials, news aggregators)
         ↓
LEVEL 5: FORUMS AND SOCIAL NETWORKS
         (Reddit, Twitter/X, forums: useful only as signals of a new bug, never as proof of fact)

You found arguments in support of claim X.
Now run a new search with the sole goal of finding the strongest peer-reviewed studies
or official data that disprove claim X.
Show the direct conflict in the data and methodology.

1. Is the goal precise enough, or does the model have to guess what you mean?
2. Is key input context missing, such as a framework version, OS, or source text?
3. Are you using the wrong ChatGPT mode, such as regular Chat instead of Search for prices?
4. Did the task require fresh web sources when Search was not explicitly requested?
5. Does the task require processing the entire file rather than an excerpt, using Library or Projects?
6. Did you forget to specify the jurisdiction or geographic market?
7. Did you specify the exact date range you care about?
8. Is the requested format too broad, such as "write everything about this" instead of specifying table columns?
9. Does the prompt contain contradictory instructions?
10. Are you asking the model to calculate something that deterministic Python code should handle?
11. Could changing the query language to English, or to the source language, improve the results?
12. Is the task too large for one prompt and should it be split into two phases, analysis and then synthesis?
13. Is the model affected by confirmation bias because the prompt suggests the answer you want?
14. Is a negative constraint missing, explaining what the model explicitly must not do?
15. Is a second independent pass needed for fact-checking and red teaming?

TASK:
[Describe exactly what you want done]

INPUT DATA / CONTEXT:
"""
[Paste text, code, a table, or a specification]
"""

CONSTRAINTS:
- Do not assume facts that are not included in the input.
- If information is missing or uncertain, label it [UNKNOWN] instead of guessing.
- For claims that depend on current conditions, check official primary sources.

OUTPUT FORMAT:
1. Summary of findings, no more than 3 sentences;
2. Detailed, structured analysis by item;
3. Identified risks and recommendations for next steps.

RESEARCH QUESTION:
[Precise question about a topic, company, product, or phenomenon]

GEOGRAPHIC SCOPE AND TIME PERIOD:
[For example, the Serbian or EU market, 2025-2026]

SOURCE HIERARCHY:
1. Official primary sources, such as institutions, laws, and company financial reports;
2. Peer-reviewed analysis and industry research;
3. Reputable business media.
Ignore uncited forums and third-party blogs.

VERIFICATION REQUIREMENTS:
- If older and newer sources conflict, show both and explain the difference.
- For every key figure, provide the exact source name and publication date.
- End with counter-evidence that challenges the leading thesis.

RESEARCH GOAL:
[A comprehensive analysis of a topic, technology, or market segment]

KEY DIMENSIONS TO COVER:
- The current technological or legal landscape and the most important changes in 2026;
- Quantitative indicators, including market size, performance, costs, and benchmark data;
- Leading players, their strategies, and how they differ;
- Main risks, regulatory barriers, and known practical problems;
- Technology or market outlook for the next three years.

DOCUMENTATION:
Provide references to primary documents for each section.
Clearly separate confirmed facts from analytical estimates and forecasts.

PROBLEM:
[Describe the fault, error, or unexpected behavior]

ENVIRONMENT / SYSTEM:
[Device model, operating system, software versions, operating conditions]

SYMPTOMS AND ERROR CODES:
"""
[Logs, stack trace, error codes, behavior before the failure]
"""

WHAT HAS ALREADY BEEN TRIED:
[List steps that did not resolve it]

DIAGNOSTIC REQUIREMENTS:
1. Create a fault tree, ordered from the most likely cause to the least likely;
2. For each cause, state:
   - Why it is suspect based on the symptoms provided;
   - The cheapest and safest test to confirm or rule it out;
3. Do not recommend random part replacement or reinstallation before confirming the fault;
4. Clearly mark where it is no longer safe to continue and an authorized service provider must be involved.

PROJECT:
[Name and purpose of the project]

CURRENT STATE AND COMPLETED WORK:
[Brief overview of completed sections]

NEXT GOAL:
[The specific feature, chapter, or module we are working on]

CONTINUITY RULES:
- Follow the existing architecture and conventions established in earlier steps;
- Before introducing new libraries or approaches, check compatibility with the existing code;
- Make every change with a minimal, reviewable diff;
- After a change, define a test to confirm the feature works and no regression was introduced.

1. CHOOSE A TEMPLATE       -> Find the closest prompt in the library below.
          ↓
2. ADD YOUR CONTEXT        -> Enter your actual versions, data, code, or contracts.
          ↓
3. ADAPT IT TO THE FIELD   -> Add domain-specific rules and prohibited assumptions.
          ↓
4. CHECK WHAT IS MISSING   -> Run the meta-prompt below to identify blind spots.
          ↓
5. CHOOSE THE RIGHT MODE   -> Select Chat, Search, Deep Research, or Work as needed.
          ↓
6. RUN THE PROMPT          -> Get a structured and precise initial result.
          ↓
7. PERFORM A FINAL QA      -> Verify numbers and sources, then test the solution in practice.

Before running the following prompt, analyze the PROMPT ITSELF.

Do not solve the underlying task yet.

CONTEXT FOR MY SPECIFIC CASE:

FIELD:
[Enter the field, such as frontend web development, Serbian labor law, cardiology, or financial due diligence]

SPECIFIC GOAL:
[Enter the outcome you want to achieve]

ENVIRONMENT / JURISDICTION / MARKET / VERSION / DEVICE:
[Enter the relevant technical or regulatory specifications]

INPUT DATA AVAILABLE TO ME:
"""
[...]
"""

CONSTRAINTS:
[...]

PROMPT I WANT TO USE:
"""
[Paste the prompt from the library that you want to adapt]
"""

YOUR TASK:
Improve this prompt specifically for my case before it is run.

Check whether the prompt is missing:
1. Important field-specific information;
2. More precise input data and technical parameters;
3. The exact technology, library, or product model version;
4. The applicable jurisdiction or exact date range;
5. Local market specifics, currency, taxes, or customs;
6. Clear metric definitions and mathematical formulas;
7. A hierarchy of authoritative sources, showing which sources take precedence;
8. An explicit request for current web sources for changing information;
9. Safety and operational constraints, including what must never be done;
10. Relevant edge cases and known failure modes in the field;
11. A method for verifying the accuracy of the result;
12. A precise final answer format ready for direct use;
13. Information the model must never assume unless it is provided.

OPTIMIZATION RULES:
- Remove generic filler and phrases that do not change model behavior.
- Do not make the prompt longer just for length. Every word should improve precision.
- For time-sensitive information, require checks against primary sources.
- If research in another language, such as English for IT documentation or German for German law, would significantly improve source quality, add that instruction and require the final answer in Serbian.

RETURN THIS FORMAT:
1. IMPROVED PROMPT, in a separate code block ready to copy;
2. BRIEF EXPLANATION of exactly what changed and why it should produce a more reliable result.

Do not run the improved prompt yet. Wait for my confirmation.

I have the following technical problem:

[DESCRIBE THE PROBLEM]

ENVIRONMENT

OS:
[Windows / Linux / macOS]

Runtime:
[Node.js / Python / Java / .NET...]

Runtime version:
[...]

Framework:
[...]

Framework version:
[...]

Relevant libraries:
[...]

Deployment:
[Vercel / AWS / Railway / Docker / local...]

Database:
[...]

ERROR

"""
[ERROR / STACK TRACE / LOG]
"""

RELEVANT CODE

"""
[CODE]
"""

GOAL

I want to:
[describe the expected behavior]

REQUIRED RULES

If the answer depends on the version of a framework, library, runtime, or API, check the current official documentation before proposing a solution.

Source priority:

1. official documentation;
2. migration guide;
3. release notes / changelog;
4. official GitHub repository;
5. official issue tracker;
6. only then Stack Overflow, blogs, and community discussions.

Do not immediately generate a large patch.

First:

1. explain what is most likely happening;
2. identify possible root causes;
3. cite evidence from the code or log for each one;
4. suggest the smallest test that could confirm the hypothesis;
5. only then suggest a minimal correction.

Organize the final result as:

ROOT CAUSE
EVIDENCE
FIX
PATCH
HOW TO TEST
REGRESSION RISKS
ROLLBACK

Do not use deprecated APIs when a current replacement exists.
Do not invent functions, configuration options, or APIs.

Act as a debugging partner.

Problem:

[PROBLEM]

Symptoms:

[SYMPTOMS]

Last known working version:

[IF KNOWN]

Recent changes:

[CHANGES]

Log:

"""
[LOG]
"""

Do not change code at random.

Use hypothesis-driven debugging.

Create a table:

HYPOTHESIS
EVIDENCE FOR
EVIDENCE AGAINST
TEST
EXPECTED RESULT

Start with hypotheses that best explain all the symptoms at once.

Prefer tests that can quickly eliminate several possible causes.

If the problem is version-specific, check the current documentation.

Once you identify the root cause:

- explain why the error occurs;
- explain why it may have worked before;
- provide a minimal patch;
- write a regression test;
- list possible side effects.

Do not redesign the architecture if a small fix is enough.

Perform a deep audit of this repository:

[REPOSITORY / ATTACHED PROJECT]

Do not suggest changes immediately.

PHASE ONE - MAPPING

First understand the project:

- entry points;
- module boundaries;
- request and data flow;
- authentication;
- authorization;
- persistence;
- networking;
- background jobs;
- caching;
- queues;
- filesystem;
- external APIs;
- error handling;
- retry logic;
- cancellation;
- recovery;
- test architecture.

Create a short architecture map before the audit.

PHASE TWO - AUDIT

Look for problems in this order:

P0 - CRITICAL

- data loss;
- privilege escalation;
- authentication bypass;
- authorization bypass;
- permanent corruption;
- remote code execution;
- secrets exposure.

P1 - HIGH

- race conditions;
- stale state;
- duplicate execution;
- incorrect retries;
- partial transaction failures;
- crash recovery;
- unsafe cancellation;
- deadlocks;
- lifecycle bugs;
- resource leaks;
- incorrect fail-open behavior.

P2 - MEDIUM

- bounds;
- validation;
- edge cases;
- performance;
- scalability;
- maintainability;
- observability.

P3 - LOW

- cosmetics;
- naming;
- minor refactoring.

For every finding, provide:

ID
SEVERITY
FILE
FUNCTION / LOCATION
PROBLEM
REAL FAILURE SCENARIO
WHY THE CURRENT CODE FAILS
MINIMAL FIX
REGRESSION TEST

IMPORTANT:

Do not report theoretical problems without a real execution path.

Do not invent issues just to make the audit look more detailed.

If a subsystem appears correct, say that you found no concrete issue.

Now forget the desire to defend the previous solution.

Act as an independent senior reviewer trying to prevent a production incident.

Audit the previous patch.

Actively try to break it.

Pay particular attention to:

- race conditions;
- TOCTOU issues;
- stale references;
- partial failures;
- rollback;
- process restart;
- crashes during writes;
- duplicate requests;
- retries after partial success;
- timeouts;
- cancellation;
- resource cleanup;
- concurrent delete/update;
- authentication state changes;
- permission changes;
- malformed inputs;
- extreme input sizes;
- integer overflow;
- null/undefined;
- backward compatibility.

For every issue you find, describe a concrete scenario:

1. initial state;
2. sequence of events;
3. where the patch fails;
4. consequence;
5. correction;
6. regression test.

If you find no new issue, list the classes of errors you checked.

Assess the project architecture before we refactor anything.

PROJECT:

[DESCRIPTION / REPOSITORY]

GOAL:

[what we want to improve]

CONSTRAINTS:

- I do not want a rewrite without a strong reason;
- backward compatibility matters;
- existing users must not lose data;
- deployment must remain stable.

First analyze the current architecture.

For each subsystem, describe:

RESPONSIBILITY
DEPENDENCIES
STATE OWNERSHIP
FAILURE MODES
SCALING LIMITS
TESTABILITY

Then identify:

- tight coupling;
- unclear ownership;
- duplicated state;
- hidden side effects;
- circular dependencies;
- weak boundaries;
- fragile initialization;
- shutdown/recovery problems.

Separate each proposal into:

MUST FIX
SHOULD FIX
NICE TO HAVE

Do not suggest a new technology just because it is more popular.

For every architectural change, explain:

- the problem it solves;
- migration cost;
- risk;
- rollback plan;
- how we will prove the change worked.

I want to upgrade:

[DEPENDENCY / FRAMEWORK]

From version:

[OLD]

To version:

[NEW]

Stack:

[STACK]

Before suggesting changes, check:

- official migration guide;
- breaking changes;
- deprecated APIs;
- removed APIs;
- changes to default behavior;
- security changes;
- configuration changes;
- peer dependency requirements;
- runtime requirements.

Then review my code and find every place affected by the migration.

Create a plan:

PHASE 1 - prerequisites
PHASE 2 - dependency changes
PHASE 3 - code migration
PHASE 4 - tests
PHASE 5 - deployment
PHASE 6 - rollback

For each change, provide:

OLD
NEW
WHY
BREAKING RISK
TEST

Do not assume an example from older documentation is still valid.

Perform a performance audit.

SYSTEM:

[DESCRIPTION]

PROBLEM:
[for example, API p95 latency is 2.4 seconds]

Do not optimize blindly.

First identify where the time is spent.

Analyze:

CPU
MEMORY
DATABASE
NETWORK
SERIALIZATION
FILESYSTEM
EXTERNAL APIS
LOCKING
QUEUEING
CACHE
RENDERING

Measure before optimizing.

For each potential bottleneck, provide:

EVIDENCE
HOW TO MEASURE
EXPECTED IMPACT
IMPLEMENTATION COST
RISK

Check in particular for:

- N+1 queries;
- unnecessary sequential awaits;
- missing indexes;
- oversized payloads;
- repeated parsing;
- redundant API calls;
- expensive React re-renders;
- cache misses;
- connection pool exhaustion;
- memory growth;
- large bundle size.

Do not suggest caching without explaining the invalidation strategy.

Rank the final optimizations by:

impact / complexity / risk.

Analyze the following database/query problem.

SCHEMA:

"""
[SCHEMA]
"""

QUERY:

"""
[SQL]
"""

GOAL:

[what the query should do]

Check correctness before optimizing.

Analyze in particular:

- table grain;
- primary keys;
- foreign keys;
- join cardinality;
- one-to-many relationships;
- many-to-many relationships;
- duplicate row multiplication;
- NULL semantics;
- transaction boundaries;
- locking;
- isolation level;
- indexes.

For each JOIN, state:

LEFT TABLE GRAIN
RIGHT TABLE GRAIN
JOIN KEY
EXPECTED CARDINALITY
DUPLICATION RISK

If the query calculates a KPI, first define:

numerator
denominator
filters
time window

Then check whether the JOIN changes the row count.

Only suggest performance optimizations after the correctness analysis.

For each index proposal, explain:

- which query uses the index;
- column order;
- write performance trade-off;
- whether the index is likely redundant.

Review this API design.

ENDPOINTS:

[API]

Analyze:

- resource naming;
- HTTP methods;
- status codes;
- idempotency;
- pagination;
- filtering;
- sorting;
- validation;
- error model;
- authentication;
- authorization;
- rate limiting;
- retries;
- timeouts;
- versioning;
- backward compatibility.

Pay special attention to retry safety.

For each endpoint, answer:

Is it idempotent?
What happens if the client retries the request?
What happens if the server completes the operation but the response never arrives?
How do we prevent duplicate side effects?

Consider idempotency keys for write operations where appropriate.

Suggest changes only when they solve a concrete problem.

Perform a deep audit of the authentication and authorization system.

AUTH FLOW:

[DESCRIPTION]

Check:

AUTHENTICATION

- credential handling;
- password hashing;
- MFA;
- session creation;
- token issuance;
- token expiration;
- refresh;
- logout;
- revocation;
- session fixation.

AUTHORIZATION

- role checks;
- permission checks;
- object-level authorization;
- tenant boundaries;
- admin operations;
- horizontal privilege escalation;
- vertical privilege escalation.

Check IDOR/BOLA in particular.

For every endpoint that accesses an object, ask:

"Do we only check that the user is signed in, or do we also check that they are allowed to access this specific object?"

For each finding, provide a concrete attack or failure scenario and a safe correction.

Do not treat a frontend check as a security control.

Analyze this project's test suite.

I care about more than the code coverage percentage.

Map tests to real failure modes.

Identify critical flows:

[flows]

For each, list:

HAPPY PATH
INVALID INPUT
BOUNDARY
CONCURRENCY
TIMEOUT
CANCELLATION
RETRY
PARTIAL FAILURE
CRASH/RESTART
RECOVERY

Find features that appear covered by tests but have no test for their most dangerous failure scenario.

Prioritize tests that prevent:

data loss;
security regression;
silent corruption;
double execution;
stale state;
incorrect recovery.

Do not suggest tests just to increase the coverage percentage.

Analyze the frontend.

Stack:

React:
[version]

Next.js:
[version]

Routing:
[App Router / other]

State:
[...]

Data fetching:
[...]

Check the current documentation for the specified versions.

Audit:

- Server vs. Client Components;
- hydration;
- data fetching;
- caching;
- invalidation;
- loading states;
- error states;
- optimistic updates;
- stale UI;
- form submission;
- duplicate submissions;
- accessibility;
- focus handling;
- keyboard navigation;
- responsive layout;
- bundle size.

Look in particular for problems where the UI says an operation succeeded before the backend confirms it.

For each problem, provide:

USER ACTION
EXPECTED
ACTUAL
ROOT CAUSE
FIX
TEST

Audit the backend with a focus on reliability.

Assume that:

- the process can crash at any time;
- the network can fail;
- an external API can time out;
- the same request can arrive twice;
- a database operation can partially succeed;
- the service can restart during a job.

For each write flow, check:

1. what is durable;
2. what exists only in memory;
3. what happens after a restart;
4. how already-completed work is detected;
5. whether a retry can create a duplicate;
6. whether partial failure leaves inconsistent state.

Look for:

idempotency;
transactions;
outbox/inbox patterns;
job ownership;
leases;
heartbeats;
recovery;
dead-letter handling;
assumptions about exactly-once execution.

Do not say "exactly once" without explaining how it is guaranteed.

Analyze the CI/CD and deployment pipeline.

Platform:
[GitHub Actions / GitLab CI / Vercel / AWS / Railway...]

Check:

- build reproducibility;
- dependency pinning;
- secrets;
- environment separation;
- preview vs. production;
- migrations;
- zero-downtime deployment;
- rollback;
- artifact integrity;
- caching;
- deployment concurrency.

Pay particular attention to database migrations.

Answer:

What happens if the old and new application versions run at the same time?

Is the schema change backward compatible during rollout?

Will an application rollback work if the migration has already run?

For each risk, suggest a safe deployment sequence.

PERFORM A TECHNICAL QA REVIEW BEFORE YOUR FINAL RESPONSE.

Check:

- whether the proposed API actually exists in the specified version;
- whether anything is deprecated;
- whether the code is conceptually compilable;
- whether an import is missing;
- whether the types match;
- whether async operations handle errors correctly;
- whether the patch introduces a race condition;
- whether it causes a regression;
- whether the change affects backward compatibility;
- whether the test is sufficient to catch the original bug.

Do not claim the problem is fixed just because the code looks correct.

Separate:

CONFIRMED
LIKELY
UNVERIFIED

Perform a deep DEFENSIVE security audit of the following application.

APPLICATION:

[DESCRIPTION]

STACK:

Frontend:
Backend:
Database:
Authentication:
Hosting:
CDN/WAF:
Storage:
External APIs:

ARCHITECTURE:

[DESCRIPTION OR REPOSITORY]

Before the analysis, map:

- internet-facing entry points;
- trust boundaries;
- authentication boundaries;
- authorization boundaries;
- privileged operations;
- user-controlled inputs;
- sensitive data;
- external integrations;
- secrets;
- persistence;
- background jobs.

Do not merely list OWASP categories.

For each potential issue, provide:

ID
SEVERITY
COMPONENT
TRUST BOUNDARY
PRECONDITIONS
ATTACK / FAILURE PATH
CURRENT CONTROL
WHY THE CONTROL FAILS
IMPACT
EVIDENCE
FIX
VERIFICATION TEST

Check in particular:

authentication;
authorization;
BOLA/IDOR;
session management;
CSRF;
XSS;
SSRF;
SQL/NoSQL injection;
command injection;
path traversal;
unsafe file uploads;
deserialization;
template injection;
open redirects;
CORS;
webhooks;
rate limits;
password reset;
email verification;
OAuth/OIDC flow;
secret exposure;
sensitive logging;
dependency risk;
error leakage.

Do not label an issue exploitable unless you can explain the conditions required to exploit it.

Separate:

CONFIRMED
LIKELY
THEORETICAL
NOT APPLICABLE

Audit the application against the relevant controls in a modern OWASP ASVS-style approach.

Do not blindly check every control.

First determine which areas apply to this system.

Analyze in particular:

- architecture;
- authentication;
- session management;
- access control;
- input validation;
- cryptography;
- error handling;
- data protection;
- communications;
- malicious code;
- business logic;
- files/resources;
- APIs/web services;
- configuration.

For each finding, provide:

CONTROL AREA
EXPECTED SECURITY PROPERTY
CURRENT IMPLEMENTATION
EVIDENCE
GAP
PRACTICAL IMPACT
REMEDIATION
TEST

If the control is already implemented correctly, mark it:

PASS

If there is not enough information:

UNVERIFIED

Do not mark FAIL just because the code needed to prove PASS was not provided.

Create a threat model for the system.

SYSTEM:

[DESCRIPTION]

USERS:

[user types]

COMPONENTS:

[...]

DATA:

[...]

EXTERNAL SERVICES:

[...]

First map:

ASSETS
ACTORS
ENTRY POINTS
TRUST BOUNDARIES
PRIVILEGED COMPONENTS
DATA FLOWS

Then analyze threats using STRIDE as a supporting framework:

Spoofing
Tampering
Repudiation
Information Disclosure
Denial of Service
Elevation of Privilege

Do not generate threats mechanically just because a category exists.

For each realistic threat, provide:

THREAT
ASSET
ATTACKER CAPABILITY
PRECONDITION
ATTACK PATH
EXISTING CONTROL
LIKELIHOOD FACTORS
IMPACT
MITIGATION
DETECTION
RESIDUAL RISK

Identify in particular where compromising one subsystem could enable lateral movement into another.

Audit the authentication system.

FLOW:

[DESCRIPTION]

PROVIDER:
[custom / Auth.js / Clerk / Cognito / Supabase / Keycloak...]

Check:

LOGIN
PASSWORD STORAGE
MFA
PASSWORD RESET
EMAIL VERIFICATION
SESSION CREATION
SESSION ROTATION
SESSION EXPIRATION
REFRESH TOKENS
LOGOUT
REVOCATION
DEVICE MANAGEMENT
OAUTH/OIDC
MAGIC LINKS

Check in particular:

- session fixation;
- token replay;
- refresh token reuse;
- token leakage through URLs, logs, or referrers;
- weak password-reset tokens;
- account enumeration;
- MFA bypass;
- stale sessions after a password change;
- logout that only clears frontend state;
- OAuth state/nonce/PKCE;
- redirect URI validation.

For each issue, describe a concrete sequence of events.

Do not assume JWTs are automatically safer than server-side sessions.

Assess the actual implementation.

Audit the authorization model.

ROLES:

[...]

RESOURCES:

[...]

TENANT MODEL:

[...]

For each endpoint and privileged action, check three levels:

1. Is the user authenticated?
2. Do they have the required permission or role?
3. Are they allowed to access this specific resource?

Look in particular for:

BOLA / IDOR;
horizontal privilege escalation;
vertical privilege escalation;
cross-tenant access;
admin-only operations;
ownership transfer bugs;
stale permissions;
deleted or deactivated users;
cached authorization decisions.

For each finding, provide a scenario:

USER A
RESOURCE B
REQUEST
CURRENT CHECK
MISSING CHECK
RESULT

Hiding something in the frontend is not an authorization control.

Treat client-provided tenantId, userId, and role values as untrusted data.

Help me analyze a security or operational incident.

Do NOT immediately conclude that it was an attack.

DATA:

Start time:
[...]

Symptoms:
[...]

Logs:
[...]

Metrics:
[...]

Changes/deployments:
[...]

Network data:
[...]

Security alerts:
[...]

First create a timeline.

Then separate the hypotheses:

ATTACK
SOFTWARE BUG
CONFIGURATION ERROR
RESOURCE EXHAUSTION
EXTERNAL PROVIDER FAILURE
USER ERROR
FALSE POSITIVE

For each hypothesis, provide:

EVIDENCE FOR
EVIDENCE AGAINST
MISSING DATA
NEXT TEST

Separate actions into:

CONTAINMENT
PRESERVATION OF EVIDENCE
ERADICATION
RECOVERY
POST-INCIDENT ACTIONS

Do not recommend deleting or reinstalling anything before considering whether that would destroy evidence we need.

Analyze the following HTTP/security logs DEFENSIVELY.

LOG:

"""
[LOG]
"""

Do not label every unusual request an attack.

For each pattern, determine:

SOURCE
TIMESTAMP
METHOD
PATH
STATUS
USER AGENT
REQUEST FREQUENCY
RESPONSE SIZE
LATENCY

Group similar requests.

Try to distinguish between:

normal crawler;
search engine bot;
vulnerability scanner;
credential stuffing;
brute force;
path enumeration;
WordPress scanning;
LFI/RFI probing;
SQLi probing;
XSS probing;
SSRF probing;
automated noise;
real exploitation attempt.

For each suspicious pattern, provide:

WHAT IT APPEARS TO TEST
EVIDENCE
DID IT SUCCEED?
RELEVANT SERVER RESPONSE
RISK
RECOMMENDED DEFENSIVE ACTION

Important:

A request that contains an exploit string is not proof that the exploit worked.

Look for evidence of successful impact on the system.

Perform a Linux server hardening audit.

DISTRIBUTION:
[...]

VERSION:
[...]

ROLE:
[web/db/VPN/reverse proxy...]

PUBLIC SERVICES:
[...]

USERS:
[...]

SSH CONFIG:
[...]

FIREWALL:
[...]

CONTAINERS:
[...]

First check current recommendations for the specified distribution and version.

Audit:

PATCH STATUS
USER ACCOUNTS
SSH
SUDO
FIREWALL
LISTENING PORTS
SYSTEMD SERVICES
FILE PERMISSIONS
SECRETS
LOGGING
AUDITING
TIME SYNC
AUTOMATIC SECURITY UPDATES
KERNEL SETTINGS
BACKUPS

For SSH, check in particular:

password authentication;
root login;
key-based access;
MFA where applicable;
AllowUsers/AllowGroups;
legacy algorithms.

Do not suggest disabling services without first establishing whether the application uses them.

For every change, provide:

SECURITY BENEFIT
BREAKAGE RISK
HOW TO TEST
ROLLBACK

Audit the Docker deployment.

FILES:

Dockerfile:
"""
[Dockerfile]
"""

Compose file:
"""
[docker-compose.yml]
"""

Check:

BASE IMAGE
IMAGE VERSION PINNING
IMAGE ORIGIN
ROOT/NON-ROOT
CAPABILITIES
PRIVILEGED MODE
HOST MOUNTS
DOCKER SOCKET
SECRETS
ENVIRONMENT VARIABLES
NETWORK EXPOSURE
READ-ONLY FILESYSTEM
TMPFS
RESOURCE LIMITS
HEALTHCHECK
SIGNAL HANDLING
BUILD SECRETS
MULTI-STAGE BUILD
DEPENDENCIES

Check in particular whether the container:

- runs as root without needing to;
- uses privileged mode;
- mounts /var/run/docker.sock;
- mounts sensitive host directories;
- has unnecessary Linux capabilities;
- includes build tools in the production image;
- contains a secret in an image layer;
- exposes a port that does not need to be public.

For each finding, provide:

CURRENT CONFIG
RISK
MINIMAL HARDENING
COMPATIBILITY RISK
TEST

Perform a security and reliability audit of the Kubernetes workload.

KUBERNETES VERSION:
[...]

MANIFESTS / HELM:

"""
[CONFIG]
"""

Check:

Pod Security Standards;
service accounts;
RBAC;
network policies;
securityContext;
runAsNonRoot;
capabilities;
privilege escalation;
seccomp;
hostNetwork;
hostPID;
hostPath;
secrets;
config maps;
image sources;
image tags/digests;
resource requests/limits;
probes;
PDB;
namespace isolation.

For each workload, determine:

Does it actually require privileged access?

Can it use the Restricted Pod Security profile?

If not, explain exactly which function prevents that.

Check in particular for:

wildcard RBAC;
cluster-admin;
default service account;
automountServiceAccountToken;
cross-namespace access;
unrestricted egress;
host filesystem access.

Do not automatically move everything to Restricted if the application would stop working.

Suggest a migration plan:

AUDIT
WARN
ENFORCE

with a compatibility check.

Audit the Terraform configuration.

CLOUD:
[AWS/Azure/GCP/...]

TERRAFORM VERSION:
[...]

CODE:
[...]

First check:

VERSION PINNING
PROVIDERS
MODULE SOURCES
STATE BACKEND
STATE LOCKING
STATE ENCRYPTION
ACCESS CONTROL
SECRETS
IAM
PUBLIC EXPOSURE
NETWORK RULES
DATABASES
STORAGE
LOGGING
BACKUPS

Analyze Terraform state in particular.

Determine:

- where it is stored;
- who can access it;
- whether it contains sensitive data;
- whether it is encrypted;
- whether locking is enabled;
- whether it was accidentally committed.

For each resource, check:

Could a change cause replacement?

Could replacement cause data loss?

For each proposed change, state the Terraform effect:

IN-PLACE
RECREATE
UNKNOWN UNTIL PLAN

Do not assume that `sensitive = true` means the value never appears in state.

Before apply, require a review of:

terraform plan

and identify any unexpected destroy/recreate operations.

Perform a least-privilege audit of cloud IAM.

PROVIDER:
[AWS/Azure/GCP]

PRINCIPALS:
[...]

ROLES/POLICIES:
[...]

First map:

HUMAN USERS
SERVICE ACCOUNTS
WORKLOAD IDENTITIES
CI/CD IDENTITIES
ADMIN ACCOUNTS
BREAK-GLASS ACCOUNTS

For each identity, determine:

WHO/WHAT USES IT
TRUST POLICY
ASSIGNED PERMISSIONS
PERMISSIONS ACTUALLY REQUIRED
CREDENTIAL TYPE
CREDENTIAL LIFETIME

Look for:

wildcard actions;
wildcard resources;
privilege escalation paths;
pass-role/assume-role issues;
unused credentials;
long-lived access keys;
cross-account trust;
service account key files;
shared credentials.

Check in particular for permission combinations that enable escalation even if each permission looks harmless by itself.

Do not recommend removing permissions without identifying the workload that uses them.

Final classification:

KEEP
REDUCE
REMOVE
INVESTIGATE

Find every way the infrastructure is accessible from the internet.

CLOUD CONFIG:
[...]

Map:

public IPs;
load balancers;
CDN;
API gateways;
storage buckets;
databases;
SSH/RDP;
Kubernetes services;
serverless endpoints;
management interfaces.

For each public endpoint, determine:

SERVICE
PORT
PROTOCOL
AUTHENTICATION
EXPECTED AUDIENCE
WHY IT IS PUBLIC
SOURCE RESTRICTION
WAF/RATE LIMIT
TLS

Identify resources that are:

INTENTIONALLY PUBLIC
UNNECESSARILY PUBLIC
UNCERTAIN

Check management ports and databases in particular.

Do not automatically label HTTP port 80 a vulnerability if it is used only for a controlled redirect to HTTPS.

Analyze the network architecture.

DIAGRAM / DESCRIPTION:

[...]

Map the zones:

INTERNET
EDGE
PUBLIC SUBNET
APPLICATION
DATABASE
MANAGEMENT
CI/CD
VPN
THIRD-PARTY

For each communication flow, state:

SOURCE
DESTINATION
PORT
PROTOCOL
PURPOSE
AUTHENTICATION
ENCRYPTION
REQUIRED?

Look for:

flat network;
unnecessary east-west access;
0.0.0.0/0 rules;
overly broad security groups;
management interfaces exposed publicly;
unrestricted egress;
database exposure;
DNS dependency;
single points of failure.

Apply the principle:

default deny where practical.

However, do not block a communication flow until you have confirmed the application does not use it.

Audit the firewall rules.

RULES:

"""
[rules]
"""

For each rule, create:

SOURCE
DESTINATION
PROTOCOL
PORT
DIRECTION
PURPOSE
OWNER
LAST KNOWN USE

Look for:

ANY/ANY;
0.0.0.0/0;
::/0;
unused rules;
shadowed rules;
duplicate rules;
management ports exposed publicly;
temporary rules that became permanent.

Classify each rule:

REQUIRED
OVERLY BROAD
POSSIBLY UNUSED
DANGEROUS
UNKNOWN

If you suggest narrowing a rule, define how to check actual traffic before changing it.

Perform a deep DEFENSIVE security audit of the following application.

APPLICATION:

[DESCRIPTION]

STACK:

Frontend:
Backend:
Database:
Authentication:
Hosting:
CDN/WAF:
Storage:
External APIs:

ARCHITECTURE:

[DESCRIPTION OR REPOSITORY]

Before the analysis, map:

- internet-facing entry points;
- trust boundaries;
- authentication boundaries;
- authorization boundaries;
- privileged operations;
- user-controlled inputs;
- sensitive data;
- external integrations;
- secrets;
- persistence;
- background jobs.

Do not merely list OWASP categories.

For each potential issue, provide:

ID
SEVERITY
COMPONENT
TRUST BOUNDARY
PRECONDITIONS
ATTACK / FAILURE PATH
CURRENT CONTROL
WHY THE CONTROL FAILS
IMPACT
EVIDENCE
FIX
VERIFICATION TEST

Check in particular:

authentication;
authorization;
BOLA/IDOR;
session management;
CSRF;
XSS;
SSRF;
SQL/NoSQL injection;
command injection;
path traversal;
unsafe file uploads;
deserialization;
template injection;
open redirects;
CORS;
webhooks;
rate limits;
password reset;
email verification;
OAuth/OIDC flow;
secret exposure;
sensitive logging;
dependency risk;
error leakage.

Do not label an issue exploitable unless you can explain the conditions required to exploit it.

Separate:

CONFIRMED
LIKELY
THEORETICAL
NOT APPLICABLE

Audit the application against the relevant controls in a modern OWASP ASVS-style approach.

Do not blindly check every control.

First determine which areas apply to this system.

Analyze in particular:

- architecture;
- authentication;
- session management;
- access control;
- input validation;
- cryptography;
- error handling;
- data protection;
- communications;
- malicious code;
- business logic;
- files/resources;
- APIs/web services;
- configuration.

For each finding, provide:

CONTROL AREA
EXPECTED SECURITY PROPERTY
CURRENT IMPLEMENTATION
EVIDENCE
GAP
PRACTICAL IMPACT
REMEDIATION
TEST

If the control is already implemented correctly, mark it:

PASS

If there is not enough information:

UNVERIFIED

Do not mark FAIL just because the code needed to prove PASS was not provided.

Create a threat model for the system.

SYSTEM:

[DESCRIPTION]

USERS:

[user types]

COMPONENTS:

[...]

DATA:

[...]

EXTERNAL SERVICES:

[...]

First map:

ASSETS
ACTORS
ENTRY POINTS
TRUST BOUNDARIES
PRIVILEGED COMPONENTS
DATA FLOWS

Then analyze threats using STRIDE as a supporting framework:

Spoofing
Tampering
Repudiation
Information Disclosure
Denial of Service
Elevation of Privilege

Do not generate threats mechanically just because a category exists.

For each realistic threat, provide:

THREAT
ASSET
ATTACKER CAPABILITY
PRECONDITION
ATTACK PATH
EXISTING CONTROL
LIKELIHOOD FACTORS
IMPACT
MITIGATION
DETECTION
RESIDUAL RISK

Identify in particular where compromising one subsystem could enable lateral movement into another.

Audit the authentication system.

FLOW:

[DESCRIPTION]

PROVIDER:
[custom / Auth.js / Clerk / Cognito / Supabase / Keycloak...]

Check:

LOGIN
PASSWORD STORAGE
MFA
PASSWORD RESET
EMAIL VERIFICATION
SESSION CREATION
SESSION ROTATION
SESSION EXPIRATION
REFRESH TOKENS
LOGOUT
REVOCATION
DEVICE MANAGEMENT
OAUTH/OIDC
MAGIC LINKS

Check in particular:

- session fixation;
- token replay;
- refresh token reuse;
- token leakage through URLs, logs, or referrers;
- weak password-reset tokens;
- account enumeration;
- MFA bypass;
- stale sessions after a password change;
- logout that only clears frontend state;
- OAuth state/nonce/PKCE;
- redirect URI validation.

For each issue, describe a concrete sequence of events.

Do not assume JWTs are automatically safer than server-side sessions.

Assess the actual implementation.

Audit the authorization model.

ROLES:

[...]

RESOURCES:

[...]

TENANT MODEL:

[...]

For each endpoint and privileged action, check three levels:

1. Is the user authenticated?
2. Do they have the required permission or role?
3. Are they allowed to access this specific resource?

Look in particular for:

BOLA / IDOR;
horizontal privilege escalation;
vertical privilege escalation;
cross-tenant access;
admin-only operations;
ownership transfer bugs;
stale permissions;
deleted or deactivated users;
cached authorization decisions.

For each finding, provide a scenario:

USER A
RESOURCE B
REQUEST
CURRENT CHECK
MISSING CHECK
RESULT

Hiding something in the frontend is not an authorization control.

Treat client-provided tenantId, userId, and role values as untrusted data.

Help me analyze a security or operational incident.

Do NOT immediately conclude that it was an attack.

DATA:

Start time:
[...]

Symptoms:
[...]

Logs:
[...]

Metrics:
[...]

Changes/deployments:
[...]

Network data:
[...]

Security alerts:
[...]

First create a timeline.

Then separate the hypotheses:

ATTACK
SOFTWARE BUG
CONFIGURATION ERROR
RESOURCE EXHAUSTION
EXTERNAL PROVIDER FAILURE
USER ERROR
FALSE POSITIVE

For each hypothesis, provide:

EVIDENCE FOR
EVIDENCE AGAINST
MISSING DATA
NEXT TEST

Separate actions into:

CONTAINMENT
PRESERVATION OF EVIDENCE
ERADICATION
RECOVERY
POST-INCIDENT ACTIONS

Do not recommend deleting or reinstalling anything before considering whether that would destroy evidence we need.

Analyze the following HTTP/security logs DEFENSIVELY.

LOG:

"""
[LOG]
"""

Do not label every unusual request an attack.

For each pattern, determine:

SOURCE
TIMESTAMP
METHOD
PATH
STATUS
USER AGENT
REQUEST FREQUENCY
RESPONSE SIZE
LATENCY

Group similar requests.

Try to distinguish between:

normal crawler;
search engine bot;
vulnerability scanner;
credential stuffing;
brute force;
path enumeration;
WordPress scanning;
LFI/RFI probing;
SQLi probing;
XSS probing;
SSRF probing;
automated noise;
real exploitation attempt.

For each suspicious pattern, provide:

WHAT IT APPEARS TO TEST
EVIDENCE
DID IT SUCCEED?
RELEVANT SERVER RESPONSE
RISK
RECOMMENDED DEFENSIVE ACTION

Important:

A request that contains an exploit string is not proof that the exploit worked.

Look for evidence of successful impact on the system.

Perform a Linux server hardening audit.

DISTRIBUTION:
[...]

VERSION:
[...]

ROLE:
[web/db/VPN/reverse proxy...]

PUBLIC SERVICES:
[...]

USERS:
[...]

SSH CONFIG:
[...]

FIREWALL:
[...]

CONTAINERS:
[...]

First check current recommendations for the specified distribution and version.

Audit:

PATCH STATUS
USER ACCOUNTS
SSH
SUDO
FIREWALL
LISTENING PORTS
SYSTEMD SERVICES
FILE PERMISSIONS
SECRETS
LOGGING
AUDITING
TIME SYNC
AUTOMATIC SECURITY UPDATES
KERNEL SETTINGS
BACKUPS

For SSH, check in particular:

password authentication;
root login;
key-based access;
MFA where applicable;
AllowUsers/AllowGroups;
legacy algorithms.

Do not suggest disabling services without first establishing whether the application uses them.

For every change, provide:

SECURITY BENEFIT
BREAKAGE RISK
HOW TO TEST
ROLLBACK

Audit the Docker deployment.

FILES:

Dockerfile:
"""
[Dockerfile]
"""

Compose file:
"""
[docker-compose.yml]
"""

Check:

BASE IMAGE
IMAGE VERSION PINNING
IMAGE ORIGIN
ROOT/NON-ROOT
CAPABILITIES
PRIVILEGED MODE
HOST MOUNTS
DOCKER SOCKET
SECRETS
ENVIRONMENT VARIABLES
NETWORK EXPOSURE
READ-ONLY FILESYSTEM
TMPFS
RESOURCE LIMITS
HEALTHCHECK
SIGNAL HANDLING
BUILD SECRETS
MULTI-STAGE BUILD
DEPENDENCIES

Check in particular whether the container:

- runs as root without needing to;
- uses privileged mode;
- mounts /var/run/docker.sock;
- mounts sensitive host directories;
- has unnecessary Linux capabilities;
- includes build tools in the production image;
- contains a secret in an image layer;
- exposes a port that does not need to be public.

For each finding, provide:

CURRENT CONFIG
RISK
MINIMAL HARDENING
COMPATIBILITY RISK
TEST

Perform a security and reliability audit of the Kubernetes workload.

KUBERNETES VERSION:
[...]

MANIFESTS / HELM:

"""
[CONFIG]
"""

Check:

Pod Security Standards;
service accounts;
RBAC;
network policies;
securityContext;
runAsNonRoot;
capabilities;
privilege escalation;
seccomp;
hostNetwork;
hostPID;
hostPath;
secrets;
config maps;
image sources;
image tags/digests;
resource requests/limits;
probes;
PDB;
namespace isolation.

For each workload, determine:

Does it actually require privileged access?

Can it use the Restricted Pod Security profile?

If not, explain exactly which function prevents that.

Check in particular for:

wildcard RBAC;
cluster-admin;
default service account;
automountServiceAccountToken;
cross-namespace access;
unrestricted egress;
host filesystem access.

Do not automatically move everything to Restricted if the application would stop working.

Suggest a migration plan:

AUDIT
WARN
ENFORCE

with a compatibility check.

Audit the Terraform configuration.

CLOUD:
[AWS/Azure/GCP/...]

TERRAFORM VERSION:
[...]

CODE:
[...]

First check:

VERSION PINNING
PROVIDERS
MODULE SOURCES
STATE BACKEND
STATE LOCKING
STATE ENCRYPTION
ACCESS CONTROL
SECRETS
IAM
PUBLIC EXPOSURE
NETWORK RULES
DATABASES
STORAGE
LOGGING
BACKUPS

Analyze Terraform state in particular.

Determine:

- where it is stored;
- who can access it;
- whether it contains sensitive data;
- whether it is encrypted;
- whether locking is enabled;
- whether it was accidentally committed.

For each resource, check:

Could a change cause replacement?

Could replacement cause data loss?

For each proposed change, state the Terraform effect:

IN-PLACE
RECREATE
UNKNOWN UNTIL PLAN

Do not assume that `sensitive = true` means the value never appears in state.

Before apply, require a review of:

terraform plan

and identify any unexpected destroy/recreate operations.

Perform a least-privilege audit of cloud IAM.

PROVIDER:
[AWS/Azure/GCP]

PRINCIPALS:
[...]

ROLES/POLICIES:
[...]

First map:

HUMAN USERS
SERVICE ACCOUNTS
WORKLOAD IDENTITIES
CI/CD IDENTITIES
ADMIN ACCOUNTS
BREAK-GLASS ACCOUNTS

For each identity, determine:

WHO/WHAT USES IT
TRUST POLICY
ASSIGNED PERMISSIONS
PERMISSIONS ACTUALLY REQUIRED
CREDENTIAL TYPE
CREDENTIAL LIFETIME

Look for:

wildcard actions;
wildcard resources;
privilege escalation paths;
pass-role/assume-role issues;
unused credentials;
long-lived access keys;
cross-account trust;
service account key files;
shared credentials.

Check in particular for permission combinations that enable escalation even if each permission looks harmless by itself.

Do not recommend removing permissions without identifying the workload that uses them.

Final classification:

KEEP
REDUCE
REMOVE
INVESTIGATE

Find every way the infrastructure is accessible from the internet.

CLOUD CONFIG:
[...]

Map:

public IPs;
load balancers;
CDN;
API gateways;
storage buckets;
databases;
SSH/RDP;
Kubernetes services;
serverless endpoints;
management interfaces.

For each public endpoint, determine:

SERVICE
PORT
PROTOCOL
AUTHENTICATION
EXPECTED AUDIENCE
WHY IT IS PUBLIC
SOURCE RESTRICTION
WAF/RATE LIMIT
TLS

Identify resources that are:

INTENTIONALLY PUBLIC
UNNECESSARILY PUBLIC
UNCERTAIN

Check management ports and databases in particular.

Do not automatically label HTTP port 80 a vulnerability if it is used only for a controlled redirect to HTTPS.

Analyze the network architecture.

DIAGRAM / DESCRIPTION:

[...]

Map the zones:

INTERNET
EDGE
PUBLIC SUBNET
APPLICATION
DATABASE
MANAGEMENT
CI/CD
VPN
THIRD-PARTY

For each communication flow, state:

SOURCE
DESTINATION
PORT
PROTOCOL
PURPOSE
AUTHENTICATION
ENCRYPTION
REQUIRED?

Look for:

flat network;
unnecessary east-west access;
0.0.0.0/0 rules;
overly broad security groups;
management interfaces exposed publicly;
unrestricted egress;
database exposure;
DNS dependency;
single points of failure.

Apply the principle:

default deny where practical.

However, do not block a communication flow until you have confirmed the application does not use it.

Audit the firewall rules.

RULES:

"""
[rules]
"""

For each rule, create:

SOURCE
DESTINATION
PROTOCOL
PORT
DIRECTION
PURPOSE
OWNER
LAST KNOWN USE

Look for:

ANY/ANY;
0.0.0.0/0;
::/0;
unused rules;
shadowed rules;
duplicate rules;
management ports exposed publicly;
temporary rules that became permanent.

Classify each rule:

REQUIRED
OVERLY BROAD
POSSIBLY UNUSED
DANGEROUS
UNKNOWN

If you suggest narrowing a rule, define how to check actual traffic before changing it.

Audit the DNS configuration for this domain:

[DOMAIN]

Check current public DNS records.

Analyze:

A
AAAA
CNAME
MX
TXT
NS
CAA
DNSSEC

If email is used, also check:

SPF
DKIM
DMARC

Look for:

dangling DNS records;
subdomain takeover risk;
old IP addresses;
inactive services;
overly broad SPF policies;
missing DMARC;
unnecessary wildcard records;
CAA mismatches;
split-brain DNS issues.

For each finding, explain:

CURRENT RECORD
EXPECTED PURPOSE
RISK
SAFE CHANGE
DNS PROPAGATION IMPACT

Do not recommend deleting a record until you determine which service uses it.

Audit the service's TLS configuration:

[HOST]

Check:

certificate chain;
hostname;
expiration;
supported protocol versions;
cipher configuration;
HSTS;
OCSP, if relevant;
HTTP to HTTPS redirect;
mixed content;
certificate automation.

In particular, check whether:

TLS termination happens at multiple layers;
the CDN and origin use different policies;
the origin can be reached directly, bypassing the CDN or WAF.

Separate:

CERTIFICATE PROBLEM
PROTOCOL PROBLEM
APPLICATION HEADER PROBLEM
ORIGIN EXPOSURE

For each issue, provide a minimal correction and a way to verify it.

Audit the reverse proxy configuration.

SOFTWARE:
[Nginx/Caddy/Traefik/Apache...]

VERSION:
[...]

CONFIG:

"""
[CONFIG]
"""

Check:

TLS;
host validation;
proxy headers;
X-Forwarded-For;
real client IP;
request size;
timeouts;
buffering;
WebSocket upgrade;
rate limiting;
error pages;
server tokens;
static file paths;
cache;
upstream failover.

Look in particular for:

host header injection;
client IP spoofing caused by incorrect trusted proxy settings;
path normalization issues;
proxy_pass or path mismatches;
unintended exposure of internal endpoints;
unbounded upload size;
timeouts that are too long.

For every change, check its effect on the application behind the proxy.

Audit the backup and disaster recovery strategy.

SYSTEM:
[...]

DATA SOURCES:

database:
object storage:
filesystem:
configuration:
secrets:

Current backup:
[...]

Define:

RPO
RTO

Then check:

BACKUP FREQUENCY
RETENTION
OFFSITE COPY
IMMUTABILITY
ENCRYPTION
ACCESS CONTROL
VERSIONING
RESTORE PROCEDURE
RESTORE TESTING
MONITORING

Most importantly:

Treat a backup that has never been restored as unverified.

Create this scenario:

1. The production database is lost.
2. The latest backup is available.
3. A new environment must be brought online.
4. The application must receive the correct secrets and configuration.
5. Data integrity must be confirmed.

Create an exact restore checklist.

Identify any data that the current backup does not cover at all.

Audit the system's observability.

Current setup:

LOGS:
[...]

METRICS:
[...]

TRACES:
[...]

ALERTS:
[...]

For each critical flow, determine:

How do we know it succeeded?

How do we know it partly succeeded?

How do we know it failed completely?

Check:

structured logging;
request IDs;
correlation IDs;
trace context;
error classification;
latency percentiles;
saturation;
queue depth;
database pool;
external dependency health.

Look in particular for:

silent failures;
errors that are logged but do not trigger alerts;
alerts with no actionable meaning;
duplicate or noisy alerts;
sensitive data in logs.

For each proposed alert, define:

CONDITION
WINDOW
THRESHOLD
SEVERITY
OWNER
ACTION

Audit the CI/CD pipeline with a focus on software supply chain security.

CI:
[GitHub Actions/GitLab/Jenkins...]

CONFIG:
[...]

Check:

third-party actions or plugins;
version pinning;
commit SHA pinning where relevant;
permissions;
OIDC;
long-lived credentials;
secrets;
fork pull request behavior;
artifact integrity;
dependency installation;
build provenance;
release signing;
deployment approvals.

Check the default CI token permissions in particular.

For each job, determine:

WHAT CODE RUNS
WHAT SECRETS IT CAN ACCESS
WHAT IT CAN WRITE
WHAT EXTERNAL CODE IT DOWNLOADS

Look for a path by which a pull request or compromised dependency could access production credentials.

Propose a least-privilege pipeline.

Do not remove functionality without providing an alternative.

Audit secrets management.

SECRET TYPES:

DB credentials
API keys
OAuth secrets
TLS private keys
cloud credentials
webhook secrets
signing keys

LOCATIONS:

.env
CI
cloud secret manager
repository
container environment
Terraform
Kubernetes

For each secret, determine:

OWNER
WHERE STORED
WHO CAN READ
HOW INJECTED
ROTATION
EXPIRATION
AUDIT LOGGING

Look for:

secrets in Git history;
secrets in Docker image layers;
secrets in Terraform state;
secrets in the client bundle;
secrets in logs;
shared credentials;
secrets that have never been rotated.

If a secret has already leaked, deleting it from the current Git file is not enough.

Mark when rotation or revocation is needed, and assess historical exposure.

Audit the application's HTTP security headers.

Server response:

"""
[HEADERS]
"""

Assess whether these are relevant:

Content-Security-Policy
Strict-Transport-Security
X-Content-Type-Options
Referrer-Policy
Permissions-Policy
frame-ancestors
COOP
COEP
CORP

Do not recommend a header just because it appears on a checklist.

Explain which specific class of problem it addresses.

For CSP:

do not immediately propose an extremely restrictive policy.

Create a migration plan:

1. Inventory actual script, style, and resource sources.
2. Use Report-Only.
3. Analyze violations.
4. Remove unnecessary sources.
5. Enforce the policy.

Avoid `unsafe-inline` and `unsafe-eval` where they can realistically be removed without breaking the application.

Design rate limiting and abuse protection for:

[API/FEATURE]

Endpoints:
[...]

Users:

anonymous;
authenticated;
premium;
internal.

Do not use one global limit for everything.

For each endpoint, assess:

COST PER REQUEST
ABUSE VALUE
NORMAL BURST
AUTH STATE
SIDE EFFECT

Consider these keys:

IP;
user ID;
API key;
tenant;
device;
resource.

Protect these in particular:

login;
password reset;
OTP;
email or SMS sending;
search;
expensive AI or API operations;
file uploads;
checkout;
coupon validation.

Define:

RATE
BURST
WINDOW
RESPONSE
RETRY-AFTER
ESCALATION

Consider users sharing a NAT address before applying aggressive IP-only throttling.

Analyze this traffic spike.

DATA:

requests/sec:
bandwidth:
CPU:
memory:
status codes:
top paths:
source distribution:
user agents:
cache hit rate:
origin latency:

Do not immediately conclude that this is a DDoS attack.

Separate the possible causes:

organic traffic;
bot crawl;
misconfigured client;
retry storm;
cache failure;
application loop;
Layer 7 attack;
volumetric attack.

For each hypothesis, identify evidence that would confirm it.

Build protection in layers:

EDGE/CDN
WAF
RATE LIMITING
APPLICATION
DATABASE
EXTERNAL DEPENDENCIES

The priority is to preserve legitimate traffic, not simply to block as many requests as possible.

There is a production incident.

SYMPTOMS:
[...]

START TIME:
[...]

LATEST DEPLOY:
[...]

METRICS:
[...]

LOGS:
[...]

Do not immediately search for the perfect root cause.

First optimize for safe recovery.

Proceed in this order:

1. Confirm user impact.
2. Determine the blast radius.
3. Check recent changes.
4. Check dependency status.
5. Find a safe mitigation.
6. Stabilize the service.
7. Only then conduct a deeper root cause analysis.

For every proposed action, state:

EXPECTED BENEFIT
RISK
REVERSIBILITY
VERIFICATION

If a rollback could cause a database or schema problem, warn me before the rollback.

Keep a timeline of decisions.

Create a blameless postmortem from the incident data.

DATA:
[...]

Structure:

SUMMARY
USER IMPACT
DURATION
DETECTION
TIMELINE
ROOT CAUSE
CONTRIBUTING FACTORS
WHAT WENT WELL
WHAT WENT POORLY
WHERE WE GOT LUCKY
CORRECTIVE ACTIONS

Do not write:

"human error"

as the root cause without further explanation.

Look for systemic factors:

Why could one human mistake cause the incident?

For each corrective action, state:

OWNER
PRIORITY
VERIFICATION CRITERION

Separate:

PREVENTION
DETECTION
MITIGATION
RECOVERY

Optimize cloud costs without compromising reliability.

CLOUD:
[...]

SERVICES:
[...]

BILLING:
[...]

First look for:

idle resources;
oversized resources;
unused disks;
stale snapshots;
unused load balancers;
unused IPs;
egress;
logging or storage growth;
database sizing;
reserved capacity opportunities.

Before shutting anything down, determine:

OWNER
LAST USE
DEPENDENCIES
PRODUCTION IMPACT
RECOVERY COST

For each optimization, provide:

CURRENT COST
EXPECTED SAVING
RISK
IMPLEMENTATION
ROLLBACK

Do not delete a resource just because it currently has low utilization.

=== SECURITY QUALITY CONTROL ===

Before your final answer:

1. Check whether the finding is real or merely hypothetical.
2. Define the preconditions.
3. Define the trust boundary.
4. Show the concrete attack or failure path.
5. Check whether an existing control already prevents the scenario.
6. Distinguish exposure from exploitability.
7. Distinguish a vulnerability from a misconfiguration.
8. Assess the blast radius.
9. Propose the smallest safe correction.
10. Define a verification test.
11. Define a rollback.
12. Check that the fix will not create a new availability problem.

If there is not enough evidence, write:

UNVERIFIED

instead of inventing certainty about the finding.

=== PRODUCTION CHANGE QA ===

Before recommending an infrastructure change, check:

- current state;
- desired state;
- dependencies;
- stateful resources;
- backward compatibility;
- active traffic;
- rollout strategy;
- failure scenario;
- monitoring;
- rollback.

Answer these questions in particular:

What if the change succeeds only 50 percent?

What if the process fails during the change?

What if the old and new versions run at the same time?

What if we need to roll back after a database migration?

How do we prove the system is healthy after the change?

Do not treat:

"deployment succeeded"

as proof that:

"the application works correctly".

Audit the previous security report as an independent reviewer.

Do not assume that the findings are correct.

For each finding, check:

1. whether an affected component actually exists;
2. whether there is actual attacker-controlled input;
3. whether the input crosses the relevant trust boundary;
4. whether a control already blocks the scenario;
5. whether the described impact really follows from the issue;
6. whether the severity matches the actual conditions.

Label each finding:

CONFIRMED
OVERSTATED
UNDERSTATED
DUPLICATE
FALSE POSITIVE
NEEDS MORE EVIDENCE

Then look for issues that the first audit completely missed.

Focus in particular on chained failures where two or more seemingly minor weaknesses combine into a serious issue.

Use current sources for the following security issue:

[TOPIC]

Search the web.

Prioritize:

1. vendor security advisories;
2. CVE/NVD/CISA or the relevant national CERT;
3. the official repository;
4. maintainer advisories;
5. official documentation;
6. high-quality independent technical analysis.

Check:

affected versions;
fixed versions;
publication date;
update date;
exploit prerequisites;
known exploitation status when a reliable source exists;
mitigation;
official patch.

Do not conclude that our system is vulnerable just because it uses the product.

First check:

our version;
our configuration;
whether we use the affected feature;
whether the vulnerable code path is reachable.

Classify the final result:

AFFECTED
NOT AFFECTED
POTENTIALLY AFFECTED
INSUFFICIENT INFORMATION

If you have no evidence for a finding, do not increase the number of findings just to make the report look complete.

Find all possible vulnerabilities.

Analyze the following economic question:

[TOPIC]

COUNTRY / REGION:
[...]

PERIOD:
[...]

Check current data before answering.

Prioritize these sources:

1. national statistical office;
2. central bank;
3. ministry of finance;
4. Eurostat;
5. OECD;
6. IMF;
7. World Bank;
8. BIS / ILO / UN when relevant.

For each indicator, determine:

- exact series name;
- definition;
- unit;
- frequency;
- nominal or real;
- seasonally adjusted or unadjusted;
- base year, if applicable;
- source;
- date of the latest data point;
- whether the series has been revised.

Do not compare two series as identical if they use different methodologies.

Separate the answer into:

DATA
METHODOLOGY
FINDINGS
INTERPRETATION
POSSIBLE EXPLANATIONS
LIMITATIONS

Analyze inflation in:

[COUNTRY]

PERIOD:
[...]

Do not use the word "inflation" without specifying the indicator.

Find and separate these measures where available:

- CPI;
- HICP;
- headline inflation;
- core inflation;
- YoY;
- MoM;
- annual average.

For each data point, state:

value;
period;
source;
methodology.

Analyze the contributions of these components in particular:

- food;
- energy;
- housing;
- services;
- transport;
- administratively regulated prices.

If comparing countries, use a comparable indicator.

Do not compare one country's national CPI with another country's HICP without a warning.

At the end, separate:

WHAT HAPPENED
WHICH COMPONENTS CONTRIBUTED
WHAT IS FACT
WHAT IS ECONOMIC INTERPRETATION

Analyze GDP for:

[COUNTRY / COUNTRIES]

PERIOD:
[...]

Be sure to distinguish:

- nominal GDP;
- real GDP;
- GDP growth;
- GDP per capita;
- PPP;
- current prices;
- constant prices.

If using a growth rate, state:

YoY / QoQ;
whether it is seasonally adjusted;
real or nominal.

Analyze the contributions if available:

household consumption;
government consumption;
investment;
exports;
imports;
changes in inventories.

If the statistical office has since revised GDP, use the latest revision.

Show:

ORIGINAL DATA, if relevant;
REVISED DATA;
DIFFERENCE.

Do not claim that a policy caused a change in GDP without additional evidence.

Analyze wage trends and living standards in:

[COUNTRY]

PERIOD:
[...]

Find:

- average earnings;
- median earnings, if available;
- nominal change;
- real change;
- CPI;
- employment;
- disposable income, if available.

Do not rely only on the average wage if the wage distribution could change the conclusion.

Distinguish:

GROSS
NET
MEDIAN
AVERAGE

When comparing periods, calculate real change:

real growth is approximately nominal growth adjusted for price growth

but use the official real series when available.

Analyze in particular whether:

the increase in average wages comes from wage growth among existing workers
or
a change in the composition of employed workers.

At the end, answer:

Did purchasing power increase?
Which groups might not share this experience?
What data are missing for a more certain conclusion?

Analyze the labor market in:

[COUNTRY]

PERIOD:
[...]

Do not use the unemployment rate alone.

Find:

- unemployment rate;
- employment rate;
- activity rate;
- labor force participation;
- number of employed people;
- number of unemployed people;
- long-term unemployment;
- youth unemployment, when relevant.

Check the labor force survey methodology.

Distinguish:

registered unemployment
from
survey-based unemployment.

Do not treat them as the same metric.

If the number of unemployed people is falling, check whether the participation rate is falling at the same time.

Explain situations in which the unemployment rate can fall without a real improvement in the labor market.

Analyze public debt:

[COUNTRY]

PERIOD:
[...]

Find at least:

- nominal public debt;
- debt as a percentage of GDP;
- debt currency;
- domestic and external debt;
- maturities;
- average interest rate, if available;
- interest costs;
- budget deficit or surplus.

Check which definition is being used:

central government debt
general government debt
public-sector debt

Do not mix them up.

If comparing countries, use the same definition.

Analyze in particular whether:

the debt-to-GDP ratio fell because debt decreased
or
because nominal GDP grew.

Separate:

LEVEL OF DEBT
COST OF DEBT
MATURITY RISK
CURRENCY RISK
REFINANCING RISK

Compare these countries:

[COUNTRIES]

BY:

[INDICATORS]

PERIOD:
[...]

Before comparing them, check whether the series are methodologically comparable.

For each indicator, check the metadata:

SOURCE
DEFINITION
UNIT
FREQUENCY
ADJUSTMENT
TIME PERIOD

If one country has no fully comparable data, label it:

NOT COMPARABLE

instead of creating a falsely precise comparison.

Do not create a ranking just because a few numbers are available.

First assess whether the indicator really measures what we are trying to compare.

Find the historical series:

INDICATOR:
[...]

COUNTRIES:
[...]

YEARS:
[...]

STRICT DATA RULES:

- do not interpolate;
- do not extrapolate;
- do not copy an adjacent year's value;
- do not estimate a number from a chart if a table is available;
- do not combine different definitions;
- do not fill a missing value with an estimate.

For each cell, return:

COUNTRY
YEAR
VALUE
UNIT
SOURCE
PUBLICATION
TABLE/PAGE, if available
STATUS

STATUS must be one of:

EXACT
DERIVED
ESTIMATE
MISSING
STRUCTURAL_NA

Use ESTIMATE only if I explicitly allow estimates.

If the source has revised the historical series, use the latest revision.

Audit this economic dataset before performing any analysis.

Check for:

- missing years;
- duplicates;
- sudden unit changes;
- percent versus percentage points;
- thousands, millions, or billions;
- current prices versus constant prices;
- nominal versus real;
- calendar year versus fiscal year;
- annual average versus end of period;
- seasonal adjustment status;
- base year changes;
- methodology changes.

For each issue, return:

COUNTRY / SERIES
PERIOD
PROBLEM
WHY IT MATTERS
SAFE ACTION

Do not automatically "clean" the data.

Flag the problem first.

Identify structural breaks in particular.

Check the following claim:

"[CLAIM]"

Do not check only whether someone has already written the same thing.

Find the original data source.

Identify:

- indicator;
- period;
- population;
- geography;
- methodology;
- unit.

Then check the claim's math.

Classify it:

TRUE
MOSTLY TRUE
MISLEADING
INCOMPLETE
FALSE
IMPOSSIBLE TO VERIFY

Explain exactly what the problem is.

Look for cherry-picked periods in particular.

If the claim says:

"the highest ever"

check the entire available time series.

If it says:

"the highest in Europe"

check which countries and definitions are included in the comparison.

Analyze this claim:

"[X causes Y]"

Do not accept causation just because X and Y move together.

Check:

TEMPORAL ORDER
CONFOUNDERS
REVERSE CAUSALITY
SELECTION EFFECTS
COMMON TREND
POLICY CHANGES
STRUCTURAL BREAKS

Find relevant research, if available.

Separate:

CORRELATION
PLAUSIBLE MECHANISM
CAUSAL EVIDENCE
UNCERTAINTY

If the available data support only a correlational conclusion, do not use causal language.

Analyze the following economic question:

[TOPIC]

COUNTRY / REGION:
[...]

PERIOD:
[...]

Check current data before answering.

Prioritize these sources:

1. national statistical office;
2. central bank;
3. ministry of finance;
4. Eurostat;
5. OECD;
6. IMF;
7. World Bank;
8. BIS / ILO / UN when relevant.

For each indicator, determine:

- exact series name;
- definition;
- unit;
- frequency;
- nominal or real;
- seasonally adjusted or unadjusted;
- base year, if applicable;
- source;
- date of the latest data point;
- whether the series has been revised.

Do not compare two series as identical if they use different methodologies.

Separate the answer into:

DATA
METHODOLOGY
FINDINGS
INTERPRETATION
POSSIBLE EXPLANATIONS
LIMITATIONS

Analyze inflation in:

[COUNTRY]

PERIOD:
[...]

Do not use the word "inflation" without specifying the indicator.

Find and separate these measures where available:

- CPI;
- HICP;
- headline inflation;
- core inflation;
- YoY;
- MoM;
- annual average.

For each data point, state:

value;
period;
source;
methodology.

Analyze the contributions of these components in particular:

- food;
- energy;
- housing;
- services;
- transport;
- administratively regulated prices.

If comparing countries, use a comparable indicator.

Do not compare one country's national CPI with another country's HICP without a warning.

At the end, separate:

WHAT HAPPENED
WHICH COMPONENTS CONTRIBUTED
WHAT IS FACT
WHAT IS ECONOMIC INTERPRETATION

Analyze GDP for:

[COUNTRY / COUNTRIES]

PERIOD:
[...]

Be sure to distinguish:

- nominal GDP;
- real GDP;
- GDP growth;
- GDP per capita;
- PPP;
- current prices;
- constant prices.

If using a growth rate, state:

YoY / QoQ;
whether it is seasonally adjusted;
real or nominal.

Analyze the contributions if available:

household consumption;
government consumption;
investment;
exports;
imports;
changes in inventories.

If the statistical office has since revised GDP, use the latest revision.

Show:

ORIGINAL DATA, if relevant;
REVISED DATA;
DIFFERENCE.

Do not claim that a policy caused a change in GDP without additional evidence.

Analyze wage trends and living standards in:

[COUNTRY]

PERIOD:
[...]

Find:

- average earnings;
- median earnings, if available;
- nominal change;
- real change;
- CPI;
- employment;
- disposable income, if available.

Do not rely only on the average wage if the wage distribution could change the conclusion.

Distinguish:

GROSS
NET
MEDIAN
AVERAGE

When comparing periods, calculate real change:

real growth is approximately nominal growth adjusted for price growth

but use the official real series when available.

Analyze in particular whether:

the increase in average wages comes from wage growth among existing workers
or
a change in the composition of employed workers.

At the end, answer:

Did purchasing power increase?
Which groups might not share this experience?
What data are missing for a more certain conclusion?

Analyze the labor market in:

[COUNTRY]

PERIOD:
[...]

Do not use the unemployment rate alone.

Find:

- unemployment rate;
- employment rate;
- activity rate;
- labor force participation;
- number of employed people;
- number of unemployed people;
- long-term unemployment;
- youth unemployment, when relevant.

Check the labor force survey methodology.

Distinguish:

registered unemployment
from
survey-based unemployment.

Do not treat them as the same metric.

If the number of unemployed people is falling, check whether the participation rate is falling at the same time.

Explain situations in which the unemployment rate can fall without a real improvement in the labor market.

Analyze public debt:

[COUNTRY]

PERIOD:
[...]

Find at least:

- nominal public debt;
- debt as a percentage of GDP;
- debt currency;
- domestic and external debt;
- maturities;
- average interest rate, if available;
- interest costs;
- budget deficit or surplus.

Check which definition is being used:

central government debt
general government debt
public-sector debt

Do not mix them up.

If comparing countries, use the same definition.

Analyze in particular whether:

the debt-to-GDP ratio fell because debt decreased
or
because nominal GDP grew.

Separate:

LEVEL OF DEBT
COST OF DEBT
MATURITY RISK
CURRENCY RISK
REFINANCING RISK

Compare these countries:

[COUNTRIES]

BY:

[INDICATORS]

PERIOD:
[...]

Before comparing them, check whether the series are methodologically comparable.

For each indicator, check the metadata:

SOURCE
DEFINITION
UNIT
FREQUENCY
ADJUSTMENT
TIME PERIOD

If one country has no fully comparable data, label it:

NOT COMPARABLE

instead of creating a falsely precise comparison.

Do not create a ranking just because a few numbers are available.

First assess whether the indicator really measures what we are trying to compare.

Find the historical series:

INDICATOR:
[...]

COUNTRIES:
[...]

YEARS:
[...]

STRICT DATA RULES:

- do not interpolate;
- do not extrapolate;
- do not copy an adjacent year's value;
- do not estimate a number from a chart if a table is available;
- do not combine different definitions;
- do not fill a missing value with an estimate.

For each cell, return:

COUNTRY
YEAR
VALUE
UNIT
SOURCE
PUBLICATION
TABLE/PAGE, if available
STATUS

STATUS must be one of:

EXACT
DERIVED
ESTIMATE
MISSING
STRUCTURAL_NA

Use ESTIMATE only if I explicitly allow estimates.

If the source has revised the historical series, use the latest revision.

Audit this economic dataset before performing any analysis.

Check for:

- missing years;
- duplicates;
- sudden unit changes;
- percent versus percentage points;
- thousands, millions, or billions;
- current prices versus constant prices;
- nominal versus real;
- calendar year versus fiscal year;
- annual average versus end of period;
- seasonal adjustment status;
- base year changes;
- methodology changes.

For each issue, return:

COUNTRY / SERIES
PERIOD
PROBLEM
WHY IT MATTERS
SAFE ACTION

Do not automatically "clean" the data.

Flag the problem first.

Identify structural breaks in particular.

Check the following claim:

"[CLAIM]"

Do not check only whether someone has already written the same thing.

Find the original data source.

Identify:

- indicator;
- period;
- population;
- geography;
- methodology;
- unit.

Then check the claim's math.

Classify it:

TRUE
MOSTLY TRUE
MISLEADING
INCOMPLETE
FALSE
IMPOSSIBLE TO VERIFY

Explain exactly what the problem is.

Look for cherry-picked periods in particular.

If the claim says:

"the highest ever"

check the entire available time series.

If it says:

"the highest in Europe"

check which countries and definitions are included in the comparison.

Analyze this claim:

"[X causes Y]"

Do not accept causation just because X and Y move together.

Check:

TEMPORAL ORDER
CONFOUNDERS
REVERSE CAUSALITY
SELECTION EFFECTS
COMMON TREND
POLICY CHANGES
STRUCTURAL BREAKS

Find relevant research, if available.

Separate:

CORRELATION
PLAUSIBLE MECHANISM
CAUSAL EVIDENCE
UNCERTAINTY

If the available data support only a correlational conclusion, do not use causal language.

Research the revision history of this indicator:

[INDICATOR]

COUNTRY:
[...]

PERIOD:
[...]

Find:

FIRST RELEASE
PREVIOUS VINTAGE
CURRENT VINTAGE

For each period, show:

original value;
current value;
revision size.

Try to determine the reason for the revision:

new source data;
benchmark revision;
methodology change;
seasonal adjustment;
reclassification;
error correction.

Do not use the original release if the official statistical institution has since revised the series, except when we are analyzing what was known at the time.

Analyze:

[METRIC]

Check in particular whether public discussion uses nominal growth as evidence of real improvement.

Find:

nominal value;
price index or deflator;
real value;
population, if relevant.

Calculate and explain:

NOMINAL CHANGE
REAL CHANGE
PER-CAPITA CHANGE

If the three conclusions differ, clearly show the difference.

Example:

total nominal GDP can grow strongly,
while real GDP per capita grows much more slowly.

Do not mix these concepts.

Analyze the exchange rate:

[CURRENCY A] / [CURRENCY B]

PERIOD:
[...]

Find official or reliable market data.

Distinguish:

spot rate;
average rate;
end-of-period rate;
central bank reference rate;
real effective exchange rate.

If analyzing the economic impact, consider:

inflation;
interest rates;
trade balance;
capital flows;
FX reserves;
monetary policy;
risk sentiment.

Do not explain every short-term movement with a single variable without evidence.

Analyze foreign trade:

[COUNTRY]

PERIOD:
[...]

Find:

exports;
imports;
trade balance;
goods and services breakdown;
top partners;
top product categories.

Be sure to check:

nominal currency;
current prices;
FOB/CIF methodology when relevant;
seasonality.

If nominal exports are growing, check whether the increase is due to:

PRICE EFFECT
VOLUME EFFECT
or both.

Do not automatically conclude that a higher nominal export value means more physical exports.

Analyze foreign direct investment for:

[COUNTRY]

PERIOD:
[...]

Distinguish:

FDI FLOW
FDI STOCK

and:

gross inflows
net inflows

Check the source methodology.

Where possible, analyze:

equity;
reinvested earnings;
intercompany debt.

Do not automatically treat all FDI as a new factory or greenfield investment.

Separate greenfield investment, M&A, and financial flows when the data allow it.

Analyze productivity in:

[COUNTRY / SECTOR]

PERIOD:
[...]

Before calculating it, define which productivity measure we are using:

GDP per worker;
GDP per hour worked;
manufacturing value added per worker;
TFP;
other.

Do not use different metrics as synonyms.

Find:

output measure;
employment or hours measure;
price adjustment;
population coverage.

If comparing countries, check PPP and methodology.

Separate:

LEVEL
GROWTH
CYCLICAL EFFECTS
STRUCTURAL TREND

Research industrial automation in:

[COUNTRIES]

PERIOD:
[...]

Prioritize these data:

operational stock of industrial robots;
annual installations;
manufacturing employment;
robot density.

If robot density is not directly available, you may calculate:

operational stock / manufacturing employment x 10,000

but only if:

- the numerator and denominator refer to the same year;
- the geographic coverage matches;
- the definitions are compatible.

Clearly label every DERIVED value.

Never present a derived value as an official published IFR value.

Do not interpolate missing values.

Create a scenario analysis for:

[TOPIC]

HORIZON:
[...]

Do not try to produce one certain forecast.

Create:

BASE CASE
UPSIDE CASE
DOWNSIDE CASE

For each scenario, define:

assumptions;
inflation;
growth;
interest rates;
exchange rate;
external demand;
energy prices;
relevant specific factors.

Then state:

WHAT WOULD MAKE THIS SCENARIO MORE LIKELY
WHAT WOULD INVALIDATE IT

Do not assign precise probability percentages without an appropriate model and data.

=== ECONOMIC DATA QA ===

Before your final answer, check:

1. whether every indicator has a clear definition;
2. whether the units match;
3. whether nominal or real is clear;
4. whether seasonal adjustment is consistent;
5. whether the comparison uses the same period;
6. whether the data have been revised;
7. whether a newer official value is available;
8. whether an average has been incorrectly used instead of a median;
9. whether percent and percentage point have been confused;
10. whether correlation is incorrectly described as causation;
11. whether there is a structural break;
12. whether any estimated value is unlabelled.

If data cannot be found:

N/A

is an acceptable result.

An invented value is not.

Analyze my financial situation for informational purposes and through scenarios.

COUNTRY / TAX RESIDENCY:
[...]

CURRENCY:
[...]

MONTHLY NET INCOME:
[...]

FIXED EXPENSES:
[...]

VARIABLE EXPENSES:
[...]

DEBTS:
[...]

SAVINGS:
[...]

INVESTMENTS:
[...]

FINANCIAL GOALS:
[...]

TIME HORIZON:
[...]

LIQUIDITY I NEED:
[...]

Do not assume information I have not provided.

Separate these items in particular:

CASH FLOW
EMERGENCY FUND
DEBT
LIQUIDITY
INVESTMENTS
TAX CONSIDERATIONS
RISKS

If any interest rate, tax rule, inflation figure, or market data may have changed, check current sources.

Create at least:

CONSERVATIVE SCENARIO
BASE SCENARIO
ADVERSE SCENARIO

State the assumptions for each scenario explicitly.

Do not present a market forecast as fact.

Analyze this loan:

PRINCIPAL:
[...]

TERM:
[...]

NOMINAL INTEREST RATE:
[...]

APR:
[...]

FIXED OR VARIABLE:
[...]

CURRENCY:
[...]

MONTHLY PAYMENT:
[...]

FEES:
[...]

INSURANCE:
[...]

EARLY REPAYMENT:
[...]

First check that all the figures are mutually consistent.

Calculate:

- total payments;
- total interest paid;
- total fees;
- actual effective cost;
- interest as a share of total repayment.

If the rate is variable, create these scenarios:

- the base rate stays the same;
- +1 percentage point;
- +2 percentage points;
- -1 percentage point.

If the loan is tied to a reference rate, explain how a change in that rate affects the payment.

Analyze early repayment in particular:

- how much interest it saves;
- whether there are penalties;
- whether reducing the payment or the term is better if the bank lets me choose.

Do not use approximate math when we can calculate the result precisely.

Compare these two loans.

LOAN A:
[...]

LOAN B:
[...]

Do not compare them only by their monthly payments.

Compare:

PRINCIPAL
TERM
NOMINAL RATE
APR
TOTAL INTEREST
FEES
TOTAL COST
CURRENCY RISK
VARIABLE-RATE TERMS
EARLY REPAYMENT TERMS

Create a table.

Then create three scenarios:

1. I keep the loan until the end;
2. I repay it early after [X] years;
3. the interest rate rises by [X].

If information is missing, do not invent it.

Label it:

MISSING FOR AN ACCURATE COMPARISON.

Analyze this home loan.

PROPERTY PRICE:
[...]

DOWN PAYMENT:
[...]

LOAN AMOUNT:
[...]

TERM:
[...]

INTEREST RATE:
[...]

APR:
[...]

CURRENCY:
[...]

MONTHLY HOUSEHOLD INCOME:
[...]

OTHER DEBTS:
[...]

Show:

loan-to-value;
debt-service-to-income;
total interest;
total cost;
monthly share of income used for payments.

Then stress-test:

- interest rate +1 pp;
- interest rate +2 pp;
- income -20%;
- living costs +15%;
- six months without one source of income.

Calculate how large an emergency fund would be needed for:

3 months
6 months
12 months

of regular obligations.

If there are current local rules or bank limits, check official sources for the country I specified.

Compare these options for my situation:

A - keeping money in an account or deposit
B - money market or short-term instruments
C - government bonds
D - broadly diversified ETF
E - another option: [...]

DETAILS:

Country:
[...]

Currency:
[...]

Amount:
[...]

Time horizon:
[...]

How much money must be available quickly:
[...]

Goal:
[...]

Do not compare only expected returns.

Analyze:

LIQUIDITY
VOLATILITY
INFLATION RISK
CURRENCY RISK
CREDIT RISK
INTEREST-RATE RISK
TAX
FEES
TIME HORIZON

Use real returns where appropriate:

real return is approximately nominal return adjusted for inflation.

Do not assume that future stock returns are guaranteed.

Perform a fundamental analysis of:

[COMPANY]

PERIOD:
[...]

Use the latest available official financial statements.

Prioritize:

- annual report;
- quarterly report;
- regulatory filings;
- investor relations;
- earnings call or transcript;
- secondary sources only after that.

Analyze at least:

REVENUE
REVENUE GROWTH
GROSS MARGIN
OPERATING MARGIN
NET MARGIN
FREE CASH FLOW
CAPEX
CASH
DEBT
NET DEBT
INTEREST EXPENSE
SHARE COUNT
DILUTION
STOCK-BASED COMPENSATION

If the company has segments, analyze each segment separately.

If it depends on particular geographies, analyze revenue exposure by region.

Find the difference in particular between:

GAAP/IFRS METRICS
ADJUSTED/NON-GAAP METRICS

Do not use an adjusted metric without explaining what has been excluded.

Analyze the company's QUALITY OF EARNINGS:

[COMPANY]

Do not look only at net profit.

Check:

- operating cash flow;
- free cash flow;
- receivables;
- inventories;
- deferred revenue;
- one-off gains;
- restructuring charges;
- asset sales;
- stock-based compensation;
- capitalization of costs;
- changes in depreciation or amortization;
- working capital effects.

Questions:

Does profit track cash flow?

Are receivables growing faster than revenue?

Is inventory growing faster than sales?

Does the company often exclude "one-time" costs from adjusted results?

Does FCF look weaker or stronger than accounting profit?

For each finding, state the figures and periods.

Separate:

NORMAL OPERATING PERFORMANCE
ACCOUNTING EFFECT
ONE-OFF EFFECT
POTENTIAL WARNING SIGN

Analyze the company's balance sheet:

[COMPANY]

Check:

CASH
TOTAL DEBT
NET DEBT
SHORT-TERM DEBT
LONG-TERM DEBT
LEASE LIABILITIES
INTEREST EXPENSE
MATURITY SCHEDULE
CURRENT ASSETS
CURRENT LIABILITIES

Calculate where possible:

NET DEBT / EBITDA
INTEREST COVERAGE
CURRENT RATIO
QUICK RATIO

Check in particular:

- how much debt matures in the next 12, 24, or 36 months;
- the interest rates on that debt;
- how much debt has a fixed rate and how much has a variable rate;
- covenants;
- refinancing options;
- dependence on capital markets.

Do not automatically treat high nominal debt as a problem.

Assess it in relation to:

cash flow;
assets;
profitability;
maturity;
interest cost.

Value this company:

[COMPANY]

Do not use only one multiple.

Consider:

P/E
FORWARD P/E
EV/EBITDA
EV/EBIT
EV/SALES
P/FCF
FCF YIELD

Use metrics that make sense for this specific business model.

If the company is:

loss-making
high-growth
capital-intensive
a financial institution

adapt the method.

Compare it with:

the company's historical multiples;
relevant competitors;
revenue growth;
margins;
FCF;
debt.

Do not conclude:

"cheap"

just because its P/E is below average.

Check whether the lower valuation reflects a weaker business, debt, cyclicality, or falling earnings.

Build a transparent DCF model for:

[COMPANY]

Do not choose assumptions to get the valuation you want in advance.

Show all assumptions:

REVENUE GROWTH
OPERATING MARGIN
TAX RATE
D&A
CAPEX
WORKING CAPITAL
WACC
TERMINAL GROWTH

Create:

BASE CASE
BULL CASE
BEAR CASE

For each scenario, show:

FCF forecast;
terminal value;
enterprise value;
net debt adjustment;
equity value;
shares outstanding;
implied value per share.

Then create a sensitivity table for:

WACC
terminal growth.

Mark which assumptions have the greatest impact on the result.

Do not present the DCF result as a precise "true value".

Treat it as a model result that depends on assumptions.

Analyze the company's latest quarterly report:

[COMPANY]

Compare:

CURRENT QUARTER
PREVIOUS QUARTER
SAME QUARTER LAST YEAR

Analyze:

revenue;
growth;
gross margin;
operating margin;
FCF;
guidance;
segment performance;
customer metrics;
capex.

Look in particular for:

what changed in the guidance;
what management has stopped saying;
which metrics changed definition;
one-time items;
FX effects;
acquisitions.

Separate:

REPORTED RESULTS
MANAGEMENT EXPLANATION
INDEPENDENT INTERPRETATION

Do not automatically accept management's explanation as proof.

Analyze the company's earnings call:

[COMPANY]

Extract:

- key statements by management;
- guidance;
- changes in tone;
- analyst questions;
- questions management did not answer directly;
- new promises;
- risks management emphasized in particular.

Compare it with the previous two earnings calls.

Find:

what changed;
what is repeated;
which earlier promise was not fulfilled.

Separate:

FACT
MANAGEMENT CLAIM
FORWARD-LOOKING STATEMENT
INTERPRETATION

Perform due diligence on:

[COMPANY]

Do not focus only on finances.

Analyze:

BUSINESS MODEL
CUSTOMERS
CUSTOMER CONCENTRATION
SUPPLIERS
COMPETITORS
REGULATION
IP
MANAGEMENT
CAPITAL STRUCTURE
DEBT
LITIGATION
GEOGRAPHIC RISK
CYCLICALITY
DEPENDENCE ON KEY PRODUCTS
SECURITY / OPERATIONAL RISKS

For each risk, provide:

DESCRIPTION
EVIDENCE
POTENTIAL IMPACT
WHAT TO MONITOR

Look in particular for information that could challenge the company's bullish narrative.

Do not ignore negative evidence.

Analyze this ETF:

[TICKER / NAME]

Check:

INDEX
TER
FUND SIZE
DOMICILE
REPLICATION
ACCUMULATING / DISTRIBUTING
CURRENCY
HEDGING
HOLDINGS
SECTOR EXPOSURE
COUNTRY EXPOSURE
CONCENTRATION
TRACKING DIFFERENCE
LIQUIDITY

If I am in:

[COUNTRY]

also analyze the relevant tax or domicile context if reliable current sources are available.

Do not conclude that two ETFs are the same just because they track a "similar" index.

Check the specific benchmark and methodology.

Compare:

COMPANY A
[COMPANY]

COMPANY B
[COMPANY]

Do not immediately choose a "winner".

Compare:

business model;
revenue growth;
margins;
FCF;
balance sheet;
competitive position;
valuation;
cyclicality;
regulatory risk;
customer concentration.

For each category, provide the data and differences without an overall ranking.

Then identify:

the scenario in which A has stronger economics;
the scenario in which B has stronger economics;
the data point that would most change the analysis.

My investment thesis is:

"""
[THESIS]
"""

Do not try to confirm it.

Perform a red-team analysis.

Find:

- hidden assumptions;
- data that contradict it;
- risks I am overlooking;
- misinterpreted metrics;
- valuation risk;
- execution risk;
- competition;
- regulation;
- balance-sheet risk.

Then create:

WHAT MUST BE TRUE
WHAT COULD BREAK THE THESIS
LEADING INDICATORS
INVALIDATION CONDITIONS

Define specifically what would mean that my original thesis is wrong.

=== FINANCIAL ANALYSIS QA ===

Before your final answer, check:

1. whether financial data are from the same period;
2. whether we are using a trailing or forward metric;
3. whether the share count reflects diluted shares;
4. whether net debt includes relevant liabilities;
5. whether adjusted EBITDA excludes material recurring costs;
6. whether revenue growth includes acquisitions;
7. whether FX affects the comparison;
8. whether the data have been revised;
9. whether the valuation multiple uses the appropriate denominator;
10. whether the scenario assumptions are consistent with one another.

Separate:

HISTORICAL FACT
CURRENT MARKET DATA
MANAGEMENT GUIDANCE
ANALYST ESTIMATE
MODEL ASSUMPTION

Analyze the following legal question:

[QUESTION]

JURISDICTION:
[country / province / state / EU]

RELEVANT DATE:
[date]

AREA:
[employment / commercial / tax / contract / criminal / administrative / other]

Before reaching a conclusion, find the currently effective primary legal sources.

Prioritize:

1. official statutory text;
2. official gazette;
3. implementing regulation;
4. official regulator;
5. case law;
6. official interpretations;
7. legal scholarship;
8. secondary articles.

For each relevant law or regulation, check:

- title;
- section or article;
- whether it is currently in force;
- latest relevant amendment;
- effective date;
- transitional provisions;
- territorial scope.

Separate the answer into:

CURRENT LAW
CASE LAW
REGULATORY INTERPRETATION
POSSIBLE INTERPRETATION
UNCERTAINTY

Do not present a legal interpretation as the literal text of a statute.

Check the legal status of this provision:

[LAW / SECTION]

JURISDICTION:
[...]

DATE:
[...]

Do not rely on old blogs or copies of legislation if an official database is available.

Check:

- whether the law is still in force;
- whether the section has been amended;
- whether the section has been repealed;
- whether it has been renumbered;
- whether it has been changed by an implementing regulation;
- whether transitional provisions apply;
- effective date;
- date of application.

Return:

ORIGINAL TEXT
CURRENT TEXT
STATUS
AMENDMENT DATE
PRACTICAL CONSEQUENCE
SOURCE

If you cannot confirm the status using a primary source, label it:

UNVERIFIED.

Analyze the following contract:

[CONTRACT]

JURISDICTION:
[...]

MY ROLE:
[buyer / seller / employee / employer / contractor / client...]

Do not provide only a summary.

For each relevant clause, identify:

- obligation;
- deadline;
- condition;
- consequence of breach;
- termination right;
- financial risk;
- liability;
- unilateral right of the other party;
- unclear wording.

Check these in particular:

PAYMENT
TERM
AUTO-RENEWAL
TERMINATION
NOTICE PERIOD
LIABILITY
INDEMNITY
WARRANTY
IP
CONFIDENTIALITY
DATA PROTECTION
NON-COMPETE
NON-SOLICIT
ASSIGNMENT
FORCE MAJEURE
GOVERNING LAW
JURISDICTION
DISPUTE RESOLUTION

For each problematic provision, return:

CLAUSE
PLAIN-LANGUAGE MEANING
WHO BENEFITS
RISK
LEGAL CONTEXT
QUESTION TO CLARIFY
POSSIBLE NEGOTIATION POINT

Do not invent rights that are not in the contract.

Act as a contract reviewer focused on negotiation.

CONTRACT:
[...]

MY SIDE:
[...]

GOALS:
[...]

Identify clauses to:

ACCEPT
CLARIFY
NEGOTIATE
REJECT
ESCALATE TO LAWYER

For each clause:

1. state the problem;
2. explain the practical risk;
3. propose alternative wording;
4. explain what the new wording changes;
5. assess how material the change is.

Do not change the style without a legal or business reason.

Prioritize:

money;
liability;
IP;
deadlines;
termination;
exclusivity;
non-compete;
indemnity;
data protection;
jurisdiction.

Analyze my financial situation for informational purposes and through scenarios.

COUNTRY / TAX RESIDENCY:
[...]

CURRENCY:
[...]

MONTHLY NET INCOME:
[...]

FIXED EXPENSES:
[...]

VARIABLE EXPENSES:
[...]

DEBTS:
[...]

SAVINGS:
[...]

INVESTMENTS:
[...]

FINANCIAL GOALS:
[...]

TIME HORIZON:
[...]

LIQUIDITY I NEED:
[...]

Do not assume information I have not provided.

Separate these items in particular:

CASH FLOW
EMERGENCY FUND
DEBT
LIQUIDITY
INVESTMENTS
TAX CONSIDERATIONS
RISKS

If any interest rate, tax rule, inflation figure, or market data may have changed, check current sources.

Create at least:

CONSERVATIVE SCENARIO
BASE SCENARIO
ADVERSE SCENARIO

State the assumptions for each scenario explicitly.

Do not present a market forecast as fact.

Analyze this loan:

PRINCIPAL:
[...]

TERM:
[...]

NOMINAL INTEREST RATE:
[...]

APR:
[...]

FIXED OR VARIABLE:
[...]

CURRENCY:
[...]

MONTHLY PAYMENT:
[...]

FEES:
[...]

INSURANCE:
[...]

EARLY REPAYMENT:
[...]

First check that all the figures are mutually consistent.

Calculate:

- total payments;
- total interest paid;
- total fees;
- actual effective cost;
- interest as a share of total repayment.

If the rate is variable, create these scenarios:

- the base rate stays the same;
- +1 percentage point;
- +2 percentage points;
- -1 percentage point.

If the loan is tied to a reference rate, explain how a change in that rate affects the payment.

Analyze early repayment in particular:

- how much interest it saves;
- whether there are penalties;
- whether reducing the payment or the term is better if the bank lets me choose.

Do not use approximate math when we can calculate the result precisely.

Compare these two loans.

LOAN A:
[...]

LOAN B:
[...]

Do not compare them only by their monthly payments.

Compare:

PRINCIPAL
TERM
NOMINAL RATE
APR
TOTAL INTEREST
FEES
TOTAL COST
CURRENCY RISK
VARIABLE-RATE TERMS
EARLY REPAYMENT TERMS

Create a table.

Then create three scenarios:

1. I keep the loan until the end;
2. I repay it early after [X] years;
3. the interest rate rises by [X].

If information is missing, do not invent it.

Label it:

MISSING FOR AN ACCURATE COMPARISON.

Analyze this home loan.

PROPERTY PRICE:
[...]

DOWN PAYMENT:
[...]

LOAN AMOUNT:
[...]

TERM:
[...]

INTEREST RATE:
[...]

APR:
[...]

CURRENCY:
[...]

MONTHLY HOUSEHOLD INCOME:
[...]

OTHER DEBTS:
[...]

Show:

loan-to-value;
debt-service-to-income;
total interest;
total cost;
monthly share of income used for payments.

Then stress-test:

- interest rate +1 pp;
- interest rate +2 pp;
- income -20%;
- living costs +15%;
- six months without one source of income.

Calculate how large an emergency fund would be needed for:

3 months
6 months
12 months

of regular obligations.

If there are current local rules or bank limits, check official sources for the country I specified.

Compare these options for my situation:

A - keeping money in an account or deposit
B - money market or short-term instruments
C - government bonds
D - broadly diversified ETF
E - another option: [...]

DETAILS:

Country:
[...]

Currency:
[...]

Amount:
[...]

Time horizon:
[...]

How much money must be available quickly:
[...]

Goal:
[...]

Do not compare only expected returns.

Analyze:

LIQUIDITY
VOLATILITY
INFLATION RISK
CURRENCY RISK
CREDIT RISK
INTEREST-RATE RISK
TAX
FEES
TIME HORIZON

Use real returns where appropriate:

real return is approximately nominal return adjusted for inflation.

Do not assume that future stock returns are guaranteed.

Perform a fundamental analysis of:

[COMPANY]

PERIOD:
[...]

Use the latest available official financial statements.

Prioritize:

- annual report;
- quarterly report;
- regulatory filings;
- investor relations;
- earnings call or transcript;
- secondary sources only after that.

Analyze at least:

REVENUE
REVENUE GROWTH
GROSS MARGIN
OPERATING MARGIN
NET MARGIN
FREE CASH FLOW
CAPEX
CASH
DEBT
NET DEBT
INTEREST EXPENSE
SHARE COUNT
DILUTION
STOCK-BASED COMPENSATION

If the company has segments, analyze each segment separately.

If it depends on particular geographies, analyze revenue exposure by region.

Find the difference in particular between:

GAAP/IFRS METRICS
ADJUSTED/NON-GAAP METRICS

Do not use an adjusted metric without explaining what has been excluded.

Analyze the company's QUALITY OF EARNINGS:

[COMPANY]

Do not look only at net profit.

Check:

- operating cash flow;
- free cash flow;
- receivables;
- inventories;
- deferred revenue;
- one-off gains;
- restructuring charges;
- asset sales;
- stock-based compensation;
- capitalization of costs;
- changes in depreciation or amortization;
- working capital effects.

Questions:

Does profit track cash flow?

Are receivables growing faster than revenue?

Is inventory growing faster than sales?

Does the company often exclude "one-time" costs from adjusted results?

Does FCF look weaker or stronger than accounting profit?

For each finding, state the figures and periods.

Separate:

NORMAL OPERATING PERFORMANCE
ACCOUNTING EFFECT
ONE-OFF EFFECT
POTENTIAL WARNING SIGN

Analyze the company's balance sheet:

[COMPANY]

Check:

CASH
TOTAL DEBT
NET DEBT
SHORT-TERM DEBT
LONG-TERM DEBT
LEASE LIABILITIES
INTEREST EXPENSE
MATURITY SCHEDULE
CURRENT ASSETS
CURRENT LIABILITIES

Calculate where possible:

NET DEBT / EBITDA
INTEREST COVERAGE
CURRENT RATIO
QUICK RATIO

Check in particular:

- how much debt matures in the next 12, 24, or 36 months;
- the interest rates on that debt;
- how much debt has a fixed rate and how much has a variable rate;
- covenants;
- refinancing options;
- dependence on capital markets.

Do not automatically treat high nominal debt as a problem.

Assess it in relation to:

cash flow;
assets;
profitability;
maturity;
interest cost.

Value this company:

[COMPANY]

Do not use only one multiple.

Consider:

P/E
FORWARD P/E
EV/EBITDA
EV/EBIT
EV/SALES
P/FCF
FCF YIELD

Use metrics that make sense for this specific business model.

If the company is:

loss-making
high-growth
capital-intensive
a financial institution

adapt the method.

Compare it with:

the company's historical multiples;
relevant competitors;
revenue growth;
margins;
FCF;
debt.

Do not conclude:

"cheap"

just because its P/E is below average.

Check whether the lower valuation reflects a weaker business, debt, cyclicality, or falling earnings.

Build a transparent DCF model for:

[COMPANY]

Do not choose assumptions to get the valuation you want in advance.

Show all assumptions:

REVENUE GROWTH
OPERATING MARGIN
TAX RATE
D&A
CAPEX
WORKING CAPITAL
WACC
TERMINAL GROWTH

Create:

BASE CASE
BULL CASE
BEAR CASE

For each scenario, show:

FCF forecast;
terminal value;
enterprise value;
net debt adjustment;
equity value;
shares outstanding;
implied value per share.

Then create a sensitivity table for:

WACC
terminal growth.

Mark which assumptions have the greatest impact on the result.

Do not present the DCF result as a precise "true value".

Treat it as a model result that depends on assumptions.

Analyze the company's latest quarterly report:

[COMPANY]

Compare:

CURRENT QUARTER
PREVIOUS QUARTER
SAME QUARTER LAST YEAR

Analyze:

revenue;
growth;
gross margin;
operating margin;
FCF;
guidance;
segment performance;
customer metrics;
capex.

Look in particular for:

what changed in the guidance;
what management has stopped saying;
which metrics changed definition;
one-time items;
FX effects;
acquisitions.

Separate:

REPORTED RESULTS
MANAGEMENT EXPLANATION
INDEPENDENT INTERPRETATION

Do not automatically accept management's explanation as proof.

Analyze the company's earnings call:

[COMPANY]

Extract:

- key statements by management;
- guidance;
- changes in tone;
- analyst questions;
- questions management did not answer directly;
- new promises;
- risks management emphasized in particular.

Compare it with the previous two earnings calls.

Find:

what changed;
what is repeated;
which earlier promise was not fulfilled.

Separate:

FACT
MANAGEMENT CLAIM
FORWARD-LOOKING STATEMENT
INTERPRETATION

Perform due diligence on:

[COMPANY]

Do not focus only on finances.

Analyze:

BUSINESS MODEL
CUSTOMERS
CUSTOMER CONCENTRATION
SUPPLIERS
COMPETITORS
REGULATION
IP
MANAGEMENT
CAPITAL STRUCTURE
DEBT
LITIGATION
GEOGRAPHIC RISK
CYCLICALITY
DEPENDENCE ON KEY PRODUCTS
SECURITY / OPERATIONAL RISKS

For each risk, provide:

DESCRIPTION
EVIDENCE
POTENTIAL IMPACT
WHAT TO MONITOR

Look in particular for information that could challenge the company's bullish narrative.

Do not ignore negative evidence.

Analyze this ETF:

[TICKER / NAME]

Check:

INDEX
TER
FUND SIZE
DOMICILE
REPLICATION
ACCUMULATING / DISTRIBUTING
CURRENCY
HEDGING
HOLDINGS
SECTOR EXPOSURE
COUNTRY EXPOSURE
CONCENTRATION
TRACKING DIFFERENCE
LIQUIDITY

If I am in:

[COUNTRY]

also analyze the relevant tax or domicile context if reliable current sources are available.

Do not conclude that two ETFs are the same just because they track a "similar" index.

Check the specific benchmark and methodology.

Compare:

COMPANY A
[COMPANY]

COMPANY B
[COMPANY]

Do not immediately choose a "winner".

Compare:

business model;
revenue growth;
margins;
FCF;
balance sheet;
competitive position;
valuation;
cyclicality;
regulatory risk;
customer concentration.

For each category, provide the data and differences without an overall ranking.

Then identify:

the scenario in which A has stronger economics;
the scenario in which B has stronger economics;
the data point that would most change the analysis.

My investment thesis is:

"""
[THESIS]
"""

Do not try to confirm it.

Perform a red-team analysis.

Find:

- hidden assumptions;
- data that contradict it;
- risks I am overlooking;
- misinterpreted metrics;
- valuation risk;
- execution risk;
- competition;
- regulation;
- balance-sheet risk.

Then create:

WHAT MUST BE TRUE
WHAT COULD BREAK THE THESIS
LEADING INDICATORS
INVALIDATION CONDITIONS

Define specifically what would mean that my original thesis is wrong.

=== FINANCIAL ANALYSIS QA ===

Before your final answer, check:

1. whether financial data are from the same period;
2. whether we are using a trailing or forward metric;
3. whether the share count reflects diluted shares;
4. whether net debt includes relevant liabilities;
5. whether adjusted EBITDA excludes material recurring costs;
6. whether revenue growth includes acquisitions;
7. whether FX affects the comparison;
8. whether the data have been revised;
9. whether the valuation multiple uses the appropriate denominator;
10. whether the scenario assumptions are consistent with one another.

Separate:

HISTORICAL FACT
CURRENT MARKET DATA
MANAGEMENT GUIDANCE
ANALYST ESTIMATE
MODEL ASSUMPTION

Analyze this employment contract.

JURISDICTION:
[...]

EMPLOYEE STATUS:
[...]

CONTRACT:
[...]

Check these in particular:

- type of employment relationship;
- duration;
- probationary period;
- working hours;
- overtime;
- salary;
- bonuses;
- annual leave;
- sick leave;
- notice period;
- grounds for termination;
- disciplinary procedures;
- non-compete;
- confidentiality;
- IP created during employment;
- remote work;
- equipment;
- reimbursement;
- processing of personal data.

For each clause, check whether the contract could give the employee fewer rights than the statutory minimum.

Clearly flag any provision that may be void or limited by law.

Separate:

CONTRACTUAL RIGHT
STATUTORY RIGHT
POTENTIAL CONFLICT

Analyze this freelance or contractor agreement.

JURISDICTION:
[...]

MY ROLE:
[...]

Check these in particular:

SCOPE
DELIVERABLES
ACCEPTANCE CRITERIA
PAYMENT
LATE PAYMENT
EXPENSES
IP OWNERSHIP
LICENSE
CONFIDENTIALITY
LIABILITY
WARRANTIES
TERMINATION
NOTICE
NON-COMPETE
NON-SOLICIT
TAX RESPONSIBILITY
INDEPENDENT CONTRACTOR STATUS

Look in particular for a situation where the agreement formally says contractor, but the actual relationship may have characteristics of employment.

Do not automatically conclude that an employment relationship exists.

List the factors that would matter in the relevant jurisdiction.

Analyze this NDA.

JURISDICTION:
[...]

MY SIDE:
[...]

Check:

DEFINITION OF CONFIDENTIAL INFORMATION
EXCLUSIONS
PERMITTED USE
WHO MAY RECEIVE INFORMATION
SECURITY OBLIGATIONS
TERM
SURVIVAL
RETURN/DESTRUCTION
REQUIRED DISCLOSURE
IP
REMEDIES
INJUNCTION
GOVERNING LAW

Check in particular whether the definition of confidential information is:

- too broad;
- unlimited in time;
- inclusive of information already in the public domain;
- inclusive of information known before signing.

Find obligations that may be practically impossible to meet.

Propose more precise wording where needed.

Analyze the Terms of Service for:

[PRODUCT]

JURISDICTION / MARKETS:
[...]

USER TYPE:
B2C / B2B

Check:

account rules;
payment;
subscriptions;
renewal;
cancellation;
refunds;
acceptable use;
user content;
IP;
service availability;
warranties;
limitation of liability;
termination;
dispute resolution;
governing law;
consumer rights.

If the service operates in multiple jurisdictions, identify clauses that may apply differently.

Check consumer protection requirements for the relevant market in particular.

Do not assume that a US template is valid in the EU or Serbia.

Audit this Privacy Policy.

MARKETS:
[...]

SERVICE TYPE:
[...]

DATA COLLECTED:
[...]

THIRD PARTIES:
[...]

Check whether the policy clearly describes:

- categories of personal data;
- purposes;
- legal bases, where relevant;
- recipients;
- processors;
- international transfers;
- retention;
- user rights;
- cookies and tracking;
- automated decisions;
- children;
- contact information.

Then compare the policy with the actual implementation.

Look for mismatches such as:

the policy says we do not collect a data point,
but the application sends it to an analytics provider.

For each mismatch, provide:

POLICY CLAIM
ACTUAL BEHAVIOR
LEGAL/COMPLIANCE RISK
REMEDIATION

Analyze GDPR obligations for the following processing:

[PROCESSING DESCRIPTION]

COMPANY LOCATION:
[...]

USERS:
[...]

DATA:
[...]

PURPOSE:
[...]

THIRD PARTIES:
[...]

First determine whether the GDPR applies to the described scenario at all.

Then identify:

CONTROLLER
PROCESSOR
JOINT CONTROLLER, where relevant

For each processing activity, determine:

DATA CATEGORY
PURPOSE
LEGAL BASIS
RETENTION
RECIPIENTS
TRANSFER
SECURITY CONTROLS

Check these in particular:

special-category data;
profiling;
automated decisions;
children;
DPIA;
DPO requirements;
data subject requests;
breach notification.

Do not use "consent" as a universal legal basis.

Assess which basis actually fits the processing purpose.

Analyze this website's cookie and tracking setup.

JURISDICTIONS:
[...]

TRACKERS:
[...]

TOOLS:
Google Analytics
Meta Pixel
Hotjar
ads
other

Classify each cookie or tracker:

STRICTLY NECESSARY
FUNCTIONAL
ANALYTICS
ADVERTISING
OTHER

For each, check:

- when it loads;
- whether it requires consent;
- whether it loads before consent;
- retention;
- third-party transfer;
- withdrawal behavior.

Check in particular whether:

"Reject all"

really prevents trackers that were not accepted.

Do not analyze only how the banner looks.

Analyze its actual technical behavior.

Analyze this DPA.

PARTIES:
[...]

SERVICE:
[...]

DATA:
[...]

Check:

controller and processor roles;
processing instructions;
confidentiality;
security measures;
subprocessors;
international transfers;
breach notification;
data subject requests;
deletion or return;
audit rights;
liability.

Check the subprocessor clauses in particular.

Answer:

Can the processor add subprocessors independently?

Does the controller receive notice?

Is there a right to object?

Does the DPA regulate transfers outside the relevant jurisdiction?

Check the following legal claim:

"[CLAIM]"

JURISDICTION:
[...]

DATE:
[...]

Do not check only whether the claim is repeated online.

Find the legal basis.

For the claim, identify:

LAW
ARTICLE
CASE LAW, if relevant
REGULATORY GUIDANCE
EXCEPTIONS

Classify it:

TRUE
MOSTLY TRUE
TOO BROAD
DEPENDS ON THE CIRCUMSTANCES
OUTDATED
FALSE
IMPOSSIBLE TO VERIFY

Explain exactly why.

If the claim applies only under certain conditions, list them.

Find relevant case law for:

[LEGAL QUESTION]

JURISDICTION:
[...]

PERIOD:
[...]

For each decision, record:

COURT
CASE NUMBER
DATE
FACTS
LEGAL ISSUE
HOLDING
REASONING
OUTCOME
PRECEDENTIAL VALUE

Do not use a judgment just because it contains the same keywords.

Check whether the legal issue really matches my case.

If there are conflicting decisions, show both lines of authority.

Clearly label:

binding authority
persuasive authority
lower-court decision
non-final decision

Answer this question:

Is the following conduct permitted?

[DESCRIPTION]

JURISDICTION:
[...]

DATE:
[...]

Do not answer only YES or NO.

Analyze:

1. which law governs the conduct;
2. whether it is prohibited;
3. whether exceptions apply;
4. whether a permit or license is required;
5. whether there are administrative obligations;
6. what penalties may apply;
7. which facts would change the answer.

Use this structure for the conclusion:

GENERAL RULE
EXCEPTIONS
FACTS THAT MATTER
POTENTIAL CONSEQUENCES
WHAT SHOULD BE VERIFIED

Research the legal and regulatory requirements for:

[TYPE OF BUSINESS]

COUNTRY:
[...]

BUSINESS MODEL:
[...]

CUSTOMERS:
B2B / B2C

PRODUCT/SERVICE:
[...]

Find:

- permitted legal forms;
- registration;
- licenses;
- tax registration;
- fiscalization;
- VAT;
- bookkeeping;
- employment obligations;
- consumer law;
- privacy;
- sector-specific regulation.

Separate:

ONE-TIME REQUIREMENTS
ONGOING REQUIREMENTS
EVENT-TRIGGERED REQUIREMENTS

Provide an official source for each obligation.

Analyze the tax treatment of:

[SCENARIO]

COUNTRY / TAX RESIDENCY:
[...]

PERIOD:
[...]

TYPE OF INCOME:
[...]

Do not use outdated tax rates.

Check current legislation and official tax sources.

Separate:

INCOME TAX
SOCIAL CONTRIBUTIONS
VAT
WITHHOLDING TAX
CORPORATE TAX
OTHER

For each, state:

TAX BASE
RATE
THRESHOLD
EXEMPTION
FILING REQUIREMENT
PAYMENT DEADLINE

Clearly mark where the result depends on a tax status that is not yet known.

Analyze this cross-border legal scenario.

PERSON/COMPANY A:
[country]

PERSON/COMPANY B:
[country]

ACTIVITY:
[...]

PAYMENT:
[...]

DATA TRANSFER:
[...]

IP:
[...]

Identify potentially relevant jurisdictions.

Do not assume that choosing one governing law in a contract automatically removes the mandatory rules of another country.

Check:

governing law;
jurisdiction;
tax;
consumer law;
employment;
privacy and data transfers;
sanctions or export rules, if relevant;
IP.

For each issue, state:

WHICH COUNTRY'S RULE MAY APPLY
WHY
CONFLICT
WHAT NEEDS PROFESSIONAL CONFIRMATION

Analyze the IP situation:

[SCENARIO]

Types of rights to check:

COPYRIGHT
TRADEMARK
PATENT
DESIGN
TRADE SECRET
DATABASE RIGHT
LICENSE

For each asset, identify:

CREATOR
CURRENT OWNER
CONTRACTUAL TRANSFER
LICENSE
TERRITORY
TERM
RESTRICTIONS

Check these in particular:

employee-created work;
contractor-created work;
open-source components;
third-party images, fonts, or code;
AI-generated material;
brand names.

Do not assume that paying a contractor automatically transfers all IP rights.

Audit the project's open-source licenses.

DEPENDENCIES:
[...]

DISTRIBUTION MODEL:
SaaS / binary / source distribution / library

For each relevant dependency, find:

LICENSE
VERSION
COPYLEFT LEVEL
NOTICE REQUIREMENT
SOURCE-DISCLOSURE REQUIREMENT
MODIFICATION RULES
PATENT CLAUSES

Check these in particular:

GPL
AGPL
LGPL
Apache-2.0
MPL
MIT
BSD
SSPL or other source-available licenses

Do not group all "open-source" licenses together.

Identify the specific obligations triggered by our distribution model.

Analyze the legal obligations for this AI feature:

[DESCRIPTION]

JURISDICTION:
[...]

INPUT DATA:
[...]

OUTPUT:
[...]

USERS:
[...]

DECISIONS:
[...]

First identify which legal areas may be relevant:

privacy;
consumer protection;
AI-specific regulation;
discrimination;
copyright;
product liability;
professional regulation;
automated decision-making.

Check these in particular:

- whether the system makes a decision about a person;
- whether the decision has legal or similarly significant effects;
- whether human review exists;
- transparency;
- logging;
- data provenance;
- model or provider contractual terms.

Do not assume every AI feature falls into the same regulatory category.

Analyze the following legal question:

[QUESTION]

JURISDICTION:
[country / province / state / EU]

RELEVANT DATE:
[date]

AREA:
[employment / commercial / tax / contract / criminal / administrative / other]

Before reaching a conclusion, find the currently effective primary legal sources.

Prioritize:

1. official statutory text;
2. official gazette;
3. implementing regulation;
4. official regulator;
5. case law;
6. official interpretations;
7. legal scholarship;
8. secondary articles.

For each relevant law or regulation, check:

- title;
- section or article;
- whether it is currently in force;
- latest relevant amendment;
- effective date;
- transitional provisions;
- territorial scope.

Separate the answer into:

CURRENT LAW
CASE LAW
REGULATORY INTERPRETATION
POSSIBLE INTERPRETATION
UNCERTAINTY

Do not present a legal interpretation as the literal text of a statute.

Check the legal status of this provision:

[LAW / SECTION]

JURISDICTION:
[...]

DATE:
[...]

Do not rely on old blogs or copies of legislation if an official database is available.

Check:

- whether the law is still in force;
- whether the section has been amended;
- whether the section has been repealed;
- whether it has been renumbered;
- whether it has been changed by an implementing regulation;
- whether transitional provisions apply;
- effective date;
- date of application.

Return:

ORIGINAL TEXT
CURRENT TEXT
STATUS
AMENDMENT DATE
PRACTICAL CONSEQUENCE
SOURCE

If you cannot confirm the status using a primary source, label it:

UNVERIFIED.

Analyze the following contract:

[CONTRACT]

JURISDICTION:
[...]

MY ROLE:
[buyer / seller / employee / employer / contractor / client...]

Do not provide only a summary.

For each relevant clause, identify:

- obligation;
- deadline;
- condition;
- consequence of breach;
- termination right;
- financial risk;
- liability;
- unilateral right of the other party;
- unclear wording.

Check these in particular:

PAYMENT
TERM
AUTO-RENEWAL
TERMINATION
NOTICE PERIOD
LIABILITY
INDEMNITY
WARRANTY
IP
CONFIDENTIALITY
DATA PROTECTION
NON-COMPETE
NON-SOLICIT
ASSIGNMENT
FORCE MAJEURE
GOVERNING LAW
JURISDICTION
DISPUTE RESOLUTION

For each problematic provision, return:

CLAUSE
PLAIN-LANGUAGE MEANING
WHO BENEFITS
RISK
LEGAL CONTEXT
QUESTION TO CLARIFY
POSSIBLE NEGOTIATION POINT

Do not invent rights that are not in the contract.

Act as a contract reviewer focused on negotiation.

CONTRACT:
[...]

MY SIDE:
[...]

GOALS:
[...]

Identify clauses to:

ACCEPT
CLARIFY
NEGOTIATE
REJECT
ESCALATE TO LAWYER

For each clause:

1. state the problem;
2. explain the practical risk;
3. propose alternative wording;
4. explain what the new wording changes;
5. assess how material the change is.

Do not change the style without a legal or business reason.

Prioritize:

money;
liability;
IP;
deadlines;
termination;
exclusivity;
non-compete;
indemnity;
data protection;
jurisdiction.

Analyze this employment contract.

JURISDICTION:
[...]

EMPLOYEE STATUS:
[...]

CONTRACT:
[...]

Check these in particular:

- type of employment relationship;
- duration;
- probationary period;
- working hours;
- overtime;
- salary;
- bonuses;
- annual leave;
- sick leave;
- notice period;
- grounds for termination;
- disciplinary procedures;
- non-compete;
- confidentiality;
- IP created during employment;
- remote work;
- equipment;
- reimbursement;
- processing of personal data.

For each clause, check whether the contract could give the employee fewer rights than the statutory minimum.

Clearly flag any provision that may be void or limited by law.

Separate:

CONTRACTUAL RIGHT
STATUTORY RIGHT
POTENTIAL CONFLICT

Analyze this freelance or contractor agreement.

JURISDICTION:
[...]

MY ROLE:
[...]

Check these in particular:

SCOPE
DELIVERABLES
ACCEPTANCE CRITERIA
PAYMENT
LATE PAYMENT
EXPENSES
IP OWNERSHIP
LICENSE
CONFIDENTIALITY
LIABILITY
WARRANTIES
TERMINATION
NOTICE
NON-COMPETE
NON-SOLICIT
TAX RESPONSIBILITY
INDEPENDENT CONTRACTOR STATUS

Look in particular for a situation where the agreement formally says contractor, but the actual relationship may have characteristics of employment.

Do not automatically conclude that an employment relationship exists.

List the factors that would matter in the relevant jurisdiction.

Analyze this NDA.

JURISDICTION:
[...]

MY SIDE:
[...]

Check:

DEFINITION OF CONFIDENTIAL INFORMATION
EXCLUSIONS
PERMITTED USE
WHO MAY RECEIVE INFORMATION
SECURITY OBLIGATIONS
TERM
SURVIVAL
RETURN/DESTRUCTION
REQUIRED DISCLOSURE
IP
REMEDIES
INJUNCTION
GOVERNING LAW

Check in particular whether the definition of confidential information is:

- too broad;
- unlimited in time;
- inclusive of information already in the public domain;
- inclusive of information known before signing.

Find obligations that may be practically impossible to meet.

Propose more precise wording where needed.

Analyze the Terms of Service for:

[PRODUCT]

JURISDICTION / MARKETS:
[...]

USER TYPE:
B2C / B2B

Check:

account rules;
payment;
subscriptions;
renewal;
cancellation;
refunds;
acceptable use;
user content;
IP;
service availability;
warranties;
limitation of liability;
termination;
dispute resolution;
governing law;
consumer rights.

If the service operates in multiple jurisdictions, identify clauses that may apply differently.

Check consumer protection requirements for the relevant market in particular.

Do not assume that a US template is valid in the EU or Serbia.

Audit this Privacy Policy.

MARKETS:
[...]

SERVICE TYPE:
[...]

DATA COLLECTED:
[...]

THIRD PARTIES:
[...]

Check whether the policy clearly describes:

- categories of personal data;
- purposes;
- legal bases, where relevant;
- recipients;
- processors;
- international transfers;
- retention;
- user rights;
- cookies and tracking;
- automated decisions;
- children;
- contact information.

Then compare the policy with the actual implementation.

Look for mismatches such as:

the policy says we do not collect a data point,
but the application sends it to an analytics provider.

For each mismatch, provide:

POLICY CLAIM
ACTUAL BEHAVIOR
LEGAL/COMPLIANCE RISK
REMEDIATION

Analyze GDPR obligations for the following processing:

[PROCESSING DESCRIPTION]

COMPANY LOCATION:
[...]

USERS:
[...]

DATA:
[...]

PURPOSE:
[...]

THIRD PARTIES:
[...]

First determine whether the GDPR applies to the described scenario at all.

Then identify:

CONTROLLER
PROCESSOR
JOINT CONTROLLER, where relevant

For each processing activity, determine:

DATA CATEGORY
PURPOSE
LEGAL BASIS
RETENTION
RECIPIENTS
TRANSFER
SECURITY CONTROLS

Check these in particular:

special-category data;
profiling;
automated decisions;
children;
DPIA;
DPO requirements;
data subject requests;
breach notification.

Do not use "consent" as a universal legal basis.

Assess which basis actually fits the processing purpose.

Analyze this website's cookie and tracking setup.

JURISDICTIONS:
[...]

TRACKERS:
[...]

TOOLS:
Google Analytics
Meta Pixel
Hotjar
ads
other

Classify each cookie or tracker:

STRICTLY NECESSARY
FUNCTIONAL
ANALYTICS
ADVERTISING
OTHER

For each, check:

- when it loads;
- whether it requires consent;
- whether it loads before consent;
- retention;
- third-party transfer;
- withdrawal behavior.

Check in particular whether:

"Reject all"

really prevents trackers that were not accepted.

Do not analyze only how the banner looks.

Analyze its actual technical behavior.

Analyze this DPA.

PARTIES:
[...]

SERVICE:
[...]

DATA:
[...]

Check:

controller and processor roles;
processing instructions;
confidentiality;
security measures;
subprocessors;
international transfers;
breach notification;
data subject requests;
deletion or return;
audit rights;
liability.

Check the subprocessor clauses in particular.

Answer:

Can the processor add subprocessors independently?

Does the controller receive notice?

Is there a right to object?

Does the DPA regulate transfers outside the relevant jurisdiction?

Check the following legal claim:

"[CLAIM]"

JURISDICTION:
[...]

DATE:
[...]

Do not check only whether the claim is repeated online.

Find the legal basis.

For the claim, identify:

LAW
ARTICLE
CASE LAW, if relevant
REGULATORY GUIDANCE
EXCEPTIONS

Classify it:

TRUE
MOSTLY TRUE
TOO BROAD
DEPENDS ON THE CIRCUMSTANCES
OUTDATED
FALSE
IMPOSSIBLE TO VERIFY

Explain exactly why.

If the claim applies only under certain conditions, list them.

Find relevant case law for:

[LEGAL QUESTION]

JURISDICTION:
[...]

PERIOD:
[...]

For each decision, record:

COURT
CASE NUMBER
DATE
FACTS
LEGAL ISSUE
HOLDING
REASONING
OUTCOME
PRECEDENTIAL VALUE

Do not use a judgment just because it contains the same keywords.

Check whether the legal issue really matches my case.

If there are conflicting decisions, show both lines of authority.

Clearly label:

binding authority
persuasive authority
lower-court decision
non-final decision

Answer this question:

Is the following conduct permitted?

[DESCRIPTION]

JURISDICTION:
[...]

DATE:
[...]

Do not answer only YES or NO.

Analyze:

1. which law governs the conduct;
2. whether it is prohibited;
3. whether exceptions apply;
4. whether a permit or license is required;
5. whether there are administrative obligations;
6. what penalties may apply;
7. which facts would change the answer.

Use this structure for the conclusion:

GENERAL RULE
EXCEPTIONS
FACTS THAT MATTER
POTENTIAL CONSEQUENCES
WHAT SHOULD BE VERIFIED

Research the legal and regulatory requirements for:

[TYPE OF BUSINESS]

COUNTRY:
[...]

BUSINESS MODEL:
[...]

CUSTOMERS:
B2B / B2C

PRODUCT/SERVICE:
[...]

Find:

- permitted legal forms;
- registration;
- licenses;
- tax registration;
- fiscalization;
- VAT;
- bookkeeping;
- employment obligations;
- consumer law;
- privacy;
- sector-specific regulation.

Separate:

ONE-TIME REQUIREMENTS
ONGOING REQUIREMENTS
EVENT-TRIGGERED REQUIREMENTS

Provide an official source for each obligation.

Analyze the tax treatment of:

[SCENARIO]

COUNTRY / TAX RESIDENCY:
[...]

PERIOD:
[...]

TYPE OF INCOME:
[...]

Do not use outdated tax rates.

Check current legislation and official tax sources.

Separate:

INCOME TAX
SOCIAL CONTRIBUTIONS
VAT
WITHHOLDING TAX
CORPORATE TAX
OTHER

For each, state:

TAX BASE
RATE
THRESHOLD
EXEMPTION
FILING REQUIREMENT
PAYMENT DEADLINE

Clearly mark where the result depends on a tax status that is not yet known.

Analyze this cross-border legal scenario.

PERSON/COMPANY A:
[country]

PERSON/COMPANY B:
[country]

ACTIVITY:
[...]

PAYMENT:
[...]

DATA TRANSFER:
[...]

IP:
[...]

Identify potentially relevant jurisdictions.

Do not assume that choosing one governing law in a contract automatically removes the mandatory rules of another country.

Check:

governing law;
jurisdiction;
tax;
consumer law;
employment;
privacy and data transfers;
sanctions or export rules, if relevant;
IP.

For each issue, state:

WHICH COUNTRY'S RULE MAY APPLY
WHY
CONFLICT
WHAT NEEDS PROFESSIONAL CONFIRMATION

Analyze the IP situation:

[SCENARIO]

Types of rights to check:

COPYRIGHT
TRADEMARK
PATENT
DESIGN
TRADE SECRET
DATABASE RIGHT
LICENSE

For each asset, identify:

CREATOR
CURRENT OWNER
CONTRACTUAL TRANSFER
LICENSE
TERRITORY
TERM
RESTRICTIONS

Check these in particular:

employee-created work;
contractor-created work;
open-source components;
third-party images, fonts, or code;
AI-generated material;
brand names.

Do not assume that paying a contractor automatically transfers all IP rights.

Audit the project's open-source licenses.

DEPENDENCIES:
[...]

DISTRIBUTION MODEL:
SaaS / binary / source distribution / library

For each relevant dependency, find:

LICENSE
VERSION
COPYLEFT LEVEL
NOTICE REQUIREMENT
SOURCE-DISCLOSURE REQUIREMENT
MODIFICATION RULES
PATENT CLAUSES

Check these in particular:

GPL
AGPL
LGPL
Apache-2.0
MPL
MIT
BSD
SSPL or other source-available licenses

Do not group all "open-source" licenses together.

Identify the specific obligations triggered by our distribution model.

Analyze the legal obligations for this AI feature:

[DESCRIPTION]

JURISDICTION:
[...]

INPUT DATA:
[...]

OUTPUT:
[...]

USERS:
[...]

DECISIONS:
[...]

First identify which legal areas may be relevant:

privacy;
consumer protection;
AI-specific regulation;
discrimination;
copyright;
product liability;
professional regulation;
automated decision-making.

Check these in particular:

- whether the system makes a decision about a person;
- whether the decision has legal or similarly significant effects;
- whether human review exists;
- transparency;
- logging;
- data provenance;
- model or provider contractual terms.

Do not assume every AI feature falls into the same regulatory category.

Perform a gap analysis between:

REQUIREMENT:
[law / standard / regulation]

and

CURRENT STATE:
[system description]

For each requirement, create:

REQUIREMENT
SOURCE
CURRENT CONTROL
EVIDENCE
STATUS
GAP
REMEDIATION
OWNER
VERIFICATION

STATUS:

COMPLIANT
PARTIALLY COMPLIANT
NON-COMPLIANT
NOT APPLICABLE
UNVERIFIED

Do not mark COMPLIANT just because the company says it has a process.

Look for evidence that it has been implemented.

Perform legal due diligence on this company.

COMPANY:
[...]

JURISDICTION:
[...]

Analyze:

corporate status;
ownership;
licenses;
material contracts;
debt and security interests;
litigation;
employment;
IP;
privacy;
regulatory matters;
tax disputes;
real estate;
insurance.

For each issue, state:

ISSUE
SOURCE
SEVERITY
POTENTIAL LIABILITY
TRANSACTION IMPACT
FOLLOW-UP DOCUMENT

Separate:

CONFIRMED ISSUE
POTENTIAL ISSUE
MISSING DOCUMENTATION

Now audit the previous legal analysis as an independent reviewer.

Look for:

- incorrect jurisdiction;
- repealed law;
- outdated version of an article;
- incorrect effective date;
- nonexistent court decision;
- a judgment that does not support the claim;
- regulatory guidance presented as law;
- a secondary source presented as primary;
- an exception that changes the conclusion;
- a mandatory rule that a contract cannot exclude.

For each claim, label it:

CONFIRMED
NEEDS QUALIFICATION
OUTDATED
UNSUPPORTED
INCORRECT

Correct the analysis before giving your final answer.

=== LEGAL QA ===

Before your final answer, check:

1. jurisdiction;
2. relevant date;
3. current status of the law;
4. latest amendments;
5. effective date;
6. transitional provisions;
7. whether the cited section says what we claim;
8. whether the court decision actually supports the conclusion;
9. whether an exception applies;
10. whether an interpretation is presented as fact;
11. whether regulatory guidance is binding or only guidance;
12. whether a contract clause may be limited by mandatory law.

If there is not enough evidence:

UNVERIFIED

is better than invented certainty.

Analyze the following health question for educational purposes:

[QUESTION]

RELEVANT DETAILS

Age:
[...]

Sex:
[...]

Symptoms:
[...]

When they started:
[...]

How they are changing:
[...]

Known conditions:
[...]

Medications and supplements:
[...]

Allergies:
[...]

Lab results:
[...]

Other relevant details:
[...]

Use current professional guidelines and high-quality medical sources when the answer depends on current recommendations.

Prioritize these sources:

1. professional clinical guidelines;
2. regulatory documentation;
3. systematic reviews;
4. meta-analyses;
5. high-quality peer-reviewed studies;
6. official health institutions.

Do not make a definite diagnosis from incomplete information.

Separate:

POSSIBLE EXPLANATIONS
WHAT SUPPORTS EACH EXPLANATION
WHAT CONTRADICTS IT
MISSING INFORMATION
RED FLAGS
WHAT A DOCTOR WOULD TYPICALLY CHECK
QUESTIONS TO ASK A DOCTOR

Clearly identify any symptoms that may require urgent medical assessment.

I have these symptoms:

[SYMPTOMS]

Do not immediately try to choose one disease.

Create a structured differential analysis.

For each possible explanation, show:

NAME
WHY IT IS POSSIBLE
WHAT FITS
WHAT DOES NOT FIT
HOW URGENT IT IS TO CONSIDER
WHICH DETAIL WOULD HELP MOST

Group the explanations into:

COMMON AND USUALLY LESS DANGEROUS
LESS COMMON
IMPORTANT TO RULE OUT
POTENTIALLY URGENT

Do not use a rare disease as the main explanation just because it can cause the same symptom.

Clearly identify symptoms that typically warrant urgent medical assessment.

Analyze this symptom:

[SYMPTOM]

I want an assessment of warning signs only.

Create two groups:

RED FLAGS: symptoms or circumstances that may require urgent or prompt medical assessment

NON-URGENT FEATURES: characteristics more often associated with less urgent causes

For each red flag, explain:

what it means;
why it matters;
how quickly assessment is typically sought.

Do not use red flags to diagnose a condition.

Their purpose is to assess urgency.

Explain these lab results:

[RESULTS]

Use the reference ranges shown on the laboratory report when available.

For each parameter, show:

PARAMETER
RESULT
REFERENCE RANGE
STATUS
WHAT THE PARAMETER MEASURES
POSSIBLE REASONS FOR AN ABNORMAL RESULT
WHICH OTHER PARAMETERS ARE CONSIDERED TOGETHER

Do not diagnose based on one abnormal value alone.

Check these in particular:

- units;
- sex;
- age;
- pregnancy, if relevant;
- sample collection time;
- fasting or non-fasting status;
- medications that may affect the result.

At the end, identify:

RESULTS THAT ARE PROBABLY NOT SIGNIFICANT IN ISOLATION
RESULTS WORTH DISCUSSING WITH A DOCTOR
RESULTS THAT MAY REQUIRE A FASTER ASSESSMENT

Analyze the complete blood count.

DETAILS:

[RESULTS]

Do not look at parameters in isolation.

Analyze these together:

RBC
HGB
HCT
MCV
MCH
MCHC
RDW

then:

WBC
NEUTROPHILS
LYMPHOCYTES
MONOCYTES
EOSINOPHILS
BASOPHILS

and:

PLT
MPV

If there is a pattern that could be consistent with anemia, infection, or another process, explain it only as a possible interpretation.

For anemia, distinguish these possible patterns in particular:

microcytic
normocytic
macrocytic

and explain which additional results would help distinguish them.

Explain these liver-related lab results:

ALT:
AST:
GGT:
ALP:
bilirubin:
albumin:
INR:
[...]

Do not automatically call all elevated enzymes "liver disease".

Analyze the pattern:

HEPATOCELLULAR
CHOLESTATIC
MIXED

if the available data allow it.

Take into account:

medications;
alcohol;
physical activity;
metabolic factors;
infections;
bile ducts.

Explain which additional results are typically considered together.

Clearly identify any results or symptoms that may require faster medical assessment.

Analyze:

creatinine:
eGFR:
urea:
potassium:
sodium:
urine:
urine albumin/protein:
[...]

Take into account:

age;
sex;
muscle mass;
hydration;
medications.

Do not diagnose chronic kidney disease based on one measurement.

Explain the difference between:

a currently abnormal result
and
a chronic problem that must persist over time.

Explain the role of these in particular:

eGFR
albuminuria/proteinuria
electrolytes.

Analyze these results:

TSH:
FT4:
FT3:
anti-TPO:
anti-Tg:
TRAb:
[...]

Consider the relationships between TSH and thyroid hormones.

Do not interpret antibodies in isolation as proof of a functional disorder.

Explain possible patterns:

primary hypothyroidism;
subclinical hypothyroidism;
hyperthyroidism;
subclinical hyperthyroidism;
central disorders;

only if the results are consistent with them.

Take into account:

pregnancy;
medications;
acute illness;
biotin supplementation, if relevant.

At the end, state what would usually be checked next.

Explain these results:

fasting glucose:
HbA1c:
OGTT:
insulin:
[...]

For each test, explain:

what it measures;
which time period it represents;
the test's limitations.

Do not interpret insulin or HOMA-IR without appropriate context.

Distinguish:

normal
prediabetes
diabetes

only according to currently applicable professional criteria.

If guidelines require confirmation with another measurement, state that.

Consider circumstances that may affect HbA1c in particular.

Analyze this lipid profile:

total cholesterol:
LDL-C:
HDL-C:
triglycerides:
non-HDL:
ApoB:
Lp(a):
[...]

Do not interpret total cholesterol alone.

Explain the role of these measures:

LDL-C
non-HDL
ApoB
Lp(a)
triglycerides

when available.

Assess their significance in the context of overall cardiovascular risk, not just one number.

Relevant factors:

age;
blood pressure;
smoking;
diabetes;
known cardiovascular disease;
family history;
kidney disease.

Do not recommend treatment based on one lipid result without the necessary clinical context.

Explain this medication:

[MEDICATION NAME]

Use current official documentation and high-quality medical sources.

Explain:

INDICATIONS
MECHANISM OF ACTION
COMMON ADVERSE REACTIONS
IMPORTANT SERIOUS ADVERSE REACTIONS
MAIN CONTRAINDICATIONS
INTERACTIONS
SPECIAL POPULATIONS
MONITORING

Separate:

common
less common
rare but serious.

Do not change a prescribed dose or recommend stopping treatment without appropriate clinical context.

If suddenly stopping a medication could be a problem, clearly warn that treatment changes should be discussed with a doctor.

Check potential interactions between:

MEDICATION 1:
[...]

MEDICATION 2:
[...]

MEDICATION 3:
[...]

SUPPLEMENTS:
[...]

Relevant health condition:
[...]

Use current reliable sources.

For each potential interaction, state:

DRUG A
DRUG B
MECHANISM
CLINICAL SIGNIFICANCE
EVIDENCE
POSSIBLE CONSEQUENCE
WHAT IS TYPICALLY MONITORED

Distinguish:

MAJOR
MODERATE
MINOR
THEORETICAL

Do not treat all interactions as equally dangerous.

Check combinations that may affect these in particular:

bleeding;
QT interval;
serotonin;
blood pressure;
kidneys;
liver;
sedation;
potassium;
glucose.

Check the interaction between:

MEDICATION:
[...]

FOOD / SUPPLEMENT:
[...]

Explain:

whether an interaction exists;
mechanism;
clinical significance;
whether it depends on dose;
whether it depends on timing;
which symptoms or results can be monitored.

Do not assume a supplement is safe just because it is "natural".

Compare two treatment options for:

[CONDITION]

OPTION A:
[...]

OPTION B:
[...]

Use current clinical guidelines and high-quality evidence.

Compare:

EFFICACY
ABSOLUTE BENEFIT, when available
SIDE EFFECTS
SERIOUS ADVERSE EVENTS
CONTRAINDICATIONS
INTERACTIONS
MONITORING
CONVENIENCE
QUALITY OF EVIDENCE

Give head-to-head studies priority when available.

If the comparison relies only on separate studies, warn that indirect comparisons have limitations.

Do not choose a treatment for a specific person without enough individual information.

Research the following clinical question using PICO.

P - POPULATION:
[...]

I - INTERVENTION:
[...]

C - COMPARISON:
[...]

O - OUTCOME:
[...]

Organize the search in this order:

1. current clinical guidelines;
2. systematic reviews;
3. meta-analyses;
4. randomized controlled trials;
5. large observational studies if RCTs are unavailable.

For each key piece of evidence, record:

STUDY
YEAR
POPULATION
N
INTERVENTION
COMPARATOR
OUTCOME
EFFECT SIZE
CONFIDENCE INTERVAL
FOLLOW-UP
LIMITATIONS

At the end, assess how directly the evidence applies to the stated population.

Conduct a structured literature review on:

[TOPIC]

PERIOD:
[...]

POPULATION:
[...]

Do not just find several papers that support the same idea.

Look for:

systematic reviews;
meta-analyses;
RCTs;
large observational studies;
relevant guidelines.

For each study, check:

study design;
sample;
population;
primary endpoint;
effect size;
confidence interval;
dropout;
follow-up;
risk of bias;
funding;
conflicts of interest.

Look in particular for studies with negative or neutral results.

If the studies disagree, analyze possible reasons for heterogeneity.

Check the following health claim:

"[CLAIM]"

Do not check how many times the claim is repeated.

Check the quality of the evidence.

Look for:

mechanistic evidence;
human observational studies;
RCTs;
systematic reviews;
clinical guidelines.

Separate:

BIOLOGICALLY PLAUSIBLE
OBSERVATIONAL ASSOCIATION
CLINICAL EVIDENCE
PROVEN BENEFIT
UNPROVEN
REFUTED

If a study shows an effect only:

in vitro
in animals
or on a biomarker

do not present it as a proven clinical effect in humans.

Analyze this claim:

"[TREATMENT REDUCES RISK BY X%]"

Find the original data.

Calculate where possible:

CONTROL EVENT RATE
TREATMENT EVENT RATE
RELATIVE RISK REDUCTION
ABSOLUTE RISK REDUCTION
NNT

If there are adverse events, calculate:

ABSOLUTE RISK INCREASE
NNH

Do not highlight a relative change without absolute risk.

Example:

"risk reduced by 50 percent"

could mean:

2% -> 1%

which is an absolute reduction of 1 percentage point.

Show both.

Analyze this screening test:

[TEST]

POPULATION:
[...]

Explain:

SENSITIVITY
SPECIFICITY
POSITIVE PREDICTIVE VALUE
NEGATIVE PREDICTIVE VALUE
PREVALENCE EFFECT
FALSE POSITIVES
FALSE NEGATIVES

Do not claim that a positive screening result automatically means disease.

If different professional organizations have recommendations, compare:

AGE
RISK GROUP
SCREENING INTERVAL
BENEFITS
HARMS

Explain this diagnostic test:

[TEST]

What is it used for?

What does a positive result mean?

What does a negative result mean?

What are its sensitivity and specificity, if reliable data are available?

Does the result depend on:

disease stage;
testing time;
previous treatment;
sample quality?

Does the test:

confirm a diagnosis
or
only increase or decrease its probability?

Explain what is typically done if the result does not match the clinical picture.

Explain the following radiology report in plain language:

[REPORT]

Do not try to make a new diagnosis from an image or short description beyond what the report says.

For each term, explain:

what it means;
where it is located;
whether the radiologist describes a definite finding or a possibility;
whether additional follow-up is recommended.

Clearly identify:

INCIDENTAL FINDINGS
URGENT FINDINGS
FOLLOW-UP RECOMMENDATIONS

If the finding depends on comparison with an older scan, say so.

Analyze the following health question for educational purposes:

[QUESTION]

RELEVANT DETAILS

Age:
[...]

Sex:
[...]

Symptoms:
[...]

When they started:
[...]

How they are changing:
[...]

Known conditions:
[...]

Medications and supplements:
[...]

Allergies:
[...]

Lab results:
[...]

Other relevant details:
[...]

Use current professional guidelines and high-quality medical sources when the answer depends on current recommendations.

Prioritize these sources:

1. professional clinical guidelines;
2. regulatory documentation;
3. systematic reviews;
4. meta-analyses;
5. high-quality peer-reviewed studies;
6. official health institutions.

Do not make a definite diagnosis from incomplete information.

Separate:

POSSIBLE EXPLANATIONS
WHAT SUPPORTS EACH EXPLANATION
WHAT CONTRADICTS IT
MISSING INFORMATION
RED FLAGS
WHAT A DOCTOR WOULD TYPICALLY CHECK
QUESTIONS TO ASK A DOCTOR

Clearly identify any symptoms that may require urgent medical assessment.

I have these symptoms:

[SYMPTOMS]

Do not immediately try to choose one disease.

Create a structured differential analysis.

For each possible explanation, show:

NAME
WHY IT IS POSSIBLE
WHAT FITS
WHAT DOES NOT FIT
HOW URGENT IT IS TO CONSIDER
WHICH DETAIL WOULD HELP MOST

Group the explanations into:

COMMON AND USUALLY LESS DANGEROUS
LESS COMMON
IMPORTANT TO RULE OUT
POTENTIALLY URGENT

Do not use a rare disease as the main explanation just because it can cause the same symptom.

Clearly identify symptoms that typically warrant urgent medical assessment.

Analyze this symptom:

[SYMPTOM]

I want an assessment of warning signs only.

Create two groups:

RED FLAGS: symptoms or circumstances that may require urgent or prompt medical assessment

NON-URGENT FEATURES: characteristics more often associated with less urgent causes

For each red flag, explain:

what it means;
why it matters;
how quickly assessment is typically sought.

Do not use red flags to diagnose a condition.

Their purpose is to assess urgency.

Explain these lab results:

[RESULTS]

Use the reference ranges shown on the laboratory report when available.

For each parameter, show:

PARAMETER
RESULT
REFERENCE RANGE
STATUS
WHAT THE PARAMETER MEASURES
POSSIBLE REASONS FOR AN ABNORMAL RESULT
WHICH OTHER PARAMETERS ARE CONSIDERED TOGETHER

Do not diagnose based on one abnormal value alone.

Check these in particular:

- units;
- sex;
- age;
- pregnancy, if relevant;
- sample collection time;
- fasting or non-fasting status;
- medications that may affect the result.

At the end, identify:

RESULTS THAT ARE PROBABLY NOT SIGNIFICANT IN ISOLATION
RESULTS WORTH DISCUSSING WITH A DOCTOR
RESULTS THAT MAY REQUIRE A FASTER ASSESSMENT

Analyze the complete blood count.

DETAILS:

[RESULTS]

Do not look at parameters in isolation.

Analyze these together:

RBC
HGB
HCT
MCV
MCH
MCHC
RDW

then:

WBC
NEUTROPHILS
LYMPHOCYTES
MONOCYTES
EOSINOPHILS
BASOPHILS

and:

PLT
MPV

If there is a pattern that could be consistent with anemia, infection, or another process, explain it only as a possible interpretation.

For anemia, distinguish these possible patterns in particular:

microcytic
normocytic
macrocytic

and explain which additional results would help distinguish them.

Explain these liver-related lab results:

ALT:
AST:
GGT:
ALP:
bilirubin:
albumin:
INR:
[...]

Do not automatically call all elevated enzymes "liver disease".

Analyze the pattern:

HEPATOCELLULAR
CHOLESTATIC
MIXED

if the available data allow it.

Take into account:

medications;
alcohol;
physical activity;
metabolic factors;
infections;
bile ducts.

Explain which additional results are typically considered together.

Clearly identify any results or symptoms that may require faster medical assessment.

Analyze:

creatinine:
eGFR:
urea:
potassium:
sodium:
urine:
urine albumin/protein:
[...]

Take into account:

age;
sex;
muscle mass;
hydration;
medications.

Do not diagnose chronic kidney disease based on one measurement.

Explain the difference between:

a currently abnormal result
and
a chronic problem that must persist over time.

Explain the role of these in particular:

eGFR
albuminuria/proteinuria
electrolytes.

Analyze these results:

TSH:
FT4:
FT3:
anti-TPO:
anti-Tg:
TRAb:
[...]

Consider the relationships between TSH and thyroid hormones.

Do not interpret antibodies in isolation as proof of a functional disorder.

Explain possible patterns:

primary hypothyroidism;
subclinical hypothyroidism;
hyperthyroidism;
subclinical hyperthyroidism;
central disorders;

only if the results are consistent with them.

Take into account:

pregnancy;
medications;
acute illness;
biotin supplementation, if relevant.

At the end, state what would usually be checked next.

Explain these results:

fasting glucose:
HbA1c:
OGTT:
insulin:
[...]

For each test, explain:

what it measures;
which time period it represents;
the test's limitations.

Do not interpret insulin or HOMA-IR without appropriate context.

Distinguish:

normal
prediabetes
diabetes

only according to currently applicable professional criteria.

If guidelines require confirmation with another measurement, state that.

Consider circumstances that may affect HbA1c in particular.

Analyze this lipid profile:

total cholesterol:
LDL-C:
HDL-C:
triglycerides:
non-HDL:
ApoB:
Lp(a):
[...]

Do not interpret total cholesterol alone.

Explain the role of these measures:

LDL-C
non-HDL
ApoB
Lp(a)
triglycerides

when available.

Assess their significance in the context of overall cardiovascular risk, not just one number.

Relevant factors:

age;
blood pressure;
smoking;
diabetes;
known cardiovascular disease;
family history;
kidney disease.

Do not recommend treatment based on one lipid result without the necessary clinical context.

Explain this medication:

[MEDICATION NAME]

Use current official documentation and high-quality medical sources.

Explain:

INDICATIONS
MECHANISM OF ACTION
COMMON ADVERSE REACTIONS
IMPORTANT SERIOUS ADVERSE REACTIONS
MAIN CONTRAINDICATIONS
INTERACTIONS
SPECIAL POPULATIONS
MONITORING

Separate:

common
less common
rare but serious.

Do not change a prescribed dose or recommend stopping treatment without appropriate clinical context.

If suddenly stopping a medication could be a problem, clearly warn that treatment changes should be discussed with a doctor.

Check potential interactions between:

MEDICATION 1:
[...]

MEDICATION 2:
[...]

MEDICATION 3:
[...]

SUPPLEMENTS:
[...]

Relevant health condition:
[...]

Use current reliable sources.

For each potential interaction, state:

DRUG A
DRUG B
MECHANISM
CLINICAL SIGNIFICANCE
EVIDENCE
POSSIBLE CONSEQUENCE
WHAT IS TYPICALLY MONITORED

Distinguish:

MAJOR
MODERATE
MINOR
THEORETICAL

Do not treat all interactions as equally dangerous.

Check combinations that may affect these in particular:

bleeding;
QT interval;
serotonin;
blood pressure;
kidneys;
liver;
sedation;
potassium;
glucose.

Check the interaction between:

MEDICATION:
[...]

FOOD / SUPPLEMENT:
[...]

Explain:

whether an interaction exists;
mechanism;
clinical significance;
whether it depends on dose;
whether it depends on timing;
which symptoms or results can be monitored.

Do not assume a supplement is safe just because it is "natural".

Compare two treatment options for:

[CONDITION]

OPTION A:
[...]

OPTION B:
[...]

Use current clinical guidelines and high-quality evidence.

Compare:

EFFICACY
ABSOLUTE BENEFIT, when available
SIDE EFFECTS
SERIOUS ADVERSE EVENTS
CONTRAINDICATIONS
INTERACTIONS
MONITORING
CONVENIENCE
QUALITY OF EVIDENCE

Give head-to-head studies priority when available.

If the comparison relies only on separate studies, warn that indirect comparisons have limitations.

Do not choose a treatment for a specific person without enough individual information.

Research the following clinical question using PICO.

P - POPULATION:
[...]

I - INTERVENTION:
[...]

C - COMPARISON:
[...]

O - OUTCOME:
[...]

Organize the search in this order:

1. current clinical guidelines;
2. systematic reviews;
3. meta-analyses;
4. randomized controlled trials;
5. large observational studies if RCTs are unavailable.

For each key piece of evidence, record:

STUDY
YEAR
POPULATION
N
INTERVENTION
COMPARATOR
OUTCOME
EFFECT SIZE
CONFIDENCE INTERVAL
FOLLOW-UP
LIMITATIONS

At the end, assess how directly the evidence applies to the stated population.

Conduct a structured literature review on:

[TOPIC]

PERIOD:
[...]

POPULATION:
[...]

Do not just find several papers that support the same idea.

Look for:

systematic reviews;
meta-analyses;
RCTs;
large observational studies;
relevant guidelines.

For each study, check:

study design;
sample;
population;
primary endpoint;
effect size;
confidence interval;
dropout;
follow-up;
risk of bias;
funding;
conflicts of interest.

Look in particular for studies with negative or neutral results.

If the studies disagree, analyze possible reasons for heterogeneity.

Check the following health claim:

"[CLAIM]"

Do not check how many times the claim is repeated.

Check the quality of the evidence.

Look for:

mechanistic evidence;
human observational studies;
RCTs;
systematic reviews;
clinical guidelines.

Separate:

BIOLOGICALLY PLAUSIBLE
OBSERVATIONAL ASSOCIATION
CLINICAL EVIDENCE
PROVEN BENEFIT
UNPROVEN
REFUTED

If a study shows an effect only:

in vitro
in animals
or on a biomarker

do not present it as a proven clinical effect in humans.

Analyze this claim:

"[TREATMENT REDUCES RISK BY X%]"

Find the original data.

Calculate where possible:

CONTROL EVENT RATE
TREATMENT EVENT RATE
RELATIVE RISK REDUCTION
ABSOLUTE RISK REDUCTION
NNT

If there are adverse events, calculate:

ABSOLUTE RISK INCREASE
NNH

Do not highlight a relative change without absolute risk.

Example:

"risk reduced by 50 percent"

could mean:

2% -> 1%

which is an absolute reduction of 1 percentage point.

Show both.

Analyze this screening test:

[TEST]

POPULATION:
[...]

Explain:

SENSITIVITY
SPECIFICITY
POSITIVE PREDICTIVE VALUE
NEGATIVE PREDICTIVE VALUE
PREVALENCE EFFECT
FALSE POSITIVES
FALSE NEGATIVES

Do not claim that a positive screening result automatically means disease.

If different professional organizations have recommendations, compare:

AGE
RISK GROUP
SCREENING INTERVAL
BENEFITS
HARMS

Explain this diagnostic test:

[TEST]

What is it used for?

What does a positive result mean?

What does a negative result mean?

What are its sensitivity and specificity, if reliable data are available?

Does the result depend on:

disease stage;
testing time;
previous treatment;
sample quality?

Does the test:

confirm a diagnosis
or
only increase or decrease its probability?

Explain what is typically done if the result does not match the clinical picture.

Explain the following radiology report in plain language:

[REPORT]

Do not try to make a new diagnosis from an image or short description beyond what the report says.

For each term, explain:

what it means;
where it is located;
whether the radiologist describes a definite finding or a possibility;
whether additional follow-up is recommended.

Clearly identify:

INCIDENTAL FINDINGS
URGENT FINDINGS
FOLLOW-UP RECOMMENDATIONS

If the finding depends on comparison with an older scan, say so.

Help me prepare information for a doctor's appointment.

PROBLEM:
[...]

Let's create a brief, structured overview:

1. main symptom;
2. date it began;
3. course over time;
4. what makes it worse;
5. what relieves it;
6. associated symptoms;
7. previous episodes;
8. relevant conditions;
9. medications;
10. allergies;
11. relevant test results.

Then create a list of no more than 10 specific questions that would make sense to ask the doctor.

Prioritize questions that could change the diagnostic or treatment plan.

I want an educational analysis of this medical plan:

DIAGNOSIS:
[...]

PROPOSED TREATMENT:
[...]

FINDINGS:
[...]

Do not automatically challenge the plan.

Check:

whether it is consistent with current guidelines;
the situations in which this treatment is used;
which alternatives exist;
which contraindications are relevant;
what monitoring is usually done;
what information is missing.

Separate:

STANDARD APPROACH
REASONABLE ALTERNATIVES
PATIENT-SPECIFIC FACTORS
QUESTIONS FOR THE TREATING CLINICIAN

Explain how this condition is typically monitored:

[CHRONIC CONDITION]

Use current professional guidelines.

Separate:

SYMPTOM MONITORING
LAB MONITORING
IMAGING
MEDICATION MONITORING
COMPLICATION SCREENING
LIFESTYLE FACTORS

For each item, state:

why it is monitored;
what a change may mean;
in what context the monitoring frequency changes.

Do not give a universal schedule if guidelines vary by risk or disease stage.

Analyze this question:

[NUTRITION / DIET]

GOAL:
[...]

Relevant health conditions:
[...]

Do not rely on influencer claims.

Prefer:

guidelines;
systematic reviews;
meta-analyses;
long-term studies when relevant.

Separate:

PROVEN EFFECT
LIKELY EFFECT
WEAK EVIDENCE
INSUFFICIENT EVIDENCE

If the result depends on total energy intake, body weight, or replacing one food with another, explain that explicitly.

Do not treat an individual food as an isolated medicine or poison.

Analyze this supplement:

[NAME]

PURPOSE OF USE:
[...]

Check:

proven benefit;
studied doses;
quality of evidence;
adverse effects;
interactions;
upper intake limits, if available;
at-risk groups.

Separate:

CORRECTING A DEFICIENCY
POSSIBLE BENEFIT WITHOUT DEFICIENCY
UNPROVEN CLAIMS

Do not assume that a higher dose means greater benefit.

Check the possibility of toxicity at high doses in particular.

Help me structure these psychological or mental health symptoms:

[DESCRIPTION]

Do not make a definitive psychiatric diagnosis from a short conversation.

Analyze:

duration;
frequency;
intensity;
effect on functioning;
sleep;
mood;
anxiety;
concentration;
substances or medications;
recent life events.

Explain what types of conditions can cause similar symptoms and which details help distinguish them.

Clearly identify situations in which urgent professional help is needed.

Do not defend the previous medical answer.

Audit it as an independent reviewer.

For each important claim, check:

- whether it matches current guidelines;
- whether the source is high enough quality;
- whether the study population is comparable to the relevant population and my case;
- whether relative risk is presented without absolute risk;
- whether an observational association is presented as causation;
- whether an animal or in vitro study is directly extrapolated to humans;
- whether a minor abnormality is presented as a diagnosis;
- whether red flags were missed.

Classify each claim:

SUPPORTED
NEEDS QUALIFICATION
WEAK EVIDENCE
UNSUPPORTED
INCORRECT

Then create a corrected version.

=== MEDICAL QUALITY CONTROL ===

Before your final answer, check:

1. whether the medical information is time-sensitive;
2. whether it uses current guidelines;
3. whether it distinguishes a guideline from an individual study;
4. whether the study population resembles the relevant population;
5. whether absolute and relative effects are presented correctly;
6. whether correlation is being turned into causation;
7. whether evidence from animals is incorrectly applied to humans;
8. whether contraindications exist;
9. whether important interactions exist;
10. whether red flags are missing;
11. whether the conclusion depends on information we do not have;
12. whether the answer sounds more certain than the evidence allows.

If the evidence is uncertain:

say that it is uncertain.

Do not fill gaps with confident medical claims.

Research this scientific question:

[TOPIC]

GOAL:
[what exactly I want to establish]

PERIOD:
[...]

POPULATION / FIELD:
[...]

Before starting, define:

- key terms;
- synonyms;
- possible alternative terms;
- relevant subdisciplines.

Do not write a conclusion immediately.

First create an evidence map.

Prioritize the following:

1. systematic reviews;
2. meta-analyses;
3. randomized studies when relevant;
4. large observational studies;
5. replication studies;
6. professional guidelines and consensus statements;
7. original papers.

For each important paper, record:

TITLE
AUTHORS
YEAR
JOURNAL
STUDY TYPE
POPULATION
SAMPLE SIZE
METHOD
PRIMARY OUTCOME
EFFECT SIZE
CONFIDENCE INTERVAL
LIMITATIONS
FUNDING
CONFLICTS OF INTEREST
PEER-REVIEWED / PREPRINT

At the end, separate:

WELL ESTABLISHED
MODERATELY SUPPORTED
LIMITED EVIDENCE
CONTRADICTORY
INSUFFICIENTLY STUDIED

Do not use the number of papers found as a substitute for evidence quality.

Conduct a structured literature review on:

[TOPIC]

RESEARCH QUESTION:
[...]

PERIOD:
[...]

INCLUDE:
[...]

EXCLUDE:
[...]

First define the search strategy:

- key terms;
- Boolean combinations;
- synonyms;
- relevant disciplines.

Then group the papers by:

THEME
METHOD
POPULATION
OUTCOME
PERIOD

For each important paper, state:

CITATION
STUDY DESIGN
N
MAIN RESULT
EFFECT SIZE
UNCERTAINTY
KEY LIMITATION

Do not create a narrative that simply lists papers by year.

Try to identify:

- dominant findings;
- contradictory results;
- methodological differences;
- gaps in the literature;
- questions that are not yet well resolved.

Act as a research assistant preparing a systematic review style overview.

TOPIC:
[...]

RESEARCH QUESTION:
[...]

Define:

POPULATION
INTERVENTION / EXPOSURE
COMPARISON
OUTCOME
STUDY TYPES
TIME PERIOD

Create:

INCLUSION CRITERIA
EXCLUSION CRITERIA

For each included study, record:

study design;
sample size;
population;
intervention or exposure;
comparison;
outcome;
follow-up;
effect estimate;
confidence interval;
statistical significance;
limitations;
risk of bias.

Do not exclude a study just because its result is not statistically significant.

Record these in particular:

NEGATIVE RESULTS
NULL RESULTS
REPLICATION FAILURES

For each paper in the analysis, check its status:

PEER-REVIEWED
PREPRINT
CONFERENCE PAPER
WORKING PAPER
RETRACTED
CORRECTED
UNKNOWN

If it is a preprint:

- clearly label it;
- do not treat it as a confirmed finding;
- check whether a later peer-reviewed version exists.

If a peer-reviewed version of the same paper exists:

compare whether the following changed:

sample;
methods;
results;
effect size;
conclusions.

If a paper has been corrected or retracted, state that explicitly.

Audit these references:

[REFERENCE LIST]

For each one, check:

- whether the paper exists;
- exact title;
- authors;
- year;
- journal;
- volume/issue;
- DOI;
- PubMed/arXiv/other identifier, if available.

Do not assume a reference exists just because it looks convincing.

Status:

VERIFIED
PARTIALLY VERIFIED
NOT FOUND
INCORRECT CITATION

If the DOI does not match the paper title, flag the problem.

Check this claim:

"[CLAIM]"

Cited paper:

[REFERENCE]

Do not check only whether the paper covers the same topic.

Determine:

- whether the claim was tested directly;
- whether the result is a primary or secondary outcome;
- whether the finding is statistically significant;
- the size of the effect;
- which population was studied;
- whether the authors themselves reach the same conclusion.

Classify it:

DIRECTLY SUPPORTED
PARTIALLY SUPPORTED
OVERSTATED
NOT SUPPORTED
CONTRADICTED

Explain exactly which part of the claim is problematic.

Explain this meta-analysis:

[DATA / PAPER]

Analyze these in particular:

NUMBER OF STUDIES
TOTAL SAMPLE
EFFECT MEASURE
POOLED EFFECT
CONFIDENCE INTERVAL
P-VALUE
HETEROGENEITY
I²
PUBLICATION BIAS
SENSITIVITY ANALYSIS

Do not conclude that a meta-analysis automatically represents strong evidence.

Check:

- quality of the included studies;
- heterogeneity;
- whether the populations differ;
- whether the interventions differ;
- whether small studies dominate;
- whether publication bias exists.

If heterogeneity is high, explain how meaningful the pooled result is.

Analyze this result:

[RESULT]

Do not focus only on the p-value.

Show:

EFFECT SIZE
CONFIDENCE INTERVAL
ABSOLUTE EFFECT
RELATIVE EFFECT
SAMPLE SIZE
CLINICAL / PRACTICAL SIGNIFICANCE

If p < 0.05 but the effect is very small, say so clearly.

If the result is not statistically significant, check whether the confidence interval still allows for a potentially important effect.

Do not confuse:

"no statistically significant effect"

with:

"proven to have no effect".

Analyze whether the study has enough statistical power.

STUDY:
[...]

Check:

SAMPLE SIZE
EXPECTED EFFECT
OBSERVED EFFECT
EVENT RATE
VARIABILITY
POWER CALCULATION
DROPOUT

Answer:

Was the study large enough to detect a practically important effect?

If no statistical significance was found, could the study simply have been underpowered?

Do not use a small sample as an automatic reason to dismiss a study, but explain what it means for uncertainty.

Audit this observational study:

[STUDY]

Identify:

EXPOSURE
OUTCOME
POPULATION
FOLLOW-UP
ADJUSTMENTS

Analyze these in particular:

CONFOUNDING
SELECTION BIAS
INFORMATION BIAS
REVERSE CAUSALITY
RESIDUAL CONFOUNDING
LOSS TO FOLLOW-UP

Check which variables the model adjusted for.

Then ask:

Could the adjustment have removed a mediator?

Is an important confounder missing?

Does the conclusion use causal language even though the study design supports only an association?

Separate:

ASSOCIATION
PLAUSIBLE CAUSALITY
CAUSAL EVIDENCE

Audit this RCT:

[STUDY]

Check:

RANDOMIZATION
ALLOCATION CONCEALMENT
BLINDING
CONTROL GROUP
BASELINE BALANCE
DROP-OUT
INTENTION-TO-TREAT
PER-PROTOCOL
PRIMARY ENDPOINT
SECONDARY ENDPOINTS
MULTIPLE TESTING
TRIAL REGISTRATION

Check in particular:

whether the final paper focuses on the same primary outcome that was registered before the study began.

If the outcome changed, flag possible reporting bias.

For each result, distinguish:

PRE-SPECIFIED
SECONDARY
EXPLORATORY
POST-HOC

Research whether this finding:

[SCIENTIFIC FINDING]

has been independently replicated.

Do not count these as fully independent replications:

- the same research team;
- the same dataset population;
- a small variation of the same analysis.

Find:

ORIGINAL STUDY
DIRECT REPLICATION
CONCEPTUAL REPLICATION
FAILED REPLICATION

For each, state:

sample;
methods;
effect size;
confidence interval.

Explain how much the replications increase or decrease confidence in the original finding.

Assess the reproducibility of this study:

[STUDY]

Check whether these are available:

DATA
CODE
PROTOCOL
PRE-REGISTRATION
ANALYSIS PLAN
MATERIALS

If data or code exist:

check whether the methods description allows the key results to be reproduced.

Identify:

undocumented transformations;
unclear exclusion criteria;
manual steps;
missing software versions;
random seed;
proprietary dependencies.

Classify it:

HIGHLY REPRODUCIBLE
PARTIALLY REPRODUCIBLE
POORLY REPRODUCIBLE
UNKNOWN

Assess the possibility of publication bias for:

[TOPIC]

Look for:

- registered trials;
- unpublished trials;
- preprints;
- conference abstracts;
- negative results;
- funnel plot analyses, if available;
- trial registries.

Compare:

NUMBER OF REGISTERED STUDIES
NUMBER OF PUBLISHED STUDIES

If positive studies dominate the literature, do not automatically conclude that the effect is real.

Explain how publication bias could change the effect estimate.

Check these papers:

[LIST]

For each one, state:

FUNDING SOURCE
AUTHOR FINANCIAL DISCLOSURES
INDUSTRY INVOLVEMENT
SPONSOR ROLE
DATA OWNERSHIP
PUBLICATION CONTROL

Do not automatically dismiss an industry-funded study.

But check whether the sponsor participated in:

study design;
analysis;
writing;
the publication decision.

Separate:

CONFLICT EXISTS
POTENTIAL BIAS MECHANISM
EVIDENCE OF ACTUAL BIAS

Audit this argument:

[ARGUMENT]

Check whether the cited papers represent the full literature.

Look in particular for:

- papers with opposing results;
- systematic reviews;
- replication failures;
- larger, newer studies;
- results that are not statistically significant.

If the argument uses only one side of the literature, label it:

POTENTIAL CHERRY-PICKING

and show what is missing.

Analyze the possibility of p-hacking or a multiple-testing problem in:

[STUDY]

Check:

NUMBER OF OUTCOMES
NUMBER OF SUBGROUPS
NUMBER OF MODELS
NUMBER OF TIMEPOINTS

Was a correction for multiple comparisons applied?

Does the paper focus on a small number of significant results among a large number of tests?

Check whether the analyses were:

PRE-REGISTERED
PRE-SPECIFIED
EXPLORATORY
POST-HOC

Do not claim p-hacking is proven without evidence.

Use this wording:

POTENTIAL MULTIPLE-TESTING RISK.

Analyze this subgroup result:

[RESULT]

Check:

Was the subgroup defined in advance?

What was the subgroup sample size?

Was there a formal interaction test?

Does one subgroup have p < 0.05 and another p > 0.05?

If so, warn:

a difference between "significant" and "not significant" does not automatically mean the difference between groups is statistically significant.

Look for a direct test of the difference between subgroups.

Analyze this outcome:

[OUTCOME]

Is it a:

CLINICAL ENDPOINT
SURROGATE ENDPOINT
BIOMARKER

If it is a surrogate:

explain whether it has been validated to predict a genuinely important clinical outcome.

Do not automatically turn an improvement in a biomarker into a claim of improvement in:

survival;
quality of life;
disease progression;
another clinical outcome.

Separate:

BIOLOGICAL EFFECT
SURROGATE EFFECT
PATIENT-IMPORTANT OUTCOME

Analyze this study:

[STUDY]

Determine the level of evidence:

IN VITRO
ANIMAL
HUMAN OBSERVATIONAL
HUMAN INTERVENTIONAL

If it is an in vitro or animal study:

do not conclude that the same effect exists in humans.

Explain:

- dose;
- route of administration;
- model;
- how similar the exposure is to human exposure;
- what would be needed to translate the result to humans.

Use a final statement that matches the level of evidence.

Check this claim:

"[X works because it affects the Y pathway]"

Separate:

MECHANISTIC PLAUSIBILITY
BIOLOGICAL EFFECT
CLINICAL EFFECT

Find whether human outcome studies exist.

Do not use a known molecular mechanism as sufficient evidence of practical benefit.

Research this scientific question:

[TOPIC]

GOAL:
[what exactly I want to establish]

PERIOD:
[...]

POPULATION / FIELD:
[...]

Before starting, define:

- key terms;
- synonyms;
- possible alternative terms;
- relevant subdisciplines.

Do not write a conclusion immediately.

First create an evidence map.

Prioritize the following:

1. systematic reviews;
2. meta-analyses;
3. randomized studies when relevant;
4. large observational studies;
5. replication studies;
6. professional guidelines and consensus statements;
7. original papers.

For each important paper, record:

TITLE
AUTHORS
YEAR
JOURNAL
STUDY TYPE
POPULATION
SAMPLE SIZE
METHOD
PRIMARY OUTCOME
EFFECT SIZE
CONFIDENCE INTERVAL
LIMITATIONS
FUNDING
CONFLICTS OF INTEREST
PEER-REVIEWED / PREPRINT

At the end, separate:

WELL ESTABLISHED
MODERATELY SUPPORTED
LIMITED EVIDENCE
CONTRADICTORY
INSUFFICIENTLY STUDIED

Do not use the number of papers found as a substitute for evidence quality.

Conduct a structured literature review on:

[TOPIC]

RESEARCH QUESTION:
[...]

PERIOD:
[...]

INCLUDE:
[...]

EXCLUDE:
[...]

First define the search strategy:

- key terms;
- Boolean combinations;
- synonyms;
- relevant disciplines.

Then group the papers by:

THEME
METHOD
POPULATION
OUTCOME
PERIOD

For each important paper, state:

CITATION
STUDY DESIGN
N
MAIN RESULT
EFFECT SIZE
UNCERTAINTY
KEY LIMITATION

Do not create a narrative that simply lists papers by year.

Try to identify:

- dominant findings;
- contradictory results;
- methodological differences;
- gaps in the literature;
- questions that are not yet well resolved.

Act as a research assistant preparing a systematic review style overview.

TOPIC:
[...]

RESEARCH QUESTION:
[...]

Define:

POPULATION
INTERVENTION / EXPOSURE
COMPARISON
OUTCOME
STUDY TYPES
TIME PERIOD

Create:

INCLUSION CRITERIA
EXCLUSION CRITERIA

For each included study, record:

study design;
sample size;
population;
intervention or exposure;
comparison;
outcome;
follow-up;
effect estimate;
confidence interval;
statistical significance;
limitations;
risk of bias.

Do not exclude a study just because its result is not statistically significant.

Record these in particular:

NEGATIVE RESULTS
NULL RESULTS
REPLICATION FAILURES

For each paper in the analysis, check its status:

PEER-REVIEWED
PREPRINT
CONFERENCE PAPER
WORKING PAPER
RETRACTED
CORRECTED
UNKNOWN

If it is a preprint:

- clearly label it;
- do not treat it as a confirmed finding;
- check whether a later peer-reviewed version exists.

If a peer-reviewed version of the same paper exists:

compare whether the following changed:

sample;
methods;
results;
effect size;
conclusions.

If a paper has been corrected or retracted, state that explicitly.

Audit these references:

[REFERENCE LIST]

For each one, check:

- whether the paper exists;
- exact title;
- authors;
- year;
- journal;
- volume/issue;
- DOI;
- PubMed/arXiv/other identifier, if available.

Do not assume a reference exists just because it looks convincing.

Status:

VERIFIED
PARTIALLY VERIFIED
NOT FOUND
INCORRECT CITATION

If the DOI does not match the paper title, flag the problem.

Check this claim:

"[CLAIM]"

Cited paper:

[REFERENCE]

Do not check only whether the paper covers the same topic.

Determine:

- whether the claim was tested directly;
- whether the result is a primary or secondary outcome;
- whether the finding is statistically significant;
- the size of the effect;
- which population was studied;
- whether the authors themselves reach the same conclusion.

Classify it:

DIRECTLY SUPPORTED
PARTIALLY SUPPORTED
OVERSTATED
NOT SUPPORTED
CONTRADICTED

Explain exactly which part of the claim is problematic.

Explain this meta-analysis:

[DATA / PAPER]

Analyze these in particular:

NUMBER OF STUDIES
TOTAL SAMPLE
EFFECT MEASURE
POOLED EFFECT
CONFIDENCE INTERVAL
P-VALUE
HETEROGENEITY
I²
PUBLICATION BIAS
SENSITIVITY ANALYSIS

Do not conclude that a meta-analysis automatically represents strong evidence.

Check:

- quality of the included studies;
- heterogeneity;
- whether the populations differ;
- whether the interventions differ;
- whether small studies dominate;
- whether publication bias exists.

If heterogeneity is high, explain how meaningful the pooled result is.

Analyze this result:

[RESULT]

Do not focus only on the p-value.

Show:

EFFECT SIZE
CONFIDENCE INTERVAL
ABSOLUTE EFFECT
RELATIVE EFFECT
SAMPLE SIZE
CLINICAL / PRACTICAL SIGNIFICANCE

If p < 0.05 but the effect is very small, say so clearly.

If the result is not statistically significant, check whether the confidence interval still allows for a potentially important effect.

Do not confuse:

"no statistically significant effect"

with:

"proven to have no effect".

Analyze whether the study has enough statistical power.

STUDY:
[...]

Check:

SAMPLE SIZE
EXPECTED EFFECT
OBSERVED EFFECT
EVENT RATE
VARIABILITY
POWER CALCULATION
DROPOUT

Answer:

Was the study large enough to detect a practically important effect?

If no statistical significance was found, could the study simply have been underpowered?

Do not use a small sample as an automatic reason to dismiss a study, but explain what it means for uncertainty.

Audit this observational study:

[STUDY]

Identify:

EXPOSURE
OUTCOME
POPULATION
FOLLOW-UP
ADJUSTMENTS

Analyze these in particular:

CONFOUNDING
SELECTION BIAS
INFORMATION BIAS
REVERSE CAUSALITY
RESIDUAL CONFOUNDING
LOSS TO FOLLOW-UP

Check which variables the model adjusted for.

Then ask:

Could the adjustment have removed a mediator?

Is an important confounder missing?

Does the conclusion use causal language even though the study design supports only an association?

Separate:

ASSOCIATION
PLAUSIBLE CAUSALITY
CAUSAL EVIDENCE

Audit this RCT:

[STUDY]

Check:

RANDOMIZATION
ALLOCATION CONCEALMENT
BLINDING
CONTROL GROUP
BASELINE BALANCE
DROP-OUT
INTENTION-TO-TREAT
PER-PROTOCOL
PRIMARY ENDPOINT
SECONDARY ENDPOINTS
MULTIPLE TESTING
TRIAL REGISTRATION

Check in particular:

whether the final paper focuses on the same primary outcome that was registered before the study began.

If the outcome changed, flag possible reporting bias.

For each result, distinguish:

PRE-SPECIFIED
SECONDARY
EXPLORATORY
POST-HOC

Research whether this finding:

[SCIENTIFIC FINDING]

has been independently replicated.

Do not count these as fully independent replications:

- the same research team;
- the same dataset population;
- a small variation of the same analysis.

Find:

ORIGINAL STUDY
DIRECT REPLICATION
CONCEPTUAL REPLICATION
FAILED REPLICATION

For each, state:

sample;
methods;
effect size;
confidence interval.

Explain how much the replications increase or decrease confidence in the original finding.

Assess the reproducibility of this study:

[STUDY]

Check whether these are available:

DATA
CODE
PROTOCOL
PRE-REGISTRATION
ANALYSIS PLAN
MATERIALS

If data or code exist:

check whether the methods description allows the key results to be reproduced.

Identify:

undocumented transformations;
unclear exclusion criteria;
manual steps;
missing software versions;
random seed;
proprietary dependencies.

Classify it:

HIGHLY REPRODUCIBLE
PARTIALLY REPRODUCIBLE
POORLY REPRODUCIBLE
UNKNOWN

Assess the possibility of publication bias for:

[TOPIC]

Look for:

- registered trials;
- unpublished trials;
- preprints;
- conference abstracts;
- negative results;
- funnel plot analyses, if available;
- trial registries.

Compare:

NUMBER OF REGISTERED STUDIES
NUMBER OF PUBLISHED STUDIES

If positive studies dominate the literature, do not automatically conclude that the effect is real.

Explain how publication bias could change the effect estimate.

Check these papers:

[LIST]

For each one, state:

FUNDING SOURCE
AUTHOR FINANCIAL DISCLOSURES
INDUSTRY INVOLVEMENT
SPONSOR ROLE
DATA OWNERSHIP
PUBLICATION CONTROL

Do not automatically dismiss an industry-funded study.

But check whether the sponsor participated in:

study design;
analysis;
writing;
the publication decision.

Separate:

CONFLICT EXISTS
POTENTIAL BIAS MECHANISM
EVIDENCE OF ACTUAL BIAS

Audit this argument:

[ARGUMENT]

Check whether the cited papers represent the full literature.

Look in particular for:

- papers with opposing results;
- systematic reviews;
- replication failures;
- larger, newer studies;
- results that are not statistically significant.

If the argument uses only one side of the literature, label it:

POTENTIAL CHERRY-PICKING

and show what is missing.

Analyze the possibility of p-hacking or a multiple-testing problem in:

[STUDY]

Check:

NUMBER OF OUTCOMES
NUMBER OF SUBGROUPS
NUMBER OF MODELS
NUMBER OF TIMEPOINTS

Was a correction for multiple comparisons applied?

Does the paper focus on a small number of significant results among a large number of tests?

Check whether the analyses were:

PRE-REGISTERED
PRE-SPECIFIED
EXPLORATORY
POST-HOC

Do not claim p-hacking is proven without evidence.

Use this wording:

POTENTIAL MULTIPLE-TESTING RISK.

Analyze this subgroup result:

[RESULT]

Check:

Was the subgroup defined in advance?

What was the subgroup sample size?

Was there a formal interaction test?

Does one subgroup have p < 0.05 and another p > 0.05?

If so, warn:

a difference between "significant" and "not significant" does not automatically mean the difference between groups is statistically significant.

Look for a direct test of the difference between subgroups.

Analyze this outcome:

[OUTCOME]

Is it a:

CLINICAL ENDPOINT
SURROGATE ENDPOINT
BIOMARKER

If it is a surrogate:

explain whether it has been validated to predict a genuinely important clinical outcome.

Do not automatically turn an improvement in a biomarker into a claim of improvement in:

survival;
quality of life;
disease progression;
another clinical outcome.

Separate:

BIOLOGICAL EFFECT
SURROGATE EFFECT
PATIENT-IMPORTANT OUTCOME

Analyze this study:

[STUDY]

Determine the level of evidence:

IN VITRO
ANIMAL
HUMAN OBSERVATIONAL
HUMAN INTERVENTIONAL

If it is an in vitro or animal study:

do not conclude that the same effect exists in humans.

Explain:

- dose;
- route of administration;
- model;
- how similar the exposure is to human exposure;
- what would be needed to translate the result to humans.

Use a final statement that matches the level of evidence.

Check this claim:

"[X works because it affects the Y pathway]"

Separate:

MECHANISTIC PLAUSIBILITY
BIOLOGICAL EFFECT
CLINICAL EFFECT

Find whether human outcome studies exist.

Do not use a known molecular mechanism as sufficient evidence of practical benefit.

Assess the current state of scientific consensus on:

[TOPIC]

Do not infer consensus from a few popular papers.

Look for:

- systematic reviews;
- meta-analyses;
- professional society statements;
- consensus statements;
- major guidelines;
- large, high-quality studies;
- relevant replications.

Then classify it:

STRONG CONSENSUS
BROAD BUT NOT COMPLETE CONSENSUS
ACTIVE SCIENTIFIC DEBATE
INSUFFICIENT EVIDENCE
FRINGE CLAIM

Explain the evidence behind the classification.

Do not confuse:

the public popularity of an idea

with:

scientific consensus.

Analyze this scientific controversy:

[TOPIC]

First identify the actual scientific positions.

Do not create false balance.

For each position, state:

WHAT IT CLAIMS
BEST SUPPORTING EVIDENCE
BEST COUNTER-EVIDENCE
QUALITY OF EVIDENCE
UNRESOLVED QUESTIONS

If one position has significantly stronger evidence, show that imbalance.

Do not present two positions as equally supported just because both exist.

Check this statistical claim:

"[CLAIM]"

From this study:

[STUDY]

Find the relevant table or figure.

Check:

sample size;
effect estimate;
standard error;
confidence interval;
p-value;
adjustments.

Recalculate the result where possible.

Check whether:

the percentage matches the raw counts;
the confidence interval matches the effect estimate;
the text matches the table.

If the abstract, main text, and table disagree, show the discrepancy.

Analyze this chart from a scientific paper:

[CHART]

Do not rely only on visual impression.

Determine:

X AXIS
Y AXIS
SCALE
LOG SCALE?
BASELINE
ERROR BARS
SAMPLE SIZE
NORMALIZATION

Check these in particular:

- truncated axis;
- different scales;
- cumulative versus instantaneous values;
- relative versus absolute change;
- smoothing;
- cherry-picked period.

Explain what the chart actually shows and what we cannot conclude from it.

Assess the overall quality of evidence for:

[QUESTION]

Use GRADE-style reasoning.

Consider:

RISK OF BIAS
INCONSISTENCY
INDIRECTNESS
IMPRECISION
PUBLICATION BIAS

Where relevant, also consider:

LARGE EFFECT
DOSE RESPONSE
RESIDUAL CONFOUNDING

Finally, classify confidence in the evidence as:

HIGH
MODERATE
LOW
VERY LOW

Explain the reason for each downgrade or upgrade.

Do not classify evidence based only on study type.

Based on the literature about:

[TOPIC]

identify genuine research gaps.

Do not generate generic phrases such as:

"more research is needed."

For each gap, explain:

WHAT IS UNKNOWN
WHY CURRENT STUDIES DO NOT ANSWER IT
WHAT STUDY DESIGN WOULD HELP
WHAT POPULATION
WHAT OUTCOME
WHAT SAMPLE / FOLLOW-UP WOULD BE RELEVANT

Separate:

MISSING DATA
CONFLICTING DATA
METHODOLOGICAL LIMITATION
UNDERSTUDIED POPULATION
LACK OF LONG-TERM DATA

Design a study for this question:

[QUESTION]

First define:

HYPOTHESIS
POPULATION
INTERVENTION / EXPOSURE
CONTROL
PRIMARY OUTCOME
SECONDARY OUTCOMES

Then define:

STUDY DESIGN
INCLUSION CRITERIA
EXCLUSION CRITERIA
RANDOMIZATION
BLINDING
SAMPLE SIZE LOGIC
FOLLOW-UP
ANALYSIS PLAN

Prevent these in particular:

selection bias;
measurement bias;
confounding;
multiple testing;
outcome switching.

Recommend pre-registration and an analysis plan defined in advance.

Audit the paper's methodology:

[PAPER]

Assess it by its method, not by its conclusion.

Analyze:

RESEARCH QUESTION
STUDY DESIGN
SAMPLING
MEASUREMENT
CONTROL GROUP
STATISTICS
MISSING DATA
CONFOUNDERS
BIAS
GENERALIZABILITY

For each problem, state:

SEVERITY
WHY IT MATTERS
DIRECTION OF POSSIBLE BIAS
CAN IT CHANGE THE CONCLUSION?

Separate:

FATAL FLAW
MAJOR LIMITATION
MINOR LIMITATION

Compare the abstract and the full paper:

[STUDY]

Check whether the abstract:

- overstates the results;
- omits negative results;
- uses stronger causal language;
- leaves out limitations;
- selects a secondary outcome.

Create a table:

CLAIM IN ABSTRACT
WHAT THE FULL PAPER SHOWS
STATUS

Status:

ACCURATE
OVERSIMPLIFIED
OVERSTATED
MISLEADING

Check this news article, which claims:

[CLAIM]

Find the original study.

Compare:

MEDIA CLAIM
STUDY QUESTION
ACTUAL RESULT
POPULATION
EFFECT SIZE
LIMITATIONS

Check in particular whether the headline turned:

association into causation
a mouse study into a claim about humans
relative risk into a dramatic absolute claim
a preprint into established fact

Finally, state:

WHAT THE STUDY ACTUALLY SHOWS
WHAT IT DOES NOT SHOW

Check all references in this text:

[TEXT]

For each reference:

1. confirm that it exists;
2. find its DOI or identifier;
3. check whether the authors match;
4. check the year;
5. check the title;
6. check whether it supports the claim.

If the reference does not exist, label it:

HALLUCINATED / NOT FOUND

If it exists but does not support the claim, label it:

MIS-CITED

Do not guess at corrections to a reference.

Use Deep Research for:

[TOPIC]

GOAL:

I want a reliable synthesis of the current state of scientific evidence.

Create a research plan first.

Define:

- research question;
- key terms;
- inclusion criteria;
- exclusion criteria;
- date range;
- study hierarchy.

Look in particular for:

systematic reviews;
meta-analyses;
RCTs;
large observational studies;
replication studies;
guidelines;
negative findings.

Check multiple sources for each key conclusion where possible.

Analyze these in particular:

risk of bias;
confounding;
heterogeneity;
publication bias;
effect size;
external validity.

Do not force a clear conclusion if the literature remains contradictory.

At the end, create:

EXECUTIVE SUMMARY
EVIDENCE MAP
STRONGEST EVIDENCE
CONFLICTING EVIDENCE
QUALITY ASSESSMENT
RESEARCH GAPS
FINAL SYNTHESIS

=== SCIENTIFIC QUALITY CONTROL ===

Before your final answer, check:

1. whether all cited papers actually exist;
2. whether the peer-reviewed or preprint status is correct;
3. whether the citation really supports the claim;
4. whether the population matches the question;
5. whether the effect size is interpreted correctly;
6. whether the confidence interval shows substantial uncertainty;
7. whether the p-value is being used as the only measure;
8. whether correlation is being turned into causation;
9. whether an animal or in vitro result is being turned into a claim about humans;
10. whether null results are being ignored;
11. whether publication bias exists;
12. whether replication failures exist;
13. whether funding or conflicts of interest may be relevant;
14. whether the result is generalized beyond what the study allows;
15. whether the conclusions are stronger than the evidence.

If the evidence is not strong enough:

weaken the wording.

Do not increase certainty just to make the answer sound more decisive.

Analyze the attached dataset.

Do not draw conclusions immediately.

PHASE 1 - DATA AUDIT

First determine:

- number of rows;
- number of columns;
- column names;
- data types;
- percentage of missing values;
- number of unique values;
- duplicates;
- minimum and maximum values;
- possible ID columns;
- date/time columns;
- potential units.

Then identify:

- impossible values;
- outliers;
- inconsistent categories;
- date format;
- unit changes;
- methodology changes;
- obvious parsing problems.

Do not automatically fix the data.

Show me the problems first.

PHASE 2 - DATA DICTIONARY

For each column, create:

COLUMN
TYPE
LIKELY MEANING
UNIT
MISSING RATE
UNIQUE COUNT
POTENTIAL ISSUES

If the meaning of a column is uncertain, label it:

UNKNOWN

Do not invent a meaning.

PHASE 3 - ANALYSIS

Only after that, analyze this goal:

[GOAL]

Perform a strict data quality audit.

DATASET:
[ATTACHED]

Check:

COMPLETENESS
ACCURACY
CONSISTENCY
UNIQUENESS
VALIDITY
TIMELINESS

For each issue, return:

COLUMN
ROW/PATTERN
PROBLEM
SEVERITY
POSSIBLE CAUSE
SAFE ACTION

Severity:

CRITICAL
HIGH
MEDIUM
LOW

Look in particular for:

- duplicated IDs;
- conflicting records;
- impossible dates;
- negative values where they are not possible;
- percentages outside the expected range;
- currency mismatch;
- timezone mismatch;
- mixed decimal separators;
- leading or trailing spaces;
- inconsistent capitalization;
- misspelled categories.

Do not automatically delete an outlier just because it is extreme.

Create a professional data dictionary from the dataset.

For each column, provide:

NAME
DATA TYPE
BUSINESS MEANING
UNIT
EXAMPLE
VALID RANGE
NULL MEANING
UNIQUE / REPEATED
KEY STATUS
RELATIONSHIP
POTENTIAL DATA QUALITY ISSUE

If the meaning is not explicitly known, label it:

INFERRED

and clearly mark it as an assumption.

Do not present an inferred meaning as confirmed fact.

We need to calculate this KPI:

[KPI]

Formally define it before calculating.

State:

BUSINESS DEFINITION
NUMERATOR
DENOMINATOR
FILTERS
TIME WINDOW
TIMEZONE
INCLUSION RULES
EXCLUSION RULES
AGGREGATION
MISSING-DATA POLICY

Then check whether the dataset contains all required columns.

If there are multiple possible definitions of the KPI, show them before calculating.

Do not choose one definition without a warning.

Analyze this funnel:

STEP 1:
[...]

STEP 2:
[...]

STEP 3:
[...]

STEP 4:
[...]

First define the exact event representing each step.

For each step, calculate:

USERS ENTERED
USERS COMPLETED
STEP CONVERSION
DROP-OFF
CUMULATIVE CONVERSION

Check these in particular:

- duplicate events;
- multiple sessions for the same user;
- event ordering;
- missing user IDs;
- cross-device identity;
- timezone;
- attribution window.

Segment the results, when there is enough data, by:

channel;
device;
country;
plan;
cohort.

Do not conclude that the segment with the highest conversion percentage necessarily produces the greatest total value.

Perform a cohort analysis.

COHORT DEFINITION:
[for example, month of registration]

SUCCESS METRIC:
[retention / revenue / activity]

INTERVAL:
[day/week/month]

First check that cohort membership does not change over time.

Create a table:

COHORT
SIZE
PERIOD 0
PERIOD 1
PERIOD 2
...

Show both:

ABSOLUTE COUNTS
PERCENTAGES

Identify these in particular:

- small cohorts;
- structural breaks;
- product changes;
- changes in acquisition-channel mix.

Do not compare a cohort of 50 users with a cohort of 50,000 without warning about uncertainty.

Analyze retention.

First define what counts as an:

ACTIVE USER

and which type of retention we are analyzing:

N-DAY
UNBOUNDED
ROLLING
CLASSIC

Do not calculate until the definition is clear.

Show these results:

D1
D7
D14
D30
D60
D90

when the data allow it.

Segment by:

acquisition source;
platform;
country;
plan;
product version.

Check in particular whether a retention change is actually caused by a change in user mix.

Analyze revenue.

Before analysis, determine:

GROSS / NET
BOOKED / RECOGNIZED
REFUNDS
TAX
DISCOUNTS
CURRENCY

Find these metrics if the dataset allows:

TOTAL REVENUE
REVENUE GROWTH
ARPU
ARPPU
MRR
ARR
CHURNED MRR
EXPANSION MRR
CONTRACTION MRR

For multiple currencies:

do not add different currencies directly.

Define the FX methodology first.

Check these in particular:

revenue concentration;
top customers;
seasonality;
one-off transactions.

Analyze churn.

Define churn first.

Distinguish:

USER CHURN
LOGO CHURN
REVENUE CHURN
GROSS REVENUE CHURN
NET REVENUE RETENTION

State the formula for each metric.

Find:

- when churn increases;
- which segments contribute most;
- how much revenue churned users represent;
- whether churn differs by tenure;
- whether new users have different churn.

Do not infer the cause of churn from correlation alone.

Analyze this A/B test.

HYPOTHESIS:
[...]

CONTROL:
[...]

VARIANT:
[...]

PRIMARY METRIC:
[...]

SECONDARY METRICS:
[...]

Before looking at the results, check:

- randomization;
- sample size;
- sample ratio mismatch;
- test duration;
- peeking;
- seasonality;
- multiple testing;
- novelty effect.

Show:

CONTROL N
VARIANT N
CONTROL RATE
VARIANT RATE
ABSOLUTE DIFFERENCE
RELATIVE DIFFERENCE
CONFIDENCE INTERVAL
P-VALUE

where relevant.

But do not stop at the p-value.

Assess:

PRACTICAL SIGNIFICANCE
STATISTICAL SIGNIFICANCE
UNCERTAINTY

Check guardrail metrics in particular.

Before reaching a final conclusion, check for Simpson's paradox.

If the overall metric shows:

A > B

check whether the following can hold within segments:

A < B

Segments:

region
channel
device
plan
age group
product category
period

Create:

AGGREGATED RESULT
SEGMENTED RESULT

If the conclusions differ, explain why.

Identify outliers.

Do not remove them automatically.

For each outlier, check whether it is:

DATA ERROR
VALID EXTREME VALUE
FRAUD
SPECIAL EVENT
SYSTEM BUG
UNKNOWN

Use multiple approaches when appropriate:

IQR
Z-SCORE
ROBUST Z-SCORE
PERCENTILES
DOMAIN RULES

For each proposed removal, explain:

WHY
IMPACT ON RESULT
RESULT WITH OUTLIERS
RESULT WITHOUT OUTLIERS

Do not hide how much the conclusion changes after removal.

Analyze missing values.

For each column, show:

MISSING COUNT
MISSING %
PATTERN

Try to assess whether the data are:

MCAR
MAR
MNAR

if there is enough information.

Do not impute automatically.

First compare rows with and without missing values.

If you propose imputation, state:

METHOD
ASSUMPTIONS
RISK
IMPACT

Show how the key conclusions change:

WITHOUT IMPUTATION
WITH IMPUTATION

Find duplicates.

Do not use only a full-row duplicate check.

Identify possible business keys:

[user_id + date]
[email]
[transaction_id]
[order_id]
[...]

Distinguish:

EXACT DUPLICATE
LIKELY DUPLICATE
LEGITIMATE REPEATED EVENT

Propose a safe deduplication rule for each type.

Do not delete repeated transactions just because they have the same amount.

Analyze this time series:

[METRIC]

PERIOD:
[...]

FREQUENCY:
[...]

First check:

missing periods;
duplicate timestamps;
timezone;
unequal intervals;
structural breaks.

Then analyze:

TREND
SEASONALITY
CYCLICAL PATTERN
OUTLIERS
CHANGE POINTS

Do not forecast until you have established whether the historical series is stable enough.

If you create a forecast:

show the prediction interval.

Do not provide just one number.

Create a forecast for:

[METRIC]

HORIZON:
[...]

First create a baseline:

NAIVE
SEASONAL NAIVE

Only then compare more complex models.

If you use:

moving average;
exponential smoothing;
ARIMA;
regression;
another approach

explain why.

Perform backtesting.

For each model, show:

MAE
RMSE
MAPE, if appropriate.

Do not use MAPE when the denominator could be zero.

Finally, show:

POINT FORECAST
LOWER BOUND
UPPER BOUND

and the main assumptions.

Analyze correlations among the numeric columns.

Do not just create a correlation matrix.

For the strongest correlations, check:

- scatter plot;
- nonlinear relationship;
- dependence on outliers;
- common time trend;
- confounders.

Distinguish:

PEARSON
SPEARMAN

when relevant.

Do not describe correlation as causation.

Check for spurious correlations in time-series data in particular.

Create a regression analysis for:

TARGET:
[...]

PREDICTORS:
[...]

Before modeling, check:

linearity;
multicollinearity;
heteroskedasticity;
residual distribution;
outliers;
missingness.

Show:

COEFFICIENT
STANDARD ERROR
CONFIDENCE INTERVAL
P-VALUE

but also explain the practical meaning of the coefficient.

Check VIF or another suitable multicollinearity indicator.

Do not interpret a coefficient as a causal effect without an appropriate identification design.

Audit this classification model.

TARGET:
[...]

CLASS DISTRIBUTION:
[...]

MODEL:
[...]

Do not use accuracy as the only metric.

Show:

CONFUSION MATRIX
PRECISION
RECALL
SPECIFICITY
F1
ROC-AUC
PR-AUC

especially for an imbalanced dataset.

Check:

class imbalance;
data leakage;
train/test contamination;
temporal leakage;
duplicate entities;
threshold choice.

If the model is probabilistic, check calibration.

Do not use a random split when real-world use requires predicting the future and a time-based split is more appropriate.

Try to find data leakage.

For each feature, ask:

Will this information be available when the model needs to make a prediction?

Look for:

future information;
post-outcome fields;
target-derived columns;
aggregates calculated using future data;
user-level leakage between train and test;
duplicates;
preprocessing before the split.

For each issue, state:

FEATURE
LEAKAGE PATH
WHY INVALID
SAFE FIX

Write SQL for:

[GOAL]

SCHEMA:
[...]

Before the query, explain:

TABLE GRAIN
JOIN KEYS
FILTERS
TIMEZONE
METRIC DEFINITION

Check many-to-many joins in particular.

After the query, add validation queries for:

- row count;
- duplicate keys;
- NULL rate;
- reconciliation total.

Do not use SELECT DISTINCT to hide a bad join.

Analyze the attached dataset.

Do not draw conclusions immediately.

PHASE 1 - DATA AUDIT

First determine:

- number of rows;
- number of columns;
- column names;
- data types;
- percentage of missing values;
- number of unique values;
- duplicates;
- minimum and maximum values;
- possible ID columns;
- date/time columns;
- potential units.

Then identify:

- impossible values;
- outliers;
- inconsistent categories;
- date format;
- unit changes;
- methodology changes;
- obvious parsing problems.

Do not automatically fix the data.

Show me the problems first.

PHASE 2 - DATA DICTIONARY

For each column, create:

COLUMN
TYPE
LIKELY MEANING
UNIT
MISSING RATE
UNIQUE COUNT
POTENTIAL ISSUES

If the meaning of a column is uncertain, label it:

UNKNOWN

Do not invent a meaning.

PHASE 3 - ANALYSIS

Only after that, analyze this goal:

[GOAL]

Perform a strict data quality audit.

DATASET:
[ATTACHED]

Check:

COMPLETENESS
ACCURACY
CONSISTENCY
UNIQUENESS
VALIDITY
TIMELINESS

For each issue, return:

COLUMN
ROW/PATTERN
PROBLEM
SEVERITY
POSSIBLE CAUSE
SAFE ACTION

Severity:

CRITICAL
HIGH
MEDIUM
LOW

Look in particular for:

- duplicated IDs;
- conflicting records;
- impossible dates;
- negative values where they are not possible;
- percentages outside the expected range;
- currency mismatch;
- timezone mismatch;
- mixed decimal separators;
- leading or trailing spaces;
- inconsistent capitalization;
- misspelled categories.

Do not automatically delete an outlier just because it is extreme.

Create a professional data dictionary from the dataset.

For each column, provide:

NAME
DATA TYPE
BUSINESS MEANING
UNIT
EXAMPLE
VALID RANGE
NULL MEANING
UNIQUE / REPEATED
KEY STATUS
RELATIONSHIP
POTENTIAL DATA QUALITY ISSUE

If the meaning is not explicitly known, label it:

INFERRED

and clearly mark it as an assumption.

Do not present an inferred meaning as confirmed fact.

We need to calculate this KPI:

[KPI]

Formally define it before calculating.

State:

BUSINESS DEFINITION
NUMERATOR
DENOMINATOR
FILTERS
TIME WINDOW
TIMEZONE
INCLUSION RULES
EXCLUSION RULES
AGGREGATION
MISSING-DATA POLICY

Then check whether the dataset contains all required columns.

If there are multiple possible definitions of the KPI, show them before calculating.

Do not choose one definition without a warning.

Analyze this funnel:

STEP 1:
[...]

STEP 2:
[...]

STEP 3:
[...]

STEP 4:
[...]

First define the exact event representing each step.

For each step, calculate:

USERS ENTERED
USERS COMPLETED
STEP CONVERSION
DROP-OFF
CUMULATIVE CONVERSION

Check these in particular:

- duplicate events;
- multiple sessions for the same user;
- event ordering;
- missing user IDs;
- cross-device identity;
- timezone;
- attribution window.

Segment the results, when there is enough data, by:

channel;
device;
country;
plan;
cohort.

Do not conclude that the segment with the highest conversion percentage necessarily produces the greatest total value.

Perform a cohort analysis.

COHORT DEFINITION:
[for example, month of registration]

SUCCESS METRIC:
[retention / revenue / activity]

INTERVAL:
[day/week/month]

First check that cohort membership does not change over time.

Create a table:

COHORT
SIZE
PERIOD 0
PERIOD 1
PERIOD 2
...

Show both:

ABSOLUTE COUNTS
PERCENTAGES

Identify these in particular:

- small cohorts;
- structural breaks;
- product changes;
- changes in acquisition-channel mix.

Do not compare a cohort of 50 users with a cohort of 50,000 without warning about uncertainty.

Analyze retention.

First define what counts as an:

ACTIVE USER

and which type of retention we are analyzing:

N-DAY
UNBOUNDED
ROLLING
CLASSIC

Do not calculate until the definition is clear.

Show these results:

D1
D7
D14
D30
D60
D90

when the data allow it.

Segment by:

acquisition source;
platform;
country;
plan;
product version.

Check in particular whether a retention change is actually caused by a change in user mix.

Analyze revenue.

Before analysis, determine:

GROSS / NET
BOOKED / RECOGNIZED
REFUNDS
TAX
DISCOUNTS
CURRENCY

Find these metrics if the dataset allows:

TOTAL REVENUE
REVENUE GROWTH
ARPU
ARPPU
MRR
ARR
CHURNED MRR
EXPANSION MRR
CONTRACTION MRR

For multiple currencies:

do not add different currencies directly.

Define the FX methodology first.

Check these in particular:

revenue concentration;
top customers;
seasonality;
one-off transactions.

Analyze churn.

Define churn first.

Distinguish:

USER CHURN
LOGO CHURN
REVENUE CHURN
GROSS REVENUE CHURN
NET REVENUE RETENTION

State the formula for each metric.

Find:

- when churn increases;
- which segments contribute most;
- how much revenue churned users represent;
- whether churn differs by tenure;
- whether new users have different churn.

Do not infer the cause of churn from correlation alone.

Analyze this A/B test.

HYPOTHESIS:
[...]

CONTROL:
[...]

VARIANT:
[...]

PRIMARY METRIC:
[...]

SECONDARY METRICS:
[...]

Before looking at the results, check:

- randomization;
- sample size;
- sample ratio mismatch;
- test duration;
- peeking;
- seasonality;
- multiple testing;
- novelty effect.

Show:

CONTROL N
VARIANT N
CONTROL RATE
VARIANT RATE
ABSOLUTE DIFFERENCE
RELATIVE DIFFERENCE
CONFIDENCE INTERVAL
P-VALUE

where relevant.

But do not stop at the p-value.

Assess:

PRACTICAL SIGNIFICANCE
STATISTICAL SIGNIFICANCE
UNCERTAINTY

Check guardrail metrics in particular.

Before reaching a final conclusion, check for Simpson's paradox.

If the overall metric shows:

A > B

check whether the following can hold within segments:

A < B

Segments:

region
channel
device
plan
age group
product category
period

Create:

AGGREGATED RESULT
SEGMENTED RESULT

If the conclusions differ, explain why.

Identify outliers.

Do not remove them automatically.

For each outlier, check whether it is:

DATA ERROR
VALID EXTREME VALUE
FRAUD
SPECIAL EVENT
SYSTEM BUG
UNKNOWN

Use multiple approaches when appropriate:

IQR
Z-SCORE
ROBUST Z-SCORE
PERCENTILES
DOMAIN RULES

For each proposed removal, explain:

WHY
IMPACT ON RESULT
RESULT WITH OUTLIERS
RESULT WITHOUT OUTLIERS

Do not hide how much the conclusion changes after removal.

Analyze missing values.

For each column, show:

MISSING COUNT
MISSING %
PATTERN

Try to assess whether the data are:

MCAR
MAR
MNAR

if there is enough information.

Do not impute automatically.

First compare rows with and without missing values.

If you propose imputation, state:

METHOD
ASSUMPTIONS
RISK
IMPACT

Show how the key conclusions change:

WITHOUT IMPUTATION
WITH IMPUTATION

Find duplicates.

Do not use only a full-row duplicate check.

Identify possible business keys:

[user_id + date]
[email]
[transaction_id]
[order_id]
[...]

Distinguish:

EXACT DUPLICATE
LIKELY DUPLICATE
LEGITIMATE REPEATED EVENT

Propose a safe deduplication rule for each type.

Do not delete repeated transactions just because they have the same amount.

Analyze this time series:

[METRIC]

PERIOD:
[...]

FREQUENCY:
[...]

First check:

missing periods;
duplicate timestamps;
timezone;
unequal intervals;
structural breaks.

Then analyze:

TREND
SEASONALITY
CYCLICAL PATTERN
OUTLIERS
CHANGE POINTS

Do not forecast until you have established whether the historical series is stable enough.

If you create a forecast:

show the prediction interval.

Do not provide just one number.

Create a forecast for:

[METRIC]

HORIZON:
[...]

First create a baseline:

NAIVE
SEASONAL NAIVE

Only then compare more complex models.

If you use:

moving average;
exponential smoothing;
ARIMA;
regression;
another approach

explain why.

Perform backtesting.

For each model, show:

MAE
RMSE
MAPE, if appropriate.

Do not use MAPE when the denominator could be zero.

Finally, show:

POINT FORECAST
LOWER BOUND
UPPER BOUND

and the main assumptions.

Analyze correlations among the numeric columns.

Do not just create a correlation matrix.

For the strongest correlations, check:

- scatter plot;
- nonlinear relationship;
- dependence on outliers;
- common time trend;
- confounders.

Distinguish:

PEARSON
SPEARMAN

when relevant.

Do not describe correlation as causation.

Check for spurious correlations in time-series data in particular.

Create a regression analysis for:

TARGET:
[...]

PREDICTORS:
[...]

Before modeling, check:

linearity;
multicollinearity;
heteroskedasticity;
residual distribution;
outliers;
missingness.

Show:

COEFFICIENT
STANDARD ERROR
CONFIDENCE INTERVAL
P-VALUE

but also explain the practical meaning of the coefficient.

Check VIF or another suitable multicollinearity indicator.

Do not interpret a coefficient as a causal effect without an appropriate identification design.

Audit this classification model.

TARGET:
[...]

CLASS DISTRIBUTION:
[...]

MODEL:
[...]

Do not use accuracy as the only metric.

Show:

CONFUSION MATRIX
PRECISION
RECALL
SPECIFICITY
F1
ROC-AUC
PR-AUC

especially for an imbalanced dataset.

Check:

class imbalance;
data leakage;
train/test contamination;
temporal leakage;
duplicate entities;
threshold choice.

If the model is probabilistic, check calibration.

Do not use a random split when real-world use requires predicting the future and a time-based split is more appropriate.

Try to find data leakage.

For each feature, ask:

Will this information be available when the model needs to make a prediction?

Look for:

future information;
post-outcome fields;
target-derived columns;
aggregates calculated using future data;
user-level leakage between train and test;
duplicates;
preprocessing before the split.

For each issue, state:

FEATURE
LEAKAGE PATH
WHY INVALID
SAFE FIX

Write SQL for:

[GOAL]

SCHEMA:
[...]

Before the query, explain:

TABLE GRAIN
JOIN KEYS
FILTERS
TIMEZONE
METRIC DEFINITION

Check many-to-many joins in particular.

After the query, add validation queries for:

- row count;
- duplicate keys;
- NULL rate;
- reconciliation total.

Do not use SELECT DISTINCT to hide a bad join.

Audit this SQL:

"""
[SQL]
"""

Check correctness first, then performance.

CORRECTNESS:

- join cardinality;
- NULL handling;
- date boundaries;
- timezone;
- double counting;
- filters;
- grouping;
- window functions.

PERFORMANCE:

- index usage;
- unnecessary scans;
- non-sargable predicates;
- repeated subqueries;
- CTE and materialization behavior;
- sort and hash cost.

Do not optimize the query if that would change its result.

For each issue, show a concrete example of an incorrect result.

I need an Excel formula for:

[PROBLEM]

Excel version:
[...]

Columns:
[...]

Provide:

1. the simplest formula;
2. a modern formula if XLOOKUP/FILTER/LET/LAMBDA are available;
3. an alternative for older Excel, if needed.

Explain how it handles:

blank cells;
errors;
duplicates;
missing matches.

Warn me if the formula can return an incorrect result when keys are duplicated.

Audit this Excel workbook.

Check:

- formulas;
- hard-coded values;
- broken references;
- external links;
- circular references;
- hidden sheets;
- hidden rows or columns;
- inconsistent formulas;
- merged cells;
- duplicated logic;
- pivot sources;
- named ranges.

Look in particular for:

a column where one formula differs from the rest of the range.

For each issue, state:

SHEET
CELL/RANGE
PROBLEM
EXPECTED
CURRENT
RISK

I want a pivot analysis for:

[GOAL]

DATA:
[...]

Propose:

ROWS
COLUMNS
VALUES
FILTERS

Check the dataset grain first.

If summing a particular column does not make business sense, do not sum it just because it is numeric.

Suggest an appropriate aggregation:

SUM
COUNT
DISTINCT COUNT
AVERAGE
MEDIAN, if the tool allows it
MIN/MAX

Design an analytics dashboard for:

[AUDIENCE]

GOAL:
[...]

Do not try to display all the data.

First define 5 to 8 key questions the dashboard should answer.

Then, for each widget, define:

QUESTION
METRIC
VISUAL
FILTER
TIME RANGE
DRILLDOWN

Divide the dashboard into:

EXECUTIVE SUMMARY
TRENDS
SEGMENTS
DIAGNOSTICS

Avoid vanity metrics with no actionable value.

Choose the best chart type for this data:

[GOAL]

DATA:
[...]

Do not automatically use a bar chart.

Consider:

line;
bar;
scatter;
histogram;
box plot;
heatmap;
area;
table.

Explain why the selected chart best fits the question.

Avoid a pie chart when there are many categories or small differences.

Audit this chart.

Check:

AXIS SCALE
ZERO BASELINE
LOG SCALE
LABELS
UNITS
SORT ORDER
TRUNCATION
COLOR DEPENDENCE
OVERPLOTTING
DUAL AXES

Look in particular for visual tricks that exaggerate differences.

Answer:

WHAT IT SHOWS
WHAT IT MAY MISLEAD ABOUT
HOW TO IMPROVE

Analyze this result:

[RESULT]

Separate:

STATISTICAL SIGNIFICANCE
PRACTICAL SIGNIFICANCE

State:

effect size;
confidence interval;
sample size;
baseline value.

If the change is statistically significant but negligible for the business, say so clearly.

If it is not statistically significant but the interval allows for a large effect, flag the uncertainty.

I want to assess whether:

[X]

causes:

[Y]

Do not use ordinary correlation as proof.

Identify:

TREATMENT / EXPOSURE
OUTCOME
CONFOUNDERS
MEDIATORS
COLLIDERS

If possible, draw a conceptual causal DAG.

Assess which methods could help:

RCT
difference-in-differences
instrumental variables
regression discontinuity
matching
synthetic control

Do not use causal terminology if the design does not support it.

Segment the users.

Do not start with clustering without a business goal.

GOAL:
[...]

Possible dimensions:

recency
frequency
monetary value
product usage
tenure
channel
plan
engagement

Before segmentation:

standardize metrics where needed;
check outliers;
check missing values.

For each segment, define:

SIZE
CHARACTERISTICS
VALUE
BEHAVIOR
ACTIONABILITY

A segment that does not lead to a distinct business action may not be useful.

Find anomalies in:

[METRIC]

Do not classify every extreme value as an incident.

Take these into account:

trend;
seasonality;
day of week;
holidays;
campaigns;
known releases.

For each anomaly, state:

TIMESTAMP
OBSERVED
EXPECTED
DEVIATION
POSSIBLE EXPLANATION
CONFIDENCE

Separate:

EXPECTED EVENT
DATA ISSUE
REAL BUSINESS ANOMALY
UNKNOWN

Analyze the dataset for suspicious patterns.

Do not label a user a fraudster just because of one outlier.

Look for combinations of signals:

velocity;
duplicate identities;
shared devices;
shared payment methods;
geographic anomalies;
rapid retries;
unusual transaction patterns.

For each signal, state:

EVIDENCE
ALTERNATIVE EXPLANATION
RISK

Finally, use:

NORMAL
REVIEW
HIGH RISK

but do not claim criminal or fraudulent intent without appropriate evidence.

Compare these two data sources:

SOURCE A:
[...]

SOURCE B:
[...]

Define a matching key.

Check:

ROW COUNT
TOTALS
UNIQUE IDS
MISSING IDS
VALUE DIFFERENCES
DATE DIFFERENCES

For each mismatch, state:

KEY
A VALUE
B VALUE
DIFFERENCE
LIKELY CAUSE

Separate:

EXPECTED DIFFERENCE
DATA LAG
MAPPING ISSUE
TRUE MISMATCH
UNKNOWN

Audit this data pipeline:

SOURCE
→
INGESTION
→
TRANSFORMATION
→
WAREHOUSE
→
BI

Check each stage for:

INPUT
OUTPUT
SCHEMA
VALIDATION
RETRY
DEDUPLICATION
LATE DATA
BACKFILL
FAILURE HANDLING

Look in particular for:

silent data loss;
duplicate ingestion;
schema drift;
partial loads;
timezone shifts;
non-idempotent retries.

For each issue, propose:

DETECTION
PREVENTION
RECOVERY

=== DATA ANALYSIS QA ===

Before your final answer, check:

1. whether the dataset grain is clear;
2. whether ID columns actually need to be unique;
3. whether a JOIN duplicates data;
4. whether the denominator matches the KPI definition;
5. whether dates use the same timezone;
6. whether a missing value means 0 or unknown;
7. whether outliers are valid;
8. whether data leakage exists;
9. whether aggregation hides segment differences;
10. whether correlation is being turned into causation;
11. whether the sample size supports the conclusion;
12. whether the confidence interval changes the interpretation;
13. whether the result is reproducible.

If the dataset does not allow a conclusion:

say so.

Do not draw a conclusion just because the user expects an answer.

I am looking for:

[PRODUCT]

COUNTRY OF PURCHASE:
[...]

BUDGET:
[...]

PRIMARY USE:
[...]

SECONDARY USE:
[...]

PRIORITIES:
1.
2.
3.
4.

IRRELEVANT CRITERIA:
[...]

REQUIRED:

- use current prices and availability;
- check the exact model, SKU, and revision;
- prefer official specifications for technical data;
- use independent tests for real-world performance;
- check known issues and long-term experiences;
- check warranty and service availability in my country.

Separate:

SPECIFICATIONS
MEASURED PERFORMANCE
KNOWN ISSUES
PRICE
WARRANTY
LONG-TERM VALUE

Do not recommend a product just because it has better numbers on paper.

Compare these products:

A:
[...]

B:
[...]

C:
[...]

Criteria:

[list criteria]

For each criterion, separate:

WHAT THE SPEC SHEET SAYS
WHAT INDEPENDENT TESTS SHOW
REAL-WORLD LIMITATIONS

Check these in particular:

- different regional versions;
- hardware revision;
- firmware status;
- panel lottery;
- storage/RAM configuration;
- bundled accessories;
- warranty.

If two models share the same name but have different components by market, warn me.

I am looking for a laptop.

BUDGET:
[...]

COUNTRY:
[...]

USE CASES:

programming:
[...]

AI/ML:
[...]

gaming:
[...]

video editing:
[...]

office:
[...]

PORTABILITY:
important / not important

BATTERY:
[...]

DISPLAY:
[...]

NOISE:
[...]

PORTS:
[...]

RAM:
[...]

SSD:
[...]

OS:
[...]

Analyze these in particular:

CPU sustained performance
GPU performance
VRAM
RAM upgradeability
SSD slots
display brightness
color accuracy
PWM/flicker
thermals
fan noise
battery life
weight
charger
ports
webcam
keyboard
build quality

Do not use a peak benchmark result as proof of sustained performance if the laptop throttles.

For AI workloads, check these in particular:

VRAM
CUDA/ROCm/Metal support
framework compatibility
memory bandwidth

At the end, separate:

BEST FOR MY USE CASE
BEST VALUE
BEST IF I PRIORITIZE BATTERY
BEST IF I PRIORITIZE PERFORMANCE

Compare these GPUs:

[MODEL A]
[MODEL B]

My use case:

resolution:
[...]

games/workloads:
[...]

AI:
[...]

video:
[...]

PSU:
[...]

CPU:
[...]

Case:
[...]

Check:

actual gaming performance;
1% lows;
ray tracing;
upscaling;
frame generation;
VRAM;
power draw;
thermals;
noise;
driver ecosystem;
compute support.

For AI, check these in particular:

framework compatibility;
VRAM limitations;
precision support;
ROCm/CUDA/other ecosystem.

Do not use only synthetic benchmarks.

Show real results in workloads relevant to my situation.

I am looking for a monitor.

BUDGET:
[...]

SIZE:
[...]

RESOLUTION:
[...]

VIEWING DISTANCE:
[...]

USE:
programming / text / gaming / design / video

REFRESH RATE:
[...]

GPU:
[...]

NIGHT USE:
[...]

EYE SENSITIVITY:
[...]

HDR:
[...]

USB-C/KVM:
[...]

Check:

panel type;
subpixel layout;
text clarity;
brightness;
contrast;
black uniformity;
PWM/flicker;
response time;
overshoot;
VRR;
actual HDR capability;
color accuracy;
viewing angles;
burn-in risk if OLED;
coating;
ergonomics.

Distinguish in particular:

marketing HDR
from
genuinely useful HDR.

If it is OLED, analyze:

text fringing;
static UI risk;
pixel refresh;
warranty against burn-in.

Compare these phones:

[MODELS]

My priorities:

camera:
[...]

battery:
[...]

display:
[...]

performance:
[...]

updates:
[...]

size:
[...]

storage:
[...]

budget:
[...]

Check:

main camera sensor;
ultrawide;
telephoto;
video;
OIS;
low-light;
shutter lag;
battery endurance;
charging;
thermal throttling;
display brightness;
PWM;
software update policy;
regional modem/bands;
storage standard;
repairability.

Do not rate the camera by megapixels alone.

Check real camera samples and independent measurements in particular.

Analyze this used phone purchase:

[MODEL]

AGE:
[...]

PRICE:
[...]

Battery health:
[...]

Condition:
[...]

Check common problems for this model.

Create a physical inspection checklist:

- screen;
- touch;
- cameras;
- stabilization;
- Face ID/fingerprint;
- speakers;
- microphone;
- charging port;
- wireless charging;
- Wi-Fi/Bluetooth;
- battery;
- water damage indicators;
- serial/IMEI status;
- parts pairing;
- repair history.

For an iPhone, also check:

battery/parts warnings;
True Tone;
Face ID;
camera stabilization;
originality of the display, camera, and battery where possible.

Compare these SSDs:

[MODELS]

USE:
OS / gaming / workstation / NAS / video editing

CAPACITY:
[...]

INTERFACE:
[...]

Check:

NAND type;
DRAM / DRAM-less;
controller;
SLC cache;
sustained write;
random performance;
thermal behavior;
endurance/TBW;
warranty;
firmware history.

Do not use only maximum sequential read/write speed.

Check performance after the SLC cache fills in particular.

Compare these RAM kits:

[MODELS]

PLATFORM:
CPU:
motherboard:

Capacity:
[...]

DIMM count:
[...]

XMP/EXPO:
[...]

Check:

frequency;
timings;
voltage;
IC/revision, if known;
QVL;
memory-controller limits;
stability;
dual-rank/single-rank, when relevant.

Do not assume a higher frequency automatically means better real-world performance.

Analyze latency and the platform-specific sweet spot.

Assess a PSU for this system:

CPU:
GPU:
drives:
fans:
future upgrade:
[...]

PSU candidate:
[...]

Do not just add up the TDP values.

Analyze:

transient spikes;
12V capacity;
ATX standard;
12V-2x6 / 12VHPWR;
efficiency;
platform/OEM;
protections;
fan behavior;
warranty.

Check independent PSU tests if available.

Check these in particular:

OCP
OVP
UVP
OTP
SCP
OPP

and the actual platform quality, not just the 80 Plus rating.

I am looking for a router or mesh system.

Apartment/house:
[...]

Area:
[...]

Floors:
[...]

Number of devices:
[...]

Internet speed:
[...]

Wi-Fi generation:
[...]

Backhaul:
wired / wireless

Budget:
[...]

Check:

radios;
bandwidth;
MIMO;
6 GHz;
channel width;
wired ports;
2.5/10GbE;
mesh backhaul;
firmware support;
security updates;
VLAN;
VPN;
SQM/QoS;
stability.

Do not use the maximum advertised AX/BE speed as an estimate of real throughput.

Compare these TVs:

[MODELS]

SCREEN SIZE:
[...]

VIEWING DISTANCE:
[...]

CONTENT:
movies / sports / gaming / daytime TV

ROOM:
bright / dark / mixed

CONSOLE/PC:
[...]

Check:

panel type;
native refresh rate;
local dimming;
peak brightness;
black level;
HDR tone mapping;
Dolby Vision/HDR10+;
input lag;
VRR;
HDMI 2.1 bandwidth;
burn-in risk;
motion handling;
upscaling.

Check regional panel differences for the same model in particular.

I am looking for headphones.

TYPE:
over-ear / in-ear / open-back

USE:
music / gaming / calls / travel

DEVICE:
[...]

BUDGET:
[...]

Priorities:
comfort;
sound;
ANC;
microphone;
battery;
latency.

Do not rely only on the manufacturer's frequency response claims.

Use independent measurements where available.

Analyze:

frequency response;
distortion;
imaging;
soundstage;
ANC;
transparency;
mic quality;
codec support;
latency;
comfort;
clamp;
weight.

I am looking for a camera:

[MODELS]

USE:
photo / video / wildlife / travel / studio / YouTube

BUDGET:
[...]

EXISTING LENSES:
[...]

Priorities:
[...]

Compare:

sensor;
dynamic range;
autofocus;
rolling shutter;
IBIS;
video modes;
crop;
overheating;
battery;
lens ecosystem;
media;
ports.

Do not compare megapixels alone.

Analyze the cost of the complete system, including the lenses I need.

Analyze this used car:

MODEL:
[...]

YEAR:
[...]

ENGINE:
[...]

TRANSMISSION:
[...]

MILEAGE:
[...]

PRICE:
[...]

COUNTRY:
[...]

Find common problems for this exact combination:

model + year + engine + transmission.

Separate:

ENGINE
TRANSMISSION
COOLING
ELECTRICS
SUSPENSION
BODY/CORROSION
EMISSIONS
ADAS
INFOTAINMENT

For each problem, state:

SYMPTOM
WHEN IT TYPICALLY APPEARS
HOW TO CHECK
REPAIR COST RANGE, if reliable local data are available
SEVERITY

Do not generalize an issue from another engine version to this one without evidence.

Create a checklist for checking whether the mileage may have been rolled back.

CAR:
[...]

YEAR:
[...]

Do not rely only on the instrument cluster.

Identify modules that may store mileage or usage data on this model, for example:

instrument cluster;
ECU;
BCM;
ABS;
transmission;
key data;
service records;
inspection records;
telematics.

But do not assume every module stores mileage on every model.

For each method, label:

RELIABILITY
CAN BE ALTERED?
WHAT TO COMPARE

Also include physical wear indicators, but emphasize that they are not proof on their own.

Help me interpret OBD diagnostics.

CAR:
[...]

ENGINE:
[...]

CODES:
[...]

LIVE DATA:
[...]

Do not replace parts based only on one DTC.

For each code, state:

WHAT IT MEANS
COMMON ROOT CAUSES
RELATED LIVE DATA
TESTS BEFORE PART REPLACEMENT

Distinguish:

ROOT CAUSE
SECONDARY CODE
CONSEQUENCE

If freeze-frame data are available, use them for context.

Compare these EVs:

[MODELS]

COUNTRY:
[...]

ANNUAL DISTANCE:
[...]

HOME CHARGING:
yes/no

CLIMATE:
[...]

TRIPS:
[...]

Analyze:

usable battery;
real range;
winter range;
highway range;
charging curve;
peak versus sustained charging;
battery chemistry;
thermal management;
battery warranty;
degradation;
efficiency;
heat pump;
fast-charging network compatibility.

Do not use only the WLTP/EPA figure without real-world tests.

I am looking for:

[refrigerator / washing machine / air conditioner / other]

BUDGET:
[...]

DIMENSIONS:
[...]

USE:
[...]

Check:

energy use;
noise;
capacity;
warranty;
spare parts;
service;
common failures;
maintenance.

Consider owner experiences only after checking technical specifications and professional tests.

Do not choose a model only because it has the most features if they add complexity or failure points without providing a benefit.

I need an air conditioner.

SPACE:
[...]

m²:
[...]

ceiling height:
[...]

insulation:
[...]

orientation:
[...]

floor:
[...]

number of people:
[...]

appliances:
[...]

location/climate:
[...]

Primary use:
cooling / heating / both

Do not determine the required capacity based only on square meters.

Consider the heat load.

Compare models by:

minimum modulation;
maximum heating capacity;
COP/SCOP;
SEER;
low-temperature performance;
noise;
defrost behavior;
service;
spare parts;
warranty.

For heating, check performance at low outdoor temperatures in particular.

Compare these tools:

[MODELS]

USE:
DIY / professional

FREQUENCY:
[...]

Material:
[...]

Priorities:
[...]

Check:

real power;
duty cycle;
ergonomics;
battery platform;
service;
parts;
accessories;
durability.

For a cordless system, analyze the cost of the whole ecosystem, not just the price of one tool.

I am looking for:

[PRODUCT]

COUNTRY OF PURCHASE:
[...]

BUDGET:
[...]

PRIMARY USE:
[...]

SECONDARY USE:
[...]

PRIORITIES:
1.
2.
3.
4.

IRRELEVANT CRITERIA:
[...]

REQUIRED:

- use current prices and availability;
- check the exact model, SKU, and revision;
- prefer official specifications for technical data;
- use independent tests for real-world performance;
- check known issues and long-term experiences;
- check warranty and service availability in my country.

Separate:

SPECIFICATIONS
MEASURED PERFORMANCE
KNOWN ISSUES
PRICE
WARRANTY
LONG-TERM VALUE

Do not recommend a product just because it has better numbers on paper.

Compare these products:

A:
[...]

B:
[...]

C:
[...]

Criteria:

[list criteria]

For each criterion, separate:

WHAT THE SPEC SHEET SAYS
WHAT INDEPENDENT TESTS SHOW
REAL-WORLD LIMITATIONS

Check these in particular:

- different regional versions;
- hardware revision;
- firmware status;
- panel lottery;
- storage/RAM configuration;
- bundled accessories;
- warranty.

If two models share the same name but have different components by market, warn me.

I am looking for a laptop.

BUDGET:
[...]

COUNTRY:
[...]

USE CASES:

programming:
[...]

AI/ML:
[...]

gaming:
[...]

video editing:
[...]

office:
[...]

PORTABILITY:
important / not important

BATTERY:
[...]

DISPLAY:
[...]

NOISE:
[...]

PORTS:
[...]

RAM:
[...]

SSD:
[...]

OS:
[...]

Analyze these in particular:

CPU sustained performance
GPU performance
VRAM
RAM upgradeability
SSD slots
display brightness
color accuracy
PWM/flicker
thermals
fan noise
battery life
weight
charger
ports
webcam
keyboard
build quality

Do not use a peak benchmark result as proof of sustained performance if the laptop throttles.

For AI workloads, check these in particular:

VRAM
CUDA/ROCm/Metal support
framework compatibility
memory bandwidth

At the end, separate:

BEST FOR MY USE CASE
BEST VALUE
BEST IF I PRIORITIZE BATTERY
BEST IF I PRIORITIZE PERFORMANCE

Compare these GPUs:

[MODEL A]
[MODEL B]

My use case:

resolution:
[...]

games/workloads:
[...]

AI:
[...]

video:
[...]

PSU:
[...]

CPU:
[...]

Case:
[...]

Check:

actual gaming performance;
1% lows;
ray tracing;
upscaling;
frame generation;
VRAM;
power draw;
thermals;
noise;
driver ecosystem;
compute support.

For AI, check these in particular:

framework compatibility;
VRAM limitations;
precision support;
ROCm/CUDA/other ecosystem.

Do not use only synthetic benchmarks.

Show real results in workloads relevant to my situation.

I am looking for a monitor.

BUDGET:
[...]

SIZE:
[...]

RESOLUTION:
[...]

VIEWING DISTANCE:
[...]

USE:
programming / text / gaming / design / video

REFRESH RATE:
[...]

GPU:
[...]

NIGHT USE:
[...]

EYE SENSITIVITY:
[...]

HDR:
[...]

USB-C/KVM:
[...]

Check:

panel type;
subpixel layout;
text clarity;
brightness;
contrast;
black uniformity;
PWM/flicker;
response time;
overshoot;
VRR;
actual HDR capability;
color accuracy;
viewing angles;
burn-in risk if OLED;
coating;
ergonomics.

Distinguish in particular:

marketing HDR
from
genuinely useful HDR.

If it is OLED, analyze:

text fringing;
static UI risk;
pixel refresh;
warranty against burn-in.

Compare these phones:

[MODELS]

My priorities:

camera:
[...]

battery:
[...]

display:
[...]

performance:
[...]

updates:
[...]

size:
[...]

storage:
[...]

budget:
[...]

Check:

main camera sensor;
ultrawide;
telephoto;
video;
OIS;
low-light;
shutter lag;
battery endurance;
charging;
thermal throttling;
display brightness;
PWM;
software update policy;
regional modem/bands;
storage standard;
repairability.

Do not rate the camera by megapixels alone.

Check real camera samples and independent measurements in particular.

Analyze this used phone purchase:

[MODEL]

AGE:
[...]

PRICE:
[...]

Battery health:
[...]

Condition:
[...]

Check common problems for this model.

Create a physical inspection checklist:

- screen;
- touch;
- cameras;
- stabilization;
- Face ID/fingerprint;
- speakers;
- microphone;
- charging port;
- wireless charging;
- Wi-Fi/Bluetooth;
- battery;
- water damage indicators;
- serial/IMEI status;
- parts pairing;
- repair history.

For an iPhone, also check:

battery/parts warnings;
True Tone;
Face ID;
camera stabilization;
originality of the display, camera, and battery where possible.

Compare these SSDs:

[MODELS]

USE:
OS / gaming / workstation / NAS / video editing

CAPACITY:
[...]

INTERFACE:
[...]

Check:

NAND type;
DRAM / DRAM-less;
controller;
SLC cache;
sustained write;
random performance;
thermal behavior;
endurance/TBW;
warranty;
firmware history.

Do not use only maximum sequential read/write speed.

Check performance after the SLC cache fills in particular.

Compare these RAM kits:

[MODELS]

PLATFORM:
CPU:
motherboard:

Capacity:
[...]

DIMM count:
[...]

XMP/EXPO:
[...]

Check:

frequency;
timings;
voltage;
IC/revision, if known;
QVL;
memory-controller limits;
stability;
dual-rank/single-rank, when relevant.

Do not assume a higher frequency automatically means better real-world performance.

Analyze latency and the platform-specific sweet spot.

Assess a PSU for this system:

CPU:
GPU:
drives:
fans:
future upgrade:
[...]

PSU candidate:
[...]

Do not just add up the TDP values.

Analyze:

transient spikes;
12V capacity;
ATX standard;
12V-2x6 / 12VHPWR;
efficiency;
platform/OEM;
protections;
fan behavior;
warranty.

Check independent PSU tests if available.

Check these in particular:

OCP
OVP
UVP
OTP
SCP
OPP

and the actual platform quality, not just the 80 Plus rating.

I am looking for a router or mesh system.

Apartment/house:
[...]

Area:
[...]

Floors:
[...]

Number of devices:
[...]

Internet speed:
[...]

Wi-Fi generation:
[...]

Backhaul:
wired / wireless

Budget:
[...]

Check:

radios;
bandwidth;
MIMO;
6 GHz;
channel width;
wired ports;
2.5/10GbE;
mesh backhaul;
firmware support;
security updates;
VLAN;
VPN;
SQM/QoS;
stability.

Do not use the maximum advertised AX/BE speed as an estimate of real throughput.

Compare these TVs:

[MODELS]

SCREEN SIZE:
[...]

VIEWING DISTANCE:
[...]

CONTENT:
movies / sports / gaming / daytime TV

ROOM:
bright / dark / mixed

CONSOLE/PC:
[...]

Check:

panel type;
native refresh rate;
local dimming;
peak brightness;
black level;
HDR tone mapping;
Dolby Vision/HDR10+;
input lag;
VRR;
HDMI 2.1 bandwidth;
burn-in risk;
motion handling;
upscaling.

Check regional panel differences for the same model in particular.

I am looking for headphones.

TYPE:
over-ear / in-ear / open-back

USE:
music / gaming / calls / travel

DEVICE:
[...]

BUDGET:
[...]

Priorities:
comfort;
sound;
ANC;
microphone;
battery;
latency.

Do not rely only on the manufacturer's frequency response claims.

Use independent measurements where available.

Analyze:

frequency response;
distortion;
imaging;
soundstage;
ANC;
transparency;
mic quality;
codec support;
latency;
comfort;
clamp;
weight.

I am looking for a camera:

[MODELS]

USE:
photo / video / wildlife / travel / studio / YouTube

BUDGET:
[...]

EXISTING LENSES:
[...]

Priorities:
[...]

Compare:

sensor;
dynamic range;
autofocus;
rolling shutter;
IBIS;
video modes;
crop;
overheating;
battery;
lens ecosystem;
media;
ports.

Do not compare megapixels alone.

Analyze the cost of the complete system, including the lenses I need.

Analyze this used car:

MODEL:
[...]

YEAR:
[...]

ENGINE:
[...]

TRANSMISSION:
[...]

MILEAGE:
[...]

PRICE:
[...]

COUNTRY:
[...]

Find common problems for this exact combination:

model + year + engine + transmission.

Separate:

ENGINE
TRANSMISSION
COOLING
ELECTRICS
SUSPENSION
BODY/CORROSION
EMISSIONS
ADAS
INFOTAINMENT

For each problem, state:

SYMPTOM
WHEN IT TYPICALLY APPEARS
HOW TO CHECK
REPAIR COST RANGE, if reliable local data are available
SEVERITY

Do not generalize an issue from another engine version to this one without evidence.

Create a checklist for checking whether the mileage may have been rolled back.

CAR:
[...]

YEAR:
[...]

Do not rely only on the instrument cluster.

Identify modules that may store mileage or usage data on this model, for example:

instrument cluster;
ECU;
BCM;
ABS;
transmission;
key data;
service records;
inspection records;
telematics.

But do not assume every module stores mileage on every model.

For each method, label:

RELIABILITY
CAN BE ALTERED?
WHAT TO COMPARE

Also include physical wear indicators, but emphasize that they are not proof on their own.

Help me interpret OBD diagnostics.

CAR:
[...]

ENGINE:
[...]

CODES:
[...]

LIVE DATA:
[...]

Do not replace parts based only on one DTC.

For each code, state:

WHAT IT MEANS
COMMON ROOT CAUSES
RELATED LIVE DATA
TESTS BEFORE PART REPLACEMENT

Distinguish:

ROOT CAUSE
SECONDARY CODE
CONSEQUENCE

If freeze-frame data are available, use them for context.

Compare these EVs:

[MODELS]

COUNTRY:
[...]

ANNUAL DISTANCE:
[...]

HOME CHARGING:
yes/no

CLIMATE:
[...]

TRIPS:
[...]

Analyze:

usable battery;
real range;
winter range;
highway range;
charging curve;
peak versus sustained charging;
battery chemistry;
thermal management;
battery warranty;
degradation;
efficiency;
heat pump;
fast-charging network compatibility.

Do not use only the WLTP/EPA figure without real-world tests.

I am looking for:

[refrigerator / washing machine / air conditioner / other]

BUDGET:
[...]

DIMENSIONS:
[...]

USE:
[...]

Check:

energy use;
noise;
capacity;
warranty;
spare parts;
service;
common failures;
maintenance.

Consider owner experiences only after checking technical specifications and professional tests.

Do not choose a model only because it has the most features if they add complexity or failure points without providing a benefit.

I need an air conditioner.

SPACE:
[...]

m²:
[...]

ceiling height:
[...]

insulation:
[...]

orientation:
[...]

floor:
[...]

number of people:
[...]

appliances:
[...]

location/climate:
[...]

Primary use:
cooling / heating / both

Do not determine the required capacity based only on square meters.

Consider the heat load.

Compare models by:

minimum modulation;
maximum heating capacity;
COP/SCOP;
SEER;
low-temperature performance;
noise;
defrost behavior;
service;
spare parts;
warranty.

For heating, check performance at low outdoor temperatures in particular.

Compare these tools:

[MODELS]

USE:
DIY / professional

FREQUENCY:
[...]

Material:
[...]

Priorities:
[...]

Check:

real power;
duty cycle;
ergonomics;
battery platform;
service;
parts;
accessories;
durability.

For a cordless system, analyze the cost of the whole ecosystem, not just the price of one tool.

Research this product:

[MODEL]

Check in particular whether the manufacturer changed any of these during the product's lifecycle:

panel;
SSD/NAND;
controller;
RAM;
battery;
Wi-Fi module;
power supply;
firmware;
another key component.

Create a table:

REVISION
REGION
COMPONENT CHANGE
PERFORMANCE IMPACT
HOW TO IDENTIFY

If the seller's listing does not identify the exact revision, flag the risk.

Audit the marketing claims for:

[PRODUCT]

MARKETING:
[...]

Classify each claim:

MEASURABLE
VAGUE
CONDITIONAL
POTENTIALLY MISLEADING

Try to find independent measurements.

Examples:

"AI-powered"
"up to 10 hours"
"2000 nits"
"40 Gbps"
"4K"
"Hi-Res"
"military grade"

Explain the conditions under which each claim can be achieved.

Find current prices for:

[PRODUCT]

COUNTRY:
[...]

Check several relevant retailers.

For each offer, state:

STORE
PRICE
STOCK
WARRANTY
EXACT MODEL/SKU
SHIPPING
EXTRA COSTS

Do not compare prices for different memory or configuration variants as if they were the same.

Calculate value according to my criterion:

[CRITERION]

If the product is significantly more expensive locally than in another country, include:

shipping;
VAT;
customs;
warranty implications

before concluding that importing it is worthwhile.

Analyze the price history of this product:

[MODEL]

If reliable data are available, find:

MSRP
TYPICAL PRICE
RECENT LOW
CURRENT PRICE

Try to determine whether the current "discount" is genuine or the reference price was artificially inflated.

Do not conclude that a deal is good just because of the discount percentage.

Compare:

NEW
REFURBISHED
USED

for:

[PRODUCT]

Analyze:

price;
warranty;
battery;
cosmetic condition;
replacement parts;
return policy;
seller reputation.

Distinguish:

manufacturer refurbished
authorized refurbished
seller refurbished
used

because they do not carry the same risk.

Analyze the warranty for:

[PRODUCT]

COUNTRY:
[...]

Check:

manufacturer warranty;
seller warranty;
duration;
battery coverage;
burn-in coverage;
dead-pixel policy;
accidental damage;
international warranty;
parts availability;
authorized service.

Do not use another country's warranty terms if I am buying the local version.

Find long-term user experiences for:

[MODEL]

Use community sources as a supplement, not as the only evidence.

Look for recurring patterns:

- the same failure;
- the same firmware problem;
- the same quality-control issue;
- long-term degradation.

Distinguish:

SINGLE ANECDOTE
REPEATED USER REPORT
KNOWN VENDOR ISSUE
INDEPENDENTLY VERIFIED ISSUE

Do not infer how common a problem is just from the number of Reddit posts.

Do not tell me which product is "best" until we define the criteria first.

Product:
[...]

Suggest 8 to 12 criteria that genuinely distinguish products in this category.

Then help me divide them into:

MUST HAVE
IMPORTANT
NICE TO HAVE
IRRELEVANT

Only then find products that fit.

Compare products using a decision matrix.

PRODUCTS:
[...]

CRITERIA:
[...]

WEIGHTS:
[...]

For every score:

provide evidence.

Do not assign a score based only on intuition.

If a criterion cannot be measured reliably:

label it UNKNOWN

instead of assigning an arbitrary number.

At the end, show how the result changes if we adjust the weights of the most important criteria.

Calculate the total cost of ownership for:

[PRODUCT]

PERIOD:
[3/5/10 years]

Include:

purchase price;
energy;
consumables;
maintenance;
subscription;
repairs;
accessories;
insurance;
resale value.

Do not compare only the initial price.

Show:

INITIAL COST
RUNNING COST
EXPECTED TOTAL COST
UNCERTAINTY

Assess how future-proof this product is:

[MODEL]

for:

[USE]

Do not use "future-proof" as a marketing phrase.

Analyze these specifics:

software support;
standards;
ports;
RAM/storage capacity;
performance headroom;
repairability;
battery replacement;
ecosystem compatibility.

Separate:

LIKELY USEFUL LIFE
TECHNICAL LIMIT
SOFTWARE SUPPORT LIMIT
ECONOMIC REPLACEMENT POINT

=== PRODUCT RESEARCH QA ===

Before your final answer, check:

1. exact model;
2. SKU;
3. region;
4. hardware revision;
5. current price;
6. current availability;
7. warranty;
8. firmware/software support;
9. independent measurements;
10. recurring known issues;
11. successor model;
12. EOL status.

Check these in particular:

- whether the review covers the same revision;
- whether the retailer uses the wrong photo or specification;
- whether the manufacturer states an "up to" value;
- whether the benchmark is a short burst instead of sustained performance;
- whether the local version has different components.

If something is unconfirmed:

UNKNOWN

is better than an assumption.

Create a travel plan.

DEPARTURE:
[...]

DESTINATION:
[...]

DEPARTURE DATE:
[...]

RETURN DATE:
[...]

NUMBER OF TRAVELERS:
[...]

CHILDREN:
[...]

BUDGET:
[...]

TRANSPORT:
[...]

ACCOMMODATION:
[...]

INTERESTS:
[...]

PACE:
relaxed / moderate / intensive

SPECIAL REQUIREMENTS:
[...]

Use current web sources for all information that may change.

Check these in particular:

- entry and visa requirements;
- flight, train, and bus schedules;
- prices;
- opening hours;
- public transport;
- local events;
- seasonal restrictions;
- possible construction or strikes;
- local rules.

Prefer official government sources for entry requirements.

At the end, create:

ITINERARY
TRANSPORT PLAN
BUDGET
RESERVATIONS NEEDED
BACKUP OPTIONS
WHAT TO RECHECK 24 TO 48 HOURS BEFORE THE TRIP

Find flights.

FROM:
[...]

TO:
[...]

DATE:
[...]

RETURN:
[...]

PASSENGERS:
[...]

BAGGAGE:
[...]

FLEXIBILITY:
[...]

MAX STOPS:
[...]

MAX TOTAL TRAVEL TIME:
[...]

BUDGET:
[...]

I prefer:
[morning / evening / no preference]

Compare:

TOTAL PRICE
BAGGAGE INCLUDED
NUMBER OF STOPS
TOTAL DURATION
LAYOVER DURATION
AIRPORT CHANGES
ARRIVAL TIME
FARE RULES

Do not compare only the base fare.

Include:

baggage;
seat fee, if relevant;
airport transfer;
booking fee;
self-transfer risk.

If the itinerary requires a self-transfer, clearly label it.

I want the most affordable period to travel.

FROM:
[...]

TO:
[...]

APPROXIMATE PERIOD:
[...]

TRIP LENGTH:
[X nights]

FLEXIBILITY:
± [number] days

Find the most affordable date combinations.

For each combination, show:

departure;
return;
airline;
total travel time;
stops;
real total price including necessary baggage.

Do not choose the cheapest option if it has an extremely poor itinerary without clearly flagging it.

Assess whether this layover is realistic.

AIRPORT:
[...]

ARRIVING FLIGHT:
[...]

DEPARTING FLIGHT:
[...]

LAYOVER:
[...]

SAME TICKET:
yes/no

BAGGAGE:
checked / cabin only

PASSPORT:
[...]

Check:

terminal change;
passport control;
security recheck;
baggage reclaim;
customs;
minimum connection time;
airport size;
historical operational issues, if relevant.

Separate:

LEGAL CONNECTION
COMFORTABLE CONNECTION
HIGH-RISK CONNECTION

If it is a self-transfer, analyze it separately.

Analyze this self-transfer itinerary.

LEG 1:
[...]

LEG 2:
[...]

AIRPORT:
[...]

LAYOVER:
[...]

TICKETS:
separate

BAGGAGE:
[...]

Check:

- whether I have to collect baggage;
- whether I have to check in again;
- whether I have to pass through immigration;
- whether I need a transit visa;
- terminal or airport change;
- check-in cutoff;
- consequences if the first flight is delayed.

Calculate a realistic buffer.

Do not treat two separate tickets as a protected connection.

Find the best options from:

[AIRPORT]

to:

[HOTEL / CITY CENTER / ADDRESS]

ARRIVAL TIME:
[...]

PASSENGERS:
[...]

BAGGAGE:
[...]

Compare:

TRAIN
METRO
BUS
TAXI
RIDESHARE
PRIVATE TRANSFER

For each option, state:

PRICE
DURATION
OPERATING HOURS
FREQUENCY
WALKING
LUGGAGE DIFFICULTY

Check in particular whether public transport is still running when I arrive.

Find a rail connection:

FROM:
[...]

TO:
[...]

DATE:
[...]

TIME:
[...]

PASSENGERS:
[...]

Use the current timetable.

Prioritize checking the official rail operator.

For each option, state:

departure;
arrival;
duration;
changes;
reservation requirement;
ticket price;
fare restrictions.

Check:

track works;
replacement buses;
strike notices;
seasonal timetable.

If the price is not available from an official source, clearly label it.

Create a road trip plan.

START:
[...]

END:
[...]

DATES:
[...]

CAR:
[...]

PASSENGERS:
[...]

MAX DRIVING PER DAY:
[...]

INTERESTS:
[...]

Check:

distance;
realistic driving time;
tolls;
vignettes;
fuel;
border requirements;
parking;
low-emission zones;
winter equipment;
ferries.

Do not use only an ideal Google-style driving time.

Include realistic breaks.

For each day, show:

ROUTE
DRIVE TIME
STOPS
FUEL/TOLLS
OVERNIGHT
BACKUP

Check the entry requirements.

PASSPORT:
[country]

DESTINATION:
[country]

PURPOSE:
tourism / business / other

DURATION:
[...]

ARRIVAL:
[...]

TRANSIT COUNTRIES:
[...]

Primarily use:

- official ministry of foreign affairs;
- embassy or consulate;
- official immigration portal;
- relevant government institutions.

Check:

visa requirement;
visa-free stay;
passport validity;
blank pages;
return/onward ticket;
proof of funds;
travel insurance;
registration requirements;
entry forms;
transit requirements.

Do not use a blog as the main evidence for visa rules.

State the date you checked.

Check whether I need a transit visa.

PASSPORT:
[...]

AIRPORT:
[...]

COUNTRY:
[...]

ONWARD DESTINATION:
[...]

TICKETS:
same / separate

BAGGAGE:
[...]

TERMINAL CHANGE:
[...]

LEAVING THE AIRPORT:
unknown

Check specifically:

airside transit;
landside transit;
baggage collection;
terminal transfer;
overnight rules.

Do not answer only based on whether I have a "connecting flight".

Explain the conditions under which a transit visa is required.

Find accommodation.

CITY:
[...]

DATES:
[...]

GUESTS:
[...]

BUDGET:
per night / total

MUST HAVE:
[...]

PREFERRED AREA:
[...]

MAX DISTANCE:
[...]

Compare:

TOTAL PRICE
TAXES/FEES
CANCELLATION
BREAKFAST
ROOM TYPE
BATHROOM
LOCATION
TRANSIT
CHECK-IN
REVIEWS

Do not use only the average rating.

Check recent reviews for recurring complaints about:

noise;
cleanliness;
air conditioning;
Wi-Fi;
room size;
elevator;
construction;
hidden fees.

If the price excludes city tax or resort fees, clearly add that.

Help me choose a neighborhood to stay in:

[CITY]

My priorities:

attractions:
[...]

nightlife:
[...]

quiet:
[...]

public transport:
[...]

budget:
[...]

walking:
[...]

Compare relevant neighborhoods by:

LOCATION
TRANSPORT
SAFETY CONTEXT
NOISE
RESTAURANTS
NIGHTLIFE
PRICE
AIRPORT ACCESS

Do not use generic "safe/unsafe" labels without context.

Explain how the neighborhood fits my specific kind of trip.

Analyze reviews for this hotel:

[HOTEL]

Focus primarily on recent reviews.

Find recurring patterns.

Group issues by:

CLEANLINESS
NOISE
STAFF
ROOM SIZE
AIR CONDITIONING
BEDS
BATHROOM
BREAKFAST
LOCATION
HIDDEN FEES

Separate:

SINGLE COMPLAINT
REPEATED COMPLAINT
RECENT TREND
OLD ISSUE

Check in particular whether the hotel was renovated after older negative reviews.

Analyze this listing:

[LISTING]

Check:

total price;
cleaning fee;
deposit;
cancellation;
check-in;
house rules;
host reviews;
location;
exact sleeping arrangements.

Look in particular for:

shared bathroom;
sofa bed presented as a bed;
no elevator;
noise;
late check-in fees;
tourist tax;
camera disclosure;
strict checkout duties.

Do not use the headline price if the final price includes large extra fees.

Create a detailed itinerary for:

[CITY]

DATES:
[...]

HOTEL AREA:
[...]

INTERESTS:
[...]

PACE:
[...]

Group each day by geographic area.

Minimize unnecessary backtracking across the city.

For each day, show:

MORNING
LUNCH AREA
AFTERNOON
EVENING

Add:

travel time;
reservation requirements;
opening hours;
ticket requirements.

Do not overfill the day.

Add buffer time.

Audit my itinerary:

[PLAN]

For each activity, check:

travel time;
queue time;
security;
meal time;
opening hours;
reservation;
realistic visit duration.

Identify:

IMPOSSIBLE
TOO TIGHT
REALISTIC
TOO EMPTY

Add buffer time for delays.

Check in particular whether moving between activities requires crossing to another part of the city.

Find the best attractions for me in:

[CITY]

INTERESTS:
[...]

DAYS:
[...]

Do not create a generic top 10.

For each attraction, state:

WHY IT MATCHES MY INTERESTS
TIME NEEDED
LOCATION
PRICE
RESERVATION
BEST TIME TO VISIT

Separate:

MUST BOOK EARLY
CAN DECIDE LATER
FREE / WALK-IN

Clearly flag tourist attractions that have a very high time cost relative to my interest in them.

Check:

[MUSEUM / ATTRACTION]

for this date:

[DATE]

Check the official website.

State:

OPENING HOURS
LAST ENTRY
TICKET PRICE
TIMED ENTRY?
CLOSED DAYS
SPECIAL CLOSURES
BAG POLICY

Do not use standard weekly opening hours if there are special hours for my date.

Find restaurants.

LOCATION:
[...]

DATE:
[...]

TIME:
[...]

PEOPLE:
[...]

CUISINE:
[...]

BUDGET:
[...]

DIETARY REQUIREMENTS:
[...]

I prefer:
local / casual / fine dining / quiet / view / other

For each restaurant, check:

current opening hours;
reservation availability, if possible;
price range;
menu;
dietary fit.

Do not recommend a restaurant just because it has a high aggregate rating.

Use recent reviews and multiple sources where needed.

I want to try typical food in:

[DESTINATION]

Find:

10 to 15 local dishes or drinks.

For each, state:

WHAT IT IS
MAIN INGREDIENTS
TYPICAL SERVING
WHERE IT IS COMMON
DIETARY CONCERNS

Then recommend the type of place where it makes sense to try the dish:

bakery;
market;
street food;
traditional restaurant;
dessert shop.

Do not send me only to tourist restaurants in the city center.

Create a budget travel plan.

DESTINATION:
[...]

DATES:
[...]

TOTAL BUDGET:
[...]

Include these in the budget:

transport to destination;
airport transfer;
accommodation;
local transport;
food;
attractions;
fees;
buffer.

Create:

FIXED COSTS
DAILY COSTS
OPTIONAL COSTS
EMERGENCY BUFFER

Do not create a plan that technically fits the budget only if I spend the absolute minimum every day.

Leave a realistic buffer.

Find flights.

FROM:
[...]

TO:
[...]

DATE:
[...]

RETURN:
[...]

PASSENGERS:
[...]

BAGGAGE:
[...]

FLEXIBILITY:
[...]

MAX STOPS:
[...]

MAX TOTAL TRAVEL TIME:
[...]

BUDGET:
[...]

I prefer:
[morning / evening / no preference]

Compare:

TOTAL PRICE
BAGGAGE INCLUDED
NUMBER OF STOPS
TOTAL DURATION
LAYOVER DURATION
AIRPORT CHANGES
ARRIVAL TIME
FARE RULES

Do not compare only the base fare.

Include:

baggage;
seat fee, if relevant;
airport transfer;
booking fee;
self-transfer risk.

If the itinerary requires a self-transfer, clearly label it.

I want the most affordable period to travel.

FROM:
[...]

TO:
[...]

APPROXIMATE PERIOD:
[...]

TRIP LENGTH:
[X nights]

FLEXIBILITY:
± [number] days

Find the most affordable date combinations.

For each combination, show:

departure;
return;
airline;
total travel time;
stops;
real total price including necessary baggage.

Do not choose the cheapest option if it has an extremely poor itinerary without clearly flagging it.

Assess whether this layover is realistic.

AIRPORT:
[...]

ARRIVING FLIGHT:
[...]

DEPARTING FLIGHT:
[...]

LAYOVER:
[...]

SAME TICKET:
yes/no

BAGGAGE:
checked / cabin only

PASSPORT:
[...]

Check:

terminal change;
passport control;
security recheck;
baggage reclaim;
customs;
minimum connection time;
airport size;
historical operational issues, if relevant.

Separate:

LEGAL CONNECTION
COMFORTABLE CONNECTION
HIGH-RISK CONNECTION

If it is a self-transfer, analyze it separately.

Analyze this self-transfer itinerary.

LEG 1:
[...]

LEG 2:
[...]

AIRPORT:
[...]

LAYOVER:
[...]

TICKETS:
separate

BAGGAGE:
[...]

Check:

- whether I have to collect baggage;
- whether I have to check in again;
- whether I have to pass through immigration;
- whether I need a transit visa;
- terminal or airport change;
- check-in cutoff;
- consequences if the first flight is delayed.

Calculate a realistic buffer.

Do not treat two separate tickets as a protected connection.

Find the best options from:

[AIRPORT]

to:

[HOTEL / CITY CENTER / ADDRESS]

ARRIVAL TIME:
[...]

PASSENGERS:
[...]

BAGGAGE:
[...]

Compare:

TRAIN
METRO
BUS
TAXI
RIDESHARE
PRIVATE TRANSFER

For each option, state:

PRICE
DURATION
OPERATING HOURS
FREQUENCY
WALKING
LUGGAGE DIFFICULTY

Check in particular whether public transport is still running when I arrive.

Find a rail connection:

FROM:
[...]

TO:
[...]

DATE:
[...]

TIME:
[...]

PASSENGERS:
[...]

Use the current timetable.

Prioritize checking the official rail operator.

For each option, state:

departure;
arrival;
duration;
changes;
reservation requirement;
ticket price;
fare restrictions.

Check:

track works;
replacement buses;
strike notices;
seasonal timetable.

If the price is not available from an official source, clearly label it.

Create a road trip plan.

START:
[...]

END:
[...]

DATES:
[...]

CAR:
[...]

PASSENGERS:
[...]

MAX DRIVING PER DAY:
[...]

INTERESTS:
[...]

Check:

distance;
realistic driving time;
tolls;
vignettes;
fuel;
border requirements;
parking;
low-emission zones;
winter equipment;
ferries.

Do not use only an ideal Google-style driving time.

Include realistic breaks.

For each day, show:

ROUTE
DRIVE TIME
STOPS
FUEL/TOLLS
OVERNIGHT
BACKUP

Check the entry requirements.

PASSPORT:
[country]

DESTINATION:
[country]

PURPOSE:
tourism / business / other

DURATION:
[...]

ARRIVAL:
[...]

TRANSIT COUNTRIES:
[...]

Primarily use:

- official ministry of foreign affairs;
- embassy or consulate;
- official immigration portal;
- relevant government institutions.

Check:

visa requirement;
visa-free stay;
passport validity;
blank pages;
return/onward ticket;
proof of funds;
travel insurance;
registration requirements;
entry forms;
transit requirements.

Do not use a blog as the main evidence for visa rules.

State the date you checked.

Check whether I need a transit visa.

PASSPORT:
[...]

AIRPORT:
[...]

COUNTRY:
[...]

ONWARD DESTINATION:
[...]

TICKETS:
same / separate

BAGGAGE:
[...]

TERMINAL CHANGE:
[...]

LEAVING THE AIRPORT:
unknown

Check specifically:

airside transit;
landside transit;
baggage collection;
terminal transfer;
overnight rules.

Do not answer only based on whether I have a "connecting flight".

Explain the conditions under which a transit visa is required.

Find accommodation.

CITY:
[...]

DATES:
[...]

GUESTS:
[...]

BUDGET:
per night / total

MUST HAVE:
[...]

PREFERRED AREA:
[...]

MAX DISTANCE:
[...]

Compare:

TOTAL PRICE
TAXES/FEES
CANCELLATION
BREAKFAST
ROOM TYPE
BATHROOM
LOCATION
TRANSIT
CHECK-IN
REVIEWS

Do not use only the average rating.

Check recent reviews for recurring complaints about:

noise;
cleanliness;
air conditioning;
Wi-Fi;
room size;
elevator;
construction;
hidden fees.

If the price excludes city tax or resort fees, clearly add that.

Help me choose a neighborhood to stay in:

[CITY]

My priorities:

attractions:
[...]

nightlife:
[...]

quiet:
[...]

public transport:
[...]

budget:
[...]

walking:
[...]

Compare relevant neighborhoods by:

LOCATION
TRANSPORT
SAFETY CONTEXT
NOISE
RESTAURANTS
NIGHTLIFE
PRICE
AIRPORT ACCESS

Do not use generic "safe/unsafe" labels without context.

Explain how the neighborhood fits my specific kind of trip.

Analyze reviews for this hotel:

[HOTEL]

Focus primarily on recent reviews.

Find recurring patterns.

Group issues by:

CLEANLINESS
NOISE
STAFF
ROOM SIZE
AIR CONDITIONING
BEDS
BATHROOM
BREAKFAST
LOCATION
HIDDEN FEES

Separate:

SINGLE COMPLAINT
REPEATED COMPLAINT
RECENT TREND
OLD ISSUE

Check in particular whether the hotel was renovated after older negative reviews.

Analyze this listing:

[LISTING]

Check:

total price;
cleaning fee;
deposit;
cancellation;
check-in;
house rules;
host reviews;
location;
exact sleeping arrangements.

Look in particular for:

shared bathroom;
sofa bed presented as a bed;
no elevator;
noise;
late check-in fees;
tourist tax;
camera disclosure;
strict checkout duties.

Do not use the headline price if the final price includes large extra fees.

Create a detailed itinerary for:

[CITY]

DATES:
[...]

HOTEL AREA:
[...]

INTERESTS:
[...]

PACE:
[...]

Group each day by geographic area.

Minimize unnecessary backtracking across the city.

For each day, show:

MORNING
LUNCH AREA
AFTERNOON
EVENING

Add:

travel time;
reservation requirements;
opening hours;
ticket requirements.

Do not overfill the day.

Add buffer time.

Audit my itinerary:

[PLAN]

For each activity, check:

travel time;
queue time;
security;
meal time;
opening hours;
reservation;
realistic visit duration.

Identify:

IMPOSSIBLE
TOO TIGHT
REALISTIC
TOO EMPTY

Add buffer time for delays.

Check in particular whether moving between activities requires crossing to another part of the city.

Find the best attractions for me in:

[CITY]

INTERESTS:
[...]

DAYS:
[...]

Do not create a generic top 10.

For each attraction, state:

WHY IT MATCHES MY INTERESTS
TIME NEEDED
LOCATION
PRICE
RESERVATION
BEST TIME TO VISIT

Separate:

MUST BOOK EARLY
CAN DECIDE LATER
FREE / WALK-IN

Clearly flag tourist attractions that have a very high time cost relative to my interest in them.

Check:

[MUSEUM / ATTRACTION]

for this date:

[DATE]

Check the official website.

State:

OPENING HOURS
LAST ENTRY
TICKET PRICE
TIMED ENTRY?
CLOSED DAYS
SPECIAL CLOSURES
BAG POLICY

Do not use standard weekly opening hours if there are special hours for my date.

Find restaurants.

LOCATION:
[...]

DATE:
[...]

TIME:
[...]

PEOPLE:
[...]

CUISINE:
[...]

BUDGET:
[...]

DIETARY REQUIREMENTS:
[...]

I prefer:
local / casual / fine dining / quiet / view / other

For each restaurant, check:

current opening hours;
reservation availability, if possible;
price range;
menu;
dietary fit.

Do not recommend a restaurant just because it has a high aggregate rating.

Use recent reviews and multiple sources where needed.

I want to try typical food in:

[DESTINATION]

Find:

10 to 15 local dishes or drinks.

For each, state:

WHAT IT IS
MAIN INGREDIENTS
TYPICAL SERVING
WHERE IT IS COMMON
DIETARY CONCERNS

Then recommend the type of place where it makes sense to try the dish:

bakery;
market;
street food;
traditional restaurant;
dessert shop.

Do not send me only to tourist restaurants in the city center.

Create a budget travel plan.

DESTINATION:
[...]

DATES:
[...]

TOTAL BUDGET:
[...]

Include these in the budget:

transport to destination;
airport transfer;
accommodation;
local transport;
food;
attractions;
fees;
buffer.

Create:

FIXED COSTS
DAILY COSTS
OPTIONAL COSTS
EMERGENCY BUFFER

Do not create a plan that technically fits the budget only if I spend the absolute minimum every day.

Leave a realistic buffer.

Create a budget travel plan.

DESTINATION:
[...]

DATES:
[...]

TOTAL BUDGET:
[...]

Include these in the budget:

transport to destination;
airport transfer;
accommodation;
local transport;
food;
attractions;
fees;
buffer.

Create:

FIXED COSTS
DAILY COSTS
OPTIONAL COSTS
EMERGENCY BUFFER

Do not create a plan that technically fits the budget only if I spend the absolute minimum every day.

Leave a realistic buffer.

Create a premium travel plan.

DESTINATION:
[...]

DATES:
[...]

PREFERENCES:
[...]

BUDGET:
[...]

Prioritize:

quality;
location;
convenience;
service;
unique experiences.

Do not automatically recommend the most expensive option.

Look for places where a higher price genuinely provides:

a better location;
better service;
less waiting;
special access;
a better room;
a genuinely better experience.

Create a family itinerary.

CHILDREN:
[ages]

DESTINATION:
[...]

DATES:
[...]

Take these into account:

nap time;
meal breaks;
toilets;
stroller;
walking distance;
weather;
child tickets;
height/age restrictions.

Do not use an adult itinerary and simply add a "kids activity".

Limit the number of major activities each day.

Add indoor backup options.

Create an itinerary for someone with limited mobility.

DESTINATION:
[...]

LIMITATIONS:
[...]

Check:

stairs;
elevators;
walking distance;
cobblestones;
wheelchair access;
public transport accessibility;
toilets;
resting places.

Label each attraction:

EASY
MODERATE
DIFFICULT

Do not assume a place is accessible just because its website says "accessible".

Check the specific access details.

I am planning a solo trip.

DESTINATION:
[...]

DATES:
[...]

Priorities:

central location;
late-night transport;
social activities;
single-room cost;
easy logistics.

Analyze:

transport after dark;
areas with active nightlife;
24-hour reception;
solo dining;
group tours;
common tourist scams.

Do not use dramatic generalizations about safety.

Focus on practical risks and prevention.

Analyze this destination for remote work.

DESTINATION:
[...]

DURATION:
[...]

WORK HOURS:
[...]

BUDGET:
[...]

Check:

internet quality;
mobile data;
coworking;
timezone;
power reliability;
long-stay accommodation;
visa/residency rules;
tax considerations at a high level;
public transport;
healthcare access.

For internet, distinguish:

advertised speed
from
real-world reliability.

Analyze timezone overlap with my work schedule in particular.

Compare internet options for:

[DESTINATION]

DURATION:
[...]

DATA NEED:
[...]

PHONE:
[...]

Compare:

LOCAL SIM
eSIM
ROAMING
POCKET WI-FI

For each, state:

PRICE
DATA
SPEED LIMIT
HOTSPOT
VALIDITY
NETWORK
ACTIVATION

Check that the phone supports the necessary bands or eSIM.

Do not rely only on an advertised "unlimited" plan without checking fair-use or throttling rules.

Find rental car options.

LOCATION:
[...]

DATES:
[...]

DRIVER AGE:
[...]

COUNTRY OF LICENSE:
[...]

CAR TYPE:
[...]

Check:

total price;
deposit;
excess;
insurance;
fuel policy;
mileage;
border crossing;
additional driver;
airport fee;
late return.

Explain these in particular:

CDW
excess
zero excess
third-party insurance

Do not treat "full insurance" as a standardized term without reading the terms.

Explain public transport in:

[CITY]

for a tourist staying:

[X days]

Check:

single ticket;
daily pass;
multi-day pass;
contactless payment;
mobile app;
airport inclusion;
zones;
validation rules;
night transport.

Calculate when a pass is worthwhile compared with individual rides.

Explain fines or risks if a ticket must be validated.

Assess whether this is worthwhile:

[CITY PASS]

My itinerary:

[...]

Calculate:

CARD PRICE
ACTIVITIES I WOULD ACTUALLY VISIT
NORMAL TICKET TOTAL
TRANSPORT VALUE

Do not count attractions I would not otherwise visit as "savings".

Check:

reservation requirements;
time slots;
daily limits;
exclusions.

Adjust the itinerary based on the forecast.

DESTINATION:
[...]

DATES:
[...]

Plan:
[...]

Check each day for:

temperature;
rain;
wind;
extreme weather.

Move:

outdoor activities to the best weather windows;
indoor activities to worse weather.

Do not treat a long-range forecast as certain.

Clearly mark the uncertainty when the date is farther away.

I am traveling for this event:

[EVENT]

CITY:
[...]

DATE:
[...]

Check:

exact venue;
doors/start time;
transport;
last public transport;
security and bag rules;
ticket rules;
nearby road closures;
expected crowd.

Recommend:

arrival buffer;
departure plan;
backup transport.

If we choose a hotel for the event, optimize for transport to the venue, not just the city center.

Create a logistics plan for:

[FESTIVAL/CONCERT]

Check:

entry rules;
ticket format;
ID requirements;
re-entry;
bag limits;
food/water rules;
power bank rules;
transport;
locker;
weather;
last train.

Create a checklist:

BEFORE LEAVING THE HOTEL
AT THE VENUE
AFTER THE EVENT

I have an overnight layover.

AIRPORT:
[...]

ARRIVAL:
[...]

DEPARTURE:
[...]

PASSPORT:
[...]

Check:

whether the airport stays open;
airside overnight rules;
immigration;
transit visa;
hotels;
sleep pods;
transport to the city;
first/last transport.

Compare:

STAY AT AIRPORT
AIRPORT HOTEL
CITY HOTEL

by cost, risk, and realistic time available for sleep.

Find common tourist scams in:

[DESTINATION]

Use current reliable sources and recent traveler reports as supplementary evidence.

For each, state:

HOW IT WORKS
WHERE IT HAPPENS
WARNING SIGNS
SAFE RESPONSE

Do not make the destination seem generally dangerous because of a few common scams.

Check health requirements for this trip.

FROM:
[...]

TO:
[...]

TRANSIT:
[...]

DATES:
[...]

Use official health and government sources.

Check:

required vaccinations;
recommended vaccinations;
health documentation for entry;
malaria risk, if relevant;
travel insurance considerations.

Separate:

REQUIRED FOR ENTRY
MEDICALLY RECOMMENDED
OPTIONAL

Do not confuse a recommendation with a mandatory entry requirement.

Use a browser workflow to find the best options for this trip.

DESTINATION:
[...]

DATES:
[...]

TRAVELERS:
[...]

BUDGET:
[...]

CONSTRAINTS:
[...]

You may:

- search;
- open results;
- compare options;
- check availability.

Do not:

- make a reservation;
- pay;
- accept a non-refundable condition;
- submit my personal information;

without my explicit confirmation.

At the end, create a shortlist with:

OPTION
TOTAL PRICE
WHAT IS INCLUDED
CANCELLATION
IMPORTANT CONDITIONS
SOURCE

When working through a browser or connected application:

before any action that creates a commitment, check:

- exact date;
- names;
- number of people;
- price;
- currency;
- refundability;
- baggage;
- cancellation;
- final total.

STOP before:

BOOK
BUY
PAY
CONFIRM
SUBMIT

if the action could create a financial or legal obligation.

Show me the final summary first.

=== TRAVEL QA ===

Before finalizing the plan, check:

1. dates;
2. local timezone;
3. visa and entry rules;
4. passport validity;
5. arrival/departure airport;
6. terminal;
7. baggage;
8. self-transfer;
9. total hotel price;
10. city taxes;
11. cancellation;
12. attraction opening hours;
13. reservation requirements;
14. public transport operating hours;
15. transfer times;
16. local holidays;
17. special closures;
18. strikes or disruptions, if any.

Check a current source for every time-sensitive detail.

If information should be checked again shortly before travel, label it:

RECHECK BEFORE TRAVEL.

Research this current topic:

[TOPIC]

Use the latest available information.

Do not just summarize the latest article.

First reconstruct what happened.

For each important detail, distinguish:

EVENT DATE
PUBLICATION DATE
LAST UPDATE DATE

Prioritize sources:

1. primary document or official statement;
2. direct data;
3. regulator or institution;
4. multiple independent, high-quality media outlets;
5. expert sources;
6. social media only as the source of a claim or signal.

Separate:

CONFIRMED FACTS
OFFICIAL CLAIMS
INDEPENDENTLY VERIFIED INFORMATION
MEDIA REPORTING
ANALYSIS
UNCONFIRMED INFORMATION

If sources disagree, show the disagreement.

Do not automatically treat the latest publication as the most reliable source.

Follow this breaking news event:

[TOPIC]

Do not draw conclusions from the first reports.

Create a timeline:

TIME
SOURCE
CLAIM
CONFIRMATION STATUS
LATER CORRECTION

Check in particular whether initial information was later:

CONFIRMED
CHANGED
WITHDRAWN
DENIED

If the number of victims, damage, participants, or another detail changed during reporting, use the latest confirmed value and explain the revision history.

Clearly label early information as preliminary.

Check this viral claim:

"[CLAIM]"

Do not check how many websites repeat the claim.

Find the earliest possible source.

Then determine:

WHO FIRST CLAIMED IT
WHEN
WHAT EXACTLY WAS CLAIMED
WHAT EVIDENCE WAS PROVIDED

Look for primary sources.

Then look for independent confirmation.

Classify it:

TRUE
MOSTLY TRUE
MISSING CONTEXT
MISLEADING
UNVERIFIED
FALSE
OUTDATED

For the result, explain exactly which part of the claim is or is not confirmed.

Check whether:

[PERSON]

actually said:

"[QUOTE]"

Do not use quote aggregator websites.

Look for:

- original video;
- audio;
- transcript;
- interview;
- speech;
- official statement;
- original post.

Check:

exact wording;
context;
date;
what was said immediately before and after the quote.

Classify it:

VERBATIM
ACCURATE PARAPHRASE
OUT OF CONTEXT
MISQUOTED
FABRICATED
UNVERIFIED

If only a clip is available, check the full context before reaching a conclusion.

Check this figure:

"[CLAIM WITH A NUMBER]"

Find the original dataset or institution.

Determine:

METRIC
POPULATION
GEOGRAPHY
TIME PERIOD
UNIT
METHODOLOGY

Check the math.

Check these in particular:

- percent versus percentage point;
- nominal versus real;
- total number versus rate;
- annual versus monthly;
- cumulative versus periodic;
- per capita versus total;
- absolute versus relative.

If the headline uses the number correctly but context changes its meaning, label it:

TECHNICALLY TRUE BUT MISLEADING.

Check this claim:

"[HIGHEST / LOWEST / RECORD]"

Do not accept it without a historical comparison.

Find the full relevant time series.

Determine:

START OF SERIES
METHODOLOGY CHANGES
BREAKS IN SERIES
NOMINAL / REAL
SEASONAL ADJUSTMENT

Check whether the "record" applies:

in nominal terms;
in real terms;
per capita;
as an absolute number;
as a percentage;
or only within a selected period.

If the claim is true only under a particular definition, say so.

Research the origin of this photo:

[IMAGE / LINK]

The goal is to establish:

- when it first appeared;
- where;
- what it actually shows;
- whether it is now being used with a different caption.

Look for:

earliest known appearance;
original uploader;
news archive;
official source;
older copies;
different captions.

Check in particular whether the photo is:

OLD IMAGE
WRONG LOCATION
WRONG EVENT
CROPPED
EDITED
CORRECTLY CONTEXTUALIZED

Do not conclude based only on visual similarity.

Check this viral video:

[VIDEO / LINK]

Determine:

original upload;
earliest date;
location;
event;
full-length version, if available.

Analyze:

audio continuity;
cuts;
cropping;
subtitles;
caption;
visible landmarks;
weather;
language;
signage;
vehicles;
uniforms;
shadows and time clues.

Do not conclude that a video was manipulated just because it looks unusual.

Separate:

CONTENT AUTHENTICITY
DATE
LOCATION
CAPTION ACCURACY

A video can be authentic but described incorrectly.

Check this screenshot:

[IMAGE]

Do not assume the screenshot is authentic.

Identify:

claimed source;
platform;
account;
date;
text;
URL, if visible.

Try to find the original post or an archived copy.

Check:

font/layout anomalies;
username;
timestamp format;
verified badge;
UI version;
cropping.

But a visual anomaly alone is not enough to prove forgery.

Final status:

ORIGINAL CONFIRMED
LIKELY AUTHENTIC
UNVERIFIED
ALTERED
FABRICATED

Analyze social media claims about:

[TOPIC]

Do not treat the number of likes, shares, or followers as evidence.

For each important post, determine:

WHO POSTED IT
ARE THEY A DIRECT WITNESS?
DO THEY PROVIDE EVIDENCE?
IS THERE INDEPENDENT CONFIRMATION?
HAS THE POST BEEN CORRECTED?

Separate:

PRIMARY WITNESS
OFFICIAL ACCOUNT
EXPERT COMMENTARY
SECOND-HAND CLAIM
ANONYMOUS CLAIM
AGGREGATOR

Use social media as a lead, not automatically as confirmation.

Analyze this official statement:

[STATEMENT]

Do not automatically treat it as a neutral description of events.

Separate:

VERIFIABLE FACTS
INSTITUTIONAL CLAIMS
LEGAL / PR INTERPRETATION
MISSING INFORMATION

Then find independent sources for key verifiable claims.

If independent confirmation is not possible:

clearly state that the information is currently only an official claim.

Compare these reports:

SOURCE A:
[...]

SOURCE B:
[...]

Do not decide which one is correct based on the outlet's reputation before checking the specific evidence.

For each disputed fact, create:

CLAIM
SOURCE A
SOURCE B
PRIMARY EVIDENCE
CURRENT STATUS

Check whether the difference can be explained by:

different publication time;
different definitions;
new information;
correction;
different eyewitness accounts.

Distinguish an actual contradiction from one article simply being older.

Analyze how different sources present:

[EVENT]

Do not use generic ideological labels as a substitute for analysis.

Compare:

HEADLINE
FACT SELECTION
SOURCE SELECTION
WORD CHOICE
WHAT IS OMITTED
WHAT IS EMPHASIZED

Separate:

differences in facts
from
differences in framing.

If two sources use the same facts but interpret them differently, show that clearly.

Reconstruct the complete timeline:

[TOPIC]

From:

[DATE]

To:

[DATE]

For each event, state:

DATE/TIME
EVENT
SOURCE
CONFIDENCE
WHY IT MATTERS

If a fact was later corrected, do not simply replace the old version.

Show:

INITIAL REPORT
CORRECTION
CURRENT UNDERSTANDING

At the end, identify the key turning points.

Compare the current state of this topic:

[TOPIC]

with its state on:

[PREVIOUS DATE]

Do not repeat information that has not changed.

Find only:

NEW FACTS
NEW DOCUMENTS
CORRECTIONS
REVERSALS
NEW STATEMENTS
STATUS CHANGES

For each new item, state:

what is new;
when it was confirmed;
source;
whether it changes the previous conclusion.

At the end, write:

WHAT ACTUALLY CHANGED.

Conduct publicly available OSINT research on:

[TOPIC / ORGANIZATION / DOMAIN]

Use only legitimately accessible public information.

Create a source map:

OFFICIAL SOURCES
CORPORATE RECORDS
DOMAIN/DNS
PUBLIC REPOSITORIES
ARCHIVED PAGES
NEWS
SOCIAL MEDIA
TECHNICAL METADATA

For each finding, state:
- source;
- date;
- evidence;
- verification level.

I need to complete this administrative task:

[TASK]

COUNTRY:
[...]

INSTITUTION:
[...]

MY STATUS:
[...]

DEADLINE:
[...]

First explain:

1. the purpose of the procedure;
2. which documents to prepare;
3. where to submit them;
4. whether an online process is available;
5. fees;
6. deadlines;
7. possible mistakes that could lead to rejection.

If the rules are time-sensitive:

check current official sources.

At the end, create a checklist:

BEFORE APPLYING
DURING THE APPLICATION
AFTER APPLYING

Help me fill in this form:

[FORM]

Do not invent information.

For each field, explain:

FIELD
WHAT IT MEANS
WHAT SHOULD BE ENTERED
FORMAT
COMMON MISTAKE

If the meaning depends on my status:

explain the different possibilities.

If I do not know something:

mark it as information I need to verify.

Write a formal request to an institution.

INSTITUTION:
[...]

SUBJECT:
[...]

FACTS:
[...]

WHAT I AM REQUESTING:
[...]

EVIDENCE:
[...]

TONE:
clear, formal, and without unnecessary drama.

Structure:

SUBJECT
BRIEF CONTEXT
FACTS
REQUEST
ATTACHMENTS

Do not include legal claims I have not verified.

Help me prepare a complaint.

PRODUCT/SERVICE:
[...]

PURCHASE DATE:
[...]

PROBLEM:
[...]

WHAT HAS ALREADY BEEN TRIED:
[...]

EVIDENCE:
[...]

DESIRED RESOLUTION:
[...]

First organize the facts chronologically.

Then write a complaint that is:

- concise;
- precise;
- documented;
- free of insults.

Separate:

WHAT HAPPENED
EVIDENCE
REQUESTED RESOLUTION
DEADLINE, if applicable

Analyze this problem with:

[internet / mobile / cable / another provider]

CONTRACT:
[...]

PROBLEM:
[...]

BILLS:
[...]

COMMUNICATION:
[...]

First determine:

what was agreed;
what was charged;
what was provided;
what differs.

Create a table:

PROMISED
ACTUAL
EVIDENCE
DIFFERENCE

Then create an escalation plan:

SUPPORT
FORMAL COMPLAINT
REGULATOR / OTHER CHANNEL

if relevant.

Check this bill.

BILL:
[...]

PREVIOUS BILLS:
[...]

AGREED PRICE:
[...]

Find:

BASE PRICE
DISCOUNTS
ADDITIONAL FEES
TAX
USAGE
ONE-OFF FEES

Recalculate the total independently.

Compare it with the previous period.

If there is a spike:

identify exactly which line item explains it.

Do not assume there is an error until the math and terms have been checked.

Research this current topic:

[TOPIC]

Use the latest available information.

Do not just summarize the latest article.

First reconstruct what happened.

For each important detail, distinguish:

EVENT DATE
PUBLICATION DATE
LAST UPDATE DATE

Prioritize sources:

1. primary document or official statement;
2. direct data;
3. regulator or institution;
4. multiple independent, high-quality media outlets;
5. expert sources;
6. social media only as the source of a claim or signal.

Separate:

CONFIRMED FACTS
OFFICIAL CLAIMS
INDEPENDENTLY VERIFIED INFORMATION
MEDIA REPORTING
ANALYSIS
UNCONFIRMED INFORMATION

If sources disagree, show the disagreement.

Do not automatically treat the latest publication as the most reliable source.

Follow this breaking news event:

[TOPIC]

Do not draw conclusions from the first reports.

Create a timeline:

TIME
SOURCE
CLAIM
CONFIRMATION STATUS
LATER CORRECTION

Check in particular whether initial information was later:

CONFIRMED
CHANGED
WITHDRAWN
DENIED

If the number of victims, damage, participants, or another detail changed during reporting, use the latest confirmed value and explain the revision history.

Clearly label early information as preliminary.

Check this viral claim:

"[CLAIM]"

Do not check how many websites repeat the claim.

Find the earliest possible source.

Then determine:

WHO FIRST CLAIMED IT
WHEN
WHAT EXACTLY WAS CLAIMED
WHAT EVIDENCE WAS PROVIDED

Look for primary sources.

Then look for independent confirmation.

Classify it:

TRUE
MOSTLY TRUE
MISSING CONTEXT
MISLEADING
UNVERIFIED
FALSE
OUTDATED

For the result, explain exactly which part of the claim is or is not confirmed.

Check whether:

[PERSON]

actually said:

"[QUOTE]"

Do not use quote aggregator websites.

Look for:

- original video;
- audio;
- transcript;
- interview;
- speech;
- official statement;
- original post.

Check:

exact wording;
context;
date;
what was said immediately before and after the quote.

Classify it:

VERBATIM
ACCURATE PARAPHRASE
OUT OF CONTEXT
MISQUOTED
FABRICATED
UNVERIFIED

If only a clip is available, check the full context before reaching a conclusion.

Check this figure:

"[CLAIM WITH A NUMBER]"

Find the original dataset or institution.

Determine:

METRIC
POPULATION
GEOGRAPHY
TIME PERIOD
UNIT
METHODOLOGY

Check the math.

Check these in particular:

- percent versus percentage point;
- nominal versus real;
- total number versus rate;
- annual versus monthly;
- cumulative versus periodic;
- per capita versus total;
- absolute versus relative.

If the headline uses the number correctly but context changes its meaning, label it:

TECHNICALLY TRUE BUT MISLEADING.

Check this claim:

"[HIGHEST / LOWEST / RECORD]"

Do not accept it without a historical comparison.

Find the full relevant time series.

Determine:

START OF SERIES
METHODOLOGY CHANGES
BREAKS IN SERIES
NOMINAL / REAL
SEASONAL ADJUSTMENT

Check whether the "record" applies:

in nominal terms;
in real terms;
per capita;
as an absolute number;
as a percentage;
or only within a selected period.

If the claim is true only under a particular definition, say so.

Research the origin of this photo:

[IMAGE / LINK]

The goal is to establish:

- when it first appeared;
- where;
- what it actually shows;
- whether it is now being used with a different caption.

Look for:

earliest known appearance;
original uploader;
news archive;
official source;
older copies;
different captions.

Check in particular whether the photo is:

OLD IMAGE
WRONG LOCATION
WRONG EVENT
CROPPED
EDITED
CORRECTLY CONTEXTUALIZED

Do not conclude based only on visual similarity.

Check this viral video:

[VIDEO / LINK]

Determine:

original upload;
earliest date;
location;
event;
full-length version, if available.

Analyze:

audio continuity;
cuts;
cropping;
subtitles;
caption;
visible landmarks;
weather;
language;
signage;
vehicles;
uniforms;
shadows and time clues.

Do not conclude that a video was manipulated just because it looks unusual.

Separate:

CONTENT AUTHENTICITY
DATE
LOCATION
CAPTION ACCURACY

A video can be authentic but described incorrectly.

Check this screenshot:

[IMAGE]

Do not assume the screenshot is authentic.

Identify:

claimed source;
platform;
account;
date;
text;
URL, if visible.

Try to find the original post or an archived copy.

Check:

font/layout anomalies;
username;
timestamp format;
verified badge;
UI version;
cropping.

But a visual anomaly alone is not enough to prove forgery.

Final status:

ORIGINAL CONFIRMED
LIKELY AUTHENTIC
UNVERIFIED
ALTERED
FABRICATED

Analyze social media claims about:

[TOPIC]

Do not treat the number of likes, shares, or followers as evidence.

For each important post, determine:

WHO POSTED IT
ARE THEY A DIRECT WITNESS?
DO THEY PROVIDE EVIDENCE?
IS THERE INDEPENDENT CONFIRMATION?
HAS THE POST BEEN CORRECTED?

Separate:

PRIMARY WITNESS
OFFICIAL ACCOUNT
EXPERT COMMENTARY
SECOND-HAND CLAIM
ANONYMOUS CLAIM
AGGREGATOR

Use social media as a lead, not automatically as confirmation.

Analyze this official statement:

[STATEMENT]

Do not automatically treat it as a neutral description of events.

Separate:

VERIFIABLE FACTS
INSTITUTIONAL CLAIMS
LEGAL / PR INTERPRETATION
MISSING INFORMATION

Then find independent sources for key verifiable claims.

If independent confirmation is not possible:

clearly state that the information is currently only an official claim.

Compare these reports:

SOURCE A:
[...]

SOURCE B:
[...]

Do not decide which one is correct based on the outlet's reputation before checking the specific evidence.

For each disputed fact, create:

CLAIM
SOURCE A
SOURCE B
PRIMARY EVIDENCE
CURRENT STATUS

Check whether the difference can be explained by:

different publication time;
different definitions;
new information;
correction;
different eyewitness accounts.

Distinguish an actual contradiction from one article simply being older.

Analyze how different sources present:

[EVENT]

Do not use generic ideological labels as a substitute for analysis.

Compare:

HEADLINE
FACT SELECTION
SOURCE SELECTION
WORD CHOICE
WHAT IS OMITTED
WHAT IS EMPHASIZED

Separate:

differences in facts
from
differences in framing.

If two sources use the same facts but interpret them differently, show that clearly.

Reconstruct the complete timeline:

[TOPIC]

From:

[DATE]

To:

[DATE]

For each event, state:

DATE/TIME
EVENT
SOURCE
CONFIDENCE
WHY IT MATTERS

If a fact was later corrected, do not simply replace the old version.

Show:

INITIAL REPORT
CORRECTION
CURRENT UNDERSTANDING

At the end, identify the key turning points.

Compare the current state of this topic:

[TOPIC]

with its state on:

[PREVIOUS DATE]

Do not repeat information that has not changed.

Find only:

NEW FACTS
NEW DOCUMENTS
CORRECTIONS
REVERSALS
NEW STATEMENTS
STATUS CHANGES

For each new item, state:

what is new;
when it was confirmed;
source;
whether it changes the previous conclusion.

At the end, write:

WHAT ACTUALLY CHANGED.

Conduct publicly available OSINT research on:

[TOPIC / ORGANIZATION / DOMAIN]

Use only legitimately accessible public information.

Create a source map:

OFFICIAL SOURCES
CORPORATE RECORDS
DOMAIN/DNS
PUBLIC REPOSITORIES
ARCHIVED PAGES
NEWS
SOCIAL MEDIA
TECHNICAL METADATA

For each finding, state:
- source;
- date;
- evidence;
- verification level.

Compare these offers:

A:
[...]

B:
[...]

C:
[...]

Do not look only at the headline price.

Calculate for each:

INITIAL COST
MONTHLY COST
MANDATORY FEES
OPTIONAL FEES
CONTRACT LENGTH
CANCELLATION COST
TOTAL COST

for this period:

[X months/years]

Check the fine print in particular.

At the end, show the trade-offs without hiding long-term costs.

Analyze this subscription:

[SERVICE]

PRICE:
[...]

HOW MUCH I USE IT:
[...]

ALTERNATIVES:
[...]

Calculate:

COST PER MONTH
COST PER YEAR
COST PER USE

Identify:

features I actually use;
features I pay for but do not use.

Recommend:

KEEP
DOWNGRADE
PAUSE
CANCEL

based only on my criteria and actual use.

Here are my subscriptions:

[LIST]

Create a table:

SERVICE
MONTHLY
ANNUAL
LAST USED
VALUE
DUPLICATE FUNCTION
CANCELLATION PRIORITY

Find:

- duplicate features;
- old subscriptions;
- trials that became paid plans;
- annual plans that are renewing.

Calculate how much I could save each year in different scenarios.

Create a realistic monthly budget.

INCOME:
[...]

FIXED EXPENSES:
[...]

VARIABLE:
[...]

DEBT:
[...]

SAVINGS:
[...]

GOALS:
[...]

Do not use the generic 50/30/20 rule if it does not fit reality.

Analyze my actual numbers first.

Create:

ESSENTIAL
FLEXIBLE
DISCRETIONARY
DEBT
SAVINGS

Then create:

BASE BUDGET
TIGHT BUDGET
COMFORTABLE BUDGET

Analyze my expenses:

[DATA]

Do not automatically criticize small purchases.

Find the largest structural sources of spending.

Classify them:

FIXED
NEGOTIABLE
REDUCIBLE
REMOVABLE

Prioritize changes by:

ANNUAL SAVINGS
EFFORT
IMPACT ON QUALITY OF LIFE

Look for large savings with a small negative effect first.

I am buying this used:

[PRODUCT]

MODEL:
[...]

AGE:
[...]

PRICE:
[...]

Create an inspection checklist for before the purchase.

Divide it into:

VISUAL CHECK
FUNCTIONAL TEST
SERIAL/IDENTITY CHECK
BATTERY/WEAR
KNOWN MODEL ISSUES
DOCUMENTATION
PRICE CHECK

For each test, explain:

what I do;
what I should normally expect;
which result is a red flag.

I received a quote for work:

[QUOTE]

JOB:
[...]

Analyze:

LABOR
MATERIAL
QUANTITY
UNIT PRICE
EXTRAS
WARRANTY

Find unclear items.

Create a list of questions to ask before accepting.

Check in particular:

what is included;
what is not included;
who pays for extra materials;
what happens if another problem is found.

I have a problem:

[PROBLEM]

DEVICE/SYSTEM:
[...]

SYMPTOM:
[...]

WHEN IT OCCURS:
[...]

WHAT I HAVE TRIED:
[...]

Create a troubleshooting tree.

Order:

1. safety check;
2. most common cause;
3. easiest test;
4. next possible cause;
5. when to stop DIY troubleshooting and call a professional.

Do not suggest opening a powered device or attempting risky repairs.

Device:

[MODEL]

Symptom:

[...]

Error code:

[...]

Find possible causes.

For each, state:

LIKELIHOOD
SAFE TEST
REPAIR COMPLEXITY
PART LIKELY NEEDED

Do not recommend buying a part before the cause is confirmed.

If the model has a service manual or official error-code description, use it as the primary source.

I am planning a renovation.

SPACE:
[...]

m²:
[...]

WORK:
[...]

BUDGET:
[...]

Create a cost estimate by category:

DEMOLITION
MATERIAL
LABOR
ELECTRICAL
PLUMBING
FINISHING
FURNITURE
CONTINGENCY

Separate:

MUST HAVE
CAN WAIT
COSMETIC

Add contingency scenarios:

+10%
+20%
+30%

Identify the items most likely to exceed the budget.

Create a moving plan.

FROM:
[...]

TO:
[...]

DATE:
[...]

HOUSEHOLD SIZE:
[...]

Create a timeline:

4 WEEKS BEFORE
2 WEEKS BEFORE
1 WEEK BEFORE
DAY BEFORE
MOVING DAY
AFTER THE MOVE

Include:

utilities;
documents;
address changes;
packing;
fragile items;
inventory;
cleaning;
keys;
internet;
mail.

Prioritize tasks with external deadlines.

Create a checklist for a trip.

DESTINATION:
[...]

DAYS:
[...]

PURPOSE:
[...]

LUGGAGE:
[...]

Categories:

DOCUMENTS
MONEY
PHONE/CHARGING
CLOTHES
MEDICINE
WORK
TRANSPORT
HOME BEFORE LEAVING

Do not generate 150 items.

Adapt the list to the actual trip.

These are my commitments today:

[LIST]

For each one, state:

DEADLINE
DURATION
IMPORTANCE
DEPENDENCY
ENERGY REQUIRED

Create a realistic schedule.

Include:

buffer;
breaks;
unexpected commitments.

Do not schedule 8 hours of focused work into 8 hours of available time.

Prioritize these tasks:

[TASK LIST]

For each, assess:

URGENCY
IMPACT
EFFORT
DEPENDENCY
CONSEQUENCE OF DELAY

Do not use only the Eisenhower matrix.

Find:

BLOCKERS
QUICK WINS
HIGH-LEVERAGE TASKS
DEADLINE CRITICAL

Create an execution order and explain the dependencies.

These are my commitments for the week:

[...]

Available time:

[...]

Create a plan that uses no more than:

[X%]

of the available time for things scheduled in advance.

Leave the rest as buffer.

Divide tasks into:

MUST COMPLETE
SHOULD COMPLETE
OPTIONAL

Identify tasks to delegate, automate, or postpone.

Review my plan:

[PLAN]

Look for:

- too many tasks;
- insufficient buffer;
- hidden dependencies;
- unrealistic time estimates;
- context switching.

Calculate how much time is actually planned.

If the plan is unrealistic:

reduce the scope instead of just shortening every activity.

I need to complete this administrative task:

[TASK]

COUNTRY:
[...]

INSTITUTION:
[...]

MY STATUS:
[...]

DEADLINE:
[...]

First explain:

1. the purpose of the procedure;
2. which documents to prepare;
3. where to submit them;
4. whether an online process is available;
5. fees;
6. deadlines;
7. possible mistakes that could lead to rejection.

If the rules are time-sensitive:

check current official sources.

At the end, create a checklist:

BEFORE APPLYING
DURING THE APPLICATION
AFTER APPLYING

Help me fill in this form:

[FORM]

Do not invent information.

For each field, explain:

FIELD
WHAT IT MEANS
WHAT SHOULD BE ENTERED
FORMAT
COMMON MISTAKE

If the meaning depends on my status:

explain the different possibilities.

If I do not know something:

mark it as information I need to verify.

Write a formal request to an institution.

INSTITUTION:
[...]

SUBJECT:
[...]

FACTS:
[...]

WHAT I AM REQUESTING:
[...]

EVIDENCE:
[...]

TONE:
clear, formal, and without unnecessary drama.

Structure:

SUBJECT
BRIEF CONTEXT
FACTS
REQUEST
ATTACHMENTS

Do not include legal claims I have not verified.

Help me prepare a complaint.

PRODUCT/SERVICE:
[...]

PURCHASE DATE:
[...]

PROBLEM:
[...]

WHAT HAS ALREADY BEEN TRIED:
[...]

EVIDENCE:
[...]

DESIRED RESOLUTION:
[...]

First organize the facts chronologically.

Then write a complaint that is:

- concise;
- precise;
- documented;
- free of insults.

Separate:

WHAT HAPPENED
EVIDENCE
REQUESTED RESOLUTION
DEADLINE, if applicable

Analyze this problem with:

[internet / mobile / cable / another provider]

CONTRACT:
[...]

PROBLEM:
[...]

BILLS:
[...]

COMMUNICATION:
[...]

First determine:

what was agreed;
what was charged;
what was provided;
what differs.

Create a table:

PROMISED
ACTUAL
EVIDENCE
DIFFERENCE

Then create an escalation plan:

SUPPORT
FORMAL COMPLAINT
REGULATOR / OTHER CHANNEL

if relevant.

Check this bill.

BILL:
[...]

PREVIOUS BILLS:
[...]

AGREED PRICE:
[...]

Find:

BASE PRICE
DISCOUNTS
ADDITIONAL FEES
TAX
USAGE
ONE-OFF FEES

Recalculate the total independently.

Compare it with the previous period.

If there is a spike:

identify exactly which line item explains it.

Do not assume there is an error until the math and terms have been checked.

Compare these offers:

A:
[...]

B:
[...]

C:
[...]

Do not look only at the headline price.

Calculate for each:

INITIAL COST
MONTHLY COST
MANDATORY FEES
OPTIONAL FEES
CONTRACT LENGTH
CANCELLATION COST
TOTAL COST

for this period:

[X months/years]

Check the fine print in particular.

At the end, show the trade-offs without hiding long-term costs.

Analyze this subscription:

[SERVICE]

PRICE:
[...]

HOW MUCH I USE IT:
[...]

ALTERNATIVES:
[...]

Calculate:

COST PER MONTH
COST PER YEAR
COST PER USE

Identify:

features I actually use;
features I pay for but do not use.

Recommend:

KEEP
DOWNGRADE
PAUSE
CANCEL

based only on my criteria and actual use.

Here are my subscriptions:

[LIST]

Create a table:

SERVICE
MONTHLY
ANNUAL
LAST USED
VALUE
DUPLICATE FUNCTION
CANCELLATION PRIORITY

Find:

- duplicate features;
- old subscriptions;
- trials that became paid plans;
- annual plans that are renewing.

Calculate how much I could save each year in different scenarios.

Create a realistic monthly budget.

INCOME:
[...]

FIXED EXPENSES:
[...]

VARIABLE:
[...]

DEBT:
[...]

SAVINGS:
[...]

GOALS:
[...]

Do not use the generic 50/30/20 rule if it does not fit reality.

Analyze my actual numbers first.

Create:

ESSENTIAL
FLEXIBLE
DISCRETIONARY
DEBT
SAVINGS

Then create:

BASE BUDGET
TIGHT BUDGET
COMFORTABLE BUDGET

Analyze my expenses:

[DATA]

Do not automatically criticize small purchases.

Find the largest structural sources of spending.

Classify them:

FIXED
NEGOTIABLE
REDUCIBLE
REMOVABLE

Prioritize changes by:

ANNUAL SAVINGS
EFFORT
IMPACT ON QUALITY OF LIFE

Look for large savings with a small negative effect first.

I am buying this used:

[PRODUCT]

MODEL:
[...]

AGE:
[...]

PRICE:
[...]

Create an inspection checklist for before the purchase.

Divide it into:

VISUAL CHECK
FUNCTIONAL TEST
SERIAL/IDENTITY CHECK
BATTERY/WEAR
KNOWN MODEL ISSUES
DOCUMENTATION
PRICE CHECK

For each test, explain:

what I do;
what I should normally expect;
which result is a red flag.

I received a quote for work:

[QUOTE]

JOB:
[...]

Analyze:

LABOR
MATERIAL
QUANTITY
UNIT PRICE
EXTRAS
WARRANTY

Find unclear items.

Create a list of questions to ask before accepting.

Check in particular:

what is included;
what is not included;
who pays for extra materials;
what happens if another problem is found.

I have a problem:

[PROBLEM]

DEVICE/SYSTEM:
[...]

SYMPTOM:
[...]

WHEN IT OCCURS:
[...]

WHAT I HAVE TRIED:
[...]

Create a troubleshooting tree.

Order:

1. safety check;
2. most common cause;
3. easiest test;
4. next possible cause;
5. when to stop DIY troubleshooting and call a professional.

Do not suggest opening a powered device or attempting risky repairs.

Device:

[MODEL]

Symptom:

[...]

Error code:

[...]

Find possible causes.

For each, state:

LIKELIHOOD
SAFE TEST
REPAIR COMPLEXITY
PART LIKELY NEEDED

Do not recommend buying a part before the cause is confirmed.

If the model has a service manual or official error-code description, use it as the primary source.

I am planning a renovation.

SPACE:
[...]

m²:
[...]

WORK:
[...]

BUDGET:
[...]

Create a cost estimate by category:

DEMOLITION
MATERIAL
LABOR
ELECTRICAL
PLUMBING
FINISHING
FURNITURE
CONTINGENCY

Separate:

MUST HAVE
CAN WAIT
COSMETIC

Add contingency scenarios:

+10%
+20%
+30%

Identify the items most likely to exceed the budget.

Create a moving plan.

FROM:
[...]

TO:
[...]

DATE:
[...]

HOUSEHOLD SIZE:
[...]

Create a timeline:

4 WEEKS BEFORE
2 WEEKS BEFORE
1 WEEK BEFORE
DAY BEFORE
MOVING DAY
AFTER THE MOVE

Include:

utilities;
documents;
address changes;
packing;
fragile items;
inventory;
cleaning;
keys;
internet;
mail.

Prioritize tasks with external deadlines.

Create a checklist for a trip.

DESTINATION:
[...]

DAYS:
[...]

PURPOSE:
[...]

LUGGAGE:
[...]

Categories:

DOCUMENTS
MONEY
PHONE/CHARGING
CLOTHES
MEDICINE
WORK
TRANSPORT
HOME BEFORE LEAVING

Do not generate 150 items.

Adapt the list to the actual trip.

These are my commitments today:

[LIST]

For each one, state:

DEADLINE
DURATION
IMPORTANCE
DEPENDENCY
ENERGY REQUIRED

Create a realistic schedule.

Include:

buffer;
breaks;
unexpected commitments.

Do not schedule 8 hours of focused work into 8 hours of available time.

Prioritize these tasks:

[TASK LIST]

For each, assess:

URGENCY
IMPACT
EFFORT
DEPENDENCY
CONSEQUENCE OF DELAY

Do not use only the Eisenhower matrix.

Find:

BLOCKERS
QUICK WINS
HIGH-LEVERAGE TASKS
DEADLINE CRITICAL

Create an execution order and explain the dependencies.

These are my commitments for the week:

[...]

Available time:

[...]

Create a plan that uses no more than:

[X%]

of the available time for things scheduled in advance.

Leave the rest as buffer.

Divide tasks into:

MUST COMPLETE
SHOULD COMPLETE
OPTIONAL

Identify tasks to delegate, automate, or postpone.

Review my plan:

[PLAN]

Look for:

- too many tasks;
- insufficient buffer;
- hidden dependencies;
- unrealistic time estimates;
- context switching.

Calculate how much time is actually planned.

If the plan is unrealistic:

reduce the scope instead of just shortening every activity.

I have this goal:

[GOAL]

DEADLINE:
[...]

Current state:
[...]

Break it into:

MILESTONES
DELIVERABLES
TASKS

Every task must have a clear definition of done.

Identify dependencies.

Find the critical path.

Do not create 100 microtasks if 15 to 20 well-defined tasks would provide better control.

I need to make a decision:

[DECISION]

OPTIONS:
[...]

MY CRITERIA:
[...]

First separate:

FACTS
PREFERENCES
UNCERTAINTIES

For each option, analyze:

BENEFITS
COSTS
RISKS
REVERSIBILITY
OPPORTUNITY COST

Do not reach a conclusion based on the number of pros and cons.

The importance of each item matters more than the count.

Analyze this decision:

[DECISION]

Assess:

REVERSIBILITY
COST OF REVERSAL
TIME TO LEARN
COST OF DELAY

If the decision is easy to reverse:

suggest a way to test it quickly.

If it is difficult to reverse:

identify which additional information would most reduce the risk before deciding.

Assume that this plan:

[PLAN]

has completely failed after:

[X months].

List the most likely reasons.

Do not use apocalyptic edge-case scenarios.

Focus on:

execution;
budget;
timing;
people;
technology;
assumptions;
dependencies.

For each, state:

EARLY WARNING
PREVENTION
CONTINGENCY

Analyze what happened.

GOAL:
[...]

PLAN:
[...]

ACTUAL OUTCOME:
[...]

Do not look for someone to blame.

Separate:

WHAT WAS KNOWN
WHAT WAS ASSUMED
WHAT CHANGED
WHAT FAILED

Find:

PROCESS ISSUE
DECISION ISSUE
EXECUTION ISSUE
EXTERNAL FACTOR

At the end, propose no more than 5 process changes.

I am preparing to negotiate:

[TOPIC]

MY GOAL:
[...]

THEIR LIKELY GOAL:
[...]

MY MINIMUM ACCEPTABLE OPTION:
[...]

ALTERNATIVE IF WE DO NOT AGREE:
[...]

Create:

MY PRIORITIES
THEIR POSSIBLE PRIORITIES
TRADEABLE ITEMS
NON-NEGOTIABLES
BATNA
QUESTIONS TO ASK

Do not focus the negotiation only on price.

Look for items that have different value to each side.

I am preparing for a salary negotiation.

POSITION:
[...]

CURRENT OFFER:
[...]

MY EXPERIENCE:
[...]

LOCATION:
[...]

COMPENSATION:
[...]

Help me define:

TARGET
ACCEPTABLE RANGE
NON-SALARY ITEMS

Possible items:

base;
bonus;
remote work;
vacation;
equipment;
education;
title;
review period.

Create a professional argument based on:

scope;
experience;
market data;
value.

Do not use personal financial needs as the main argument.

I want to negotiate the price of:

[PRODUCT/SERVICE]

ASKING PRICE:
[...]

MARKET PRICE:
[...]

PROBLEMS / LIMITATIONS:
[...]

Create an argument using:

comparable offers;
condition;
missing features;
payment terms;
volume;
timing.

Do not invent a competing offer.

Suggest an opening, target, and maximum price.

I want to progress toward:

[ROLE]

CURRENTLY:
[...]

EXPERIENCE:
[...]

SKILLS:
[...]

GOAL:
[...]

DEADLINE:
[...]

First create a gap analysis.

Separate:

ALREADY STRONG
NEEDS IMPROVEMENT
MISSING
NOT ACTUALLY REQUIRED

Then create a plan for:

skills;
projects;
portfolio;
interviews;
networking.

Do not create a learning plan covering 50 technologies if they are not needed for the target position.

Analyze this job posting:

[JOB DESCRIPTION]

Find:

MUST HAVE
LIKELY MUST HAVE
NICE TO HAVE
GENERIC HR LANGUAGE

Then compare it with my experience:

[CV / EXPERIENCE]

Create:

STRONG MATCH
PARTIAL MATCH
GAP

Do not treat every item in the posting as equally important.

Audit my CV:

[CV]

TARGET ROLE:
[...]

Focus on:

- clarity;
- relevance;
- measurable results;
- technical signal;
- seniority signal;
- unnecessary information.

For each bullet point, check:

ACTION
SCOPE
TECHNOLOGY
RESULT

Do not invent metrics.

If the result is unknown:

suggest a question I should ask myself to find a real result.

Rewrite this CV bullet:

[ORIGINAL]

Context:
[...]

Preserve the truth.

Structure:

ACTION
WHAT
HOW
RESULT

Do not add percentages, users, revenue, or performance numbers I have not provided.

Compare:

MY CV:
[...]

JOB:
[...]

For each important requirement, state:

REQUIREMENT
EVIDENCE IN CV
MATCH
MISSING SIGNAL

Then suggest CV changes that better highlight experience I actually have.

Do not change facts just for an ATS.

Optimize my CV for ATS for:

[JOB]

But do not use keyword stuffing.

Include relevant terms only where they match my real experience.

Keep the CV natural and readable for a person.

Identify key terms from the posting that are currently missing but truthfully apply to my experience.

Audit my portfolio for:

[TARGET ROLE]

URL/CONTENT:
[...]

For each project, check:

PROBLEM
WHAT I BUILT
MY ROLE
TECH STACK
DIFFICULT PART
OUTCOME
LIVE DEMO
CODE

Find projects that do not add a new signal.

Suggest an order for the projects based on relevance to the target role.

Audit this GitHub profile from an employer's perspective.

Target role:
[...]

Check:

pinned repositories;
README;
recent activity;
project completeness;
tests;
documentation;
CI;
issues;
code quality.

Do not judge only by the number of commits.

Identify what an employer could conclude after reviewing the profile for 2 minutes.

Create an application strategy for:

[TARGET ROLE]

LOCATION:
[...]

REMOTE:
[...]

EXPERIENCE:
[...]

Do not suggest sending 100 random applications per day.

Segment job postings into:

HIGH MATCH
STRETCH
LOW MATCH

Define:

applications/week;
tailored applications;
network outreach;
portfolio work;
interview preparation.

Track conversion between:

APPLICATION
SCREEN
INTERVIEW
OFFER

and adjust the strategy based on the funnel.

Write a cover letter for:

[JOB]

MY CV:
[...]

Do not retell my entire CV.

Focus on:

WHY THIS ROLE
MOST RELEVANT EXPERIENCE
SPECIFIC EVIDENCE
WHY THIS COMPANY

Do not invent enthusiasm for the product if there is no real reason.

Maximum:
[X words]

Write a short message to a recruiter.

ROLE:
[...]

MY SIGNAL:
[...]

GOAL:
[...]

The message must:

- quickly explain who I am;
- explain why I am relevant;
- include a clear call to action.

Avoid generic phrases such as:
"I hope this message finds you well."

Write a follow-up after an interview.

COMPANY:
[...]

ROLE:
[...]

INTERVIEW:
[...]

TOPICS:
[...]

Tone:
brief, professional, natural.

Mention one specific part of the conversation.

Do not sound desperate or overly formal.

This happened in the interview:

[NOTES]

Analyze:

WHAT WENT WELL
WEAK ANSWERS
QUESTIONS I MISUNDERSTOOD
SIGNALS FROM THE INTERVIEWER
FOLLOW-UP NEEDED

For each weak answer, write:

WHAT WAS MISSING
BETTER STRUCTURE

Do not change the content of my experience.

Analyze this offer:

BASE:
[...]

BONUS:
[...]

EQUITY:
[...]

BENEFITS:
[...]

VACATION:
[...]

REMOTE:
[...]

NOTICE:
[...]

OTHER:
[...]

Calculate total compensation where possible.

Separate:

GUARANTEED
VARIABLE
CONDITIONAL
UNKNOWN VALUE

Analyze conditions that may have substantial value or risk in particular.

Analyze this freelance job.

SCOPE:
[...]

RATE:
[...]

HOURS:
[...]

DEADLINE:
[...]

REVISIONS:
[...]

PAYMENT TERMS:
[...]

Calculate the realistic effective hourly rate, including:

communication;
meetings;
revisions;
admin;
tax overhead, if relevant.

Find scope creep risks.

Create questions to resolve before accepting.

Create a proposal structure for a client.

PROBLEM:
[...]

SCOPE:
[...]

DELIVERABLES:
[...]

TIMELINE:
[...]

PRICE:
[...]

Do not focus the proposal only on a list of technical tasks.

Structure:

CLIENT PROBLEM
PROPOSED OUTCOME
SCOPE
DELIVERABLES
TIMELINE
PRICE
ASSUMPTIONS
OUT OF SCOPE
NEXT STEP

I need to estimate this project:

[REQUIREMENTS]

Do not immediately give one number.

Separate:

KNOWN WORK
UNKNOWN WORK
DEPENDENCIES
RISKS

For each epic, provide:

BEST CASE
LIKELY
WORST REASONABLE

Explain which unknowns affect the estimate most.

Add a discovery phase if the requirements are too unclear.

Compare:

ORIGINAL SCOPE:
[...]

NEW REQUEST:
[...]

Identify:

IN SCOPE
AMBIGUOUS
OUT OF SCOPE

For out-of-scope items, estimate:

TIME
COST
DEPENDENCIES

Write a professional response to the client explaining the scope change without a confrontational tone.

Prepare me for a meeting.

TOPIC:
[...]

PARTICIPANTS:
[...]

GOAL:
[...]

Context:
[...]

Create:

OBJECTIVE
3 KEY POINTS
QUESTIONS
RISKS
DECISIONS NEEDED

Suggest the order of discussion.

Do not fill the meeting with information that does not need discussion.

Create an agenda.

GOAL:
[...]

DURATION:
[...]

PARTICIPANTS:
[...]

For each item, state:

TIME
OWNER
DESIRED OUTPUT

Separate:

INFORMATION
DISCUSSION
DECISION

If an item has no required outcome:

it probably does not belong in the meeting.

These are the meeting notes:

[NOTES]

Turn them into:

DECISIONS
ACTION ITEMS
OWNER
DEADLINE
OPEN QUESTIONS

Do not label every discussed idea as a decision.

If an owner or deadline is not stated:

UNKNOWN

Analyze this list of emails:

[...]

Sort them into:

ACTION REQUIRED
WAITING
REFERENCE
IGNORE

For ACTION REQUIRED, state:

TASK
DEADLINE
ESTIMATED EFFORT
PRIORITY

Do not treat every unread email as an obligation.

Help me respond to this message:

[MESSAGE]

CONTEXT:
[...]

My goal:
[...]

Do not escalate the conflict.

First identify:

FACTUAL ISSUE
EMOTIONAL TONE
REQUEST
MISUNDERSTANDING

Write a response that:

- acknowledges the facts that are correct;
- corrects the inaccurate points;
- clearly sets a boundary;
- suggests a next step.

Avoid passive-aggressive wording.

I need to communicate:

[BAD NEWS / REJECTION / DELAY]

To whom:
[...]

Reason:
[...]

Next step:
[...]

Write a message that is:

direct;
honest;
brief;
without unnecessary justification.

Structure:

DECISION/STATUS
BRIEF REASON
IMPACT
NEXT STEP

Write a polite rejection of:

[BUSINESS / JOB / SALES OFFER]

Reason:
[...]

I want to:
[stay in touch / close the matter]

Do not use a false phrase like:
"maybe in the future"

if I do not really mean it.

Write a message to negotiate a discount.

SERVICE/PRODUCT:
[...]

PRICE:
[...]

CONTEXT:
[...]

My argument:
[...]

The message should be brief and non-confrontational.

If there is no room for a lower price, suggest an alternative:

longer term;
upfront payment;
smaller scope;
bundle;
different plan.

Decision:

[...]

Unknowns:

[...]

Create a table:

UNKNOWN
HOW MUCH IT MATTERS
COST TO LEARN
TIME TO LEARN

Find the 1 to 3 pieces of information with the greatest expected value of information.

Do not research everything.

Focus on information that could realistically change the decision.

Analyze this decision:

[DECISION]

Do not look only at the direct consequence.

Create:

FIRST ORDER EFFECT
SECOND ORDER EFFECT
POSSIBLE THIRD ORDER EFFECT

Analyze these in particular:

incentives;
behavior change;
maintenance;
future constraints;
lock-in.

Do not invent exotic consequences without a realistic mechanism.

If I choose:

[OPTION]

what will I be unable to do because of the:

TIME
MONEY
ATTENTION
COMMITMENT

Compare it with the best realistic alternative.

Do not compare it with an idealized fantasy option.

Help me organize notes about:

[TOPIC]

I do not want a complicated system.

Suggest a minimal structure:

INBOX
ACTIVE
REFERENCE
ARCHIVE

For each new note, determine:

PROJECT?
ACTION?
REFERENCE?
DELETE?

Goal:
to find information later, not to spend time organizing the organization.

These are my notes for today:

[NOTES]

Extract:

DONE
DECISIONS
OPEN TASKS
FOLLOW-UPS
IDEAS

Do not turn every thought into a task.

If there is no clear action:

put it in NOTES.

This was my week:

[CALENDAR / TASKS / NOTES]

Analyze:

WHAT I PLANNED
WHAT I COMPLETED
WHAT SLIPPED
WHY

Find patterns:

underestimation;
interruptions;
overcommitment;
low-value work.

Suggest no more than 3 changes for next week.

Do not create a completely new productivity system every week.

I am putting off:

[TASK]

Do not give me motivational quotes.

Try to determine whether the problem is:

UNCLEAR NEXT STEP
TASK TOO LARGE
FEAR OF ERROR
BORING
DEPENDENCY
LOW VALUE
LOW ENERGY

For the most likely reason, suggest the smallest next physical action that takes less than 15 minutes.

I have:

[X minutes]

to work on:

[TASK]

Define only one goal for the session.

Create:

START CONDITION
WORK TARGET
STOP CONDITION

Remove activities that are not needed to complete the goal.

The session must produce a concrete artifact or result.

I need to learn this quickly:

[TOOL]

GOAL:
[...]

I do not need a complete course.

Identify the 20% of features that cover 80% of my use case.

Create a plan:

DAY 1
DAY 2
DAY 3
...

Each day must end with a practical result.

Do not include advanced features until I have mastered the basic workflow.

I am switching from:

[TOOL A]

to:

[TOOL B]

My current use:

[...]

Map:

A FEATURE
→
B EQUIVALENT
→
DIFFERENCE

Identify:

MISSING FEATURES
NEW FEATURES
DATA MIGRATION
WORKFLOW CHANGES
TRAINING REQUIRED

Do not assume the same workflow should be copied one to one.

Turn this process:

[PROCESS]

into an SOP.

Structure:

PURPOSE
SCOPE
PREREQUISITES
RESPONSIBILITIES
PROCEDURE
EXPECTED RESULT
ERROR HANDLING
ESCALATION
AUDIT EVIDENCE

Every step must be a concrete action.

Avoid:
"check that everything is okay"

without defining what "okay" means.

Audit this SOP:

[PROCEDURE]

Check for:

AMBIGUOUS STEPS
MISSING OWNER
MISSING INPUT
MISSING OUTPUT
NO FAILURE PATH
NO VERIFICATION
OUTDATED STEP

Suggest the smallest correction for each issue.

Turn this into a short operational checklist:

[PROCESS]

The checklist should not explain the theory.

Each item must be:

ACTIONABLE
BINARY
VERIFIABLE

Example:

Bad:
"Check security."

Good:
"Confirm that MFA is enabled for administrator accounts."

Problem:

[PROBLEM]

Do not stop at the first obvious cause.

Create:

SYMPTOM
IMMEDIATE CAUSE
CONTRIBUTING FACTORS
ROOT CAUSE

Use:

5 WHYS

only as long as the answers have a genuine cause-and-effect relationship.

At the end, separate:

FIX SYMPTOM
FIX ROOT CAUSE
PREVENT RECURRENCE

I have these problems:

[LIST/DATA]

Find the small number of causes that explain the largest share of consequences.

If numeric data are available:

calculate the cumulative contribution.

If not:

do not pretend to calculate 80/20.

Create a qualitative prioritization and state its limitations clearly.

Main plan:

[...]

Identify key failure points.

For each, state:

TRIGGER
PLAN B
DECISION DEADLINE

Do not create a backup plan for every small problem.

Focus only on failure modes that could stop the entire goal.

This is my plan:

[PLAN]

Act as an independent reviewer.

Do not just confirm that the plan looks good.

Look for:

missing dependency;
hidden cost;
legal issue;
security issue;
resource bottleneck;
bad assumption;
edge case;
maintenance burden.

For each finding, explain how realistically it could change the outcome.

For:

[PLAN]

create:

BEST CASE
BASE CASE
WORST REASONABLE CASE

Do not use extreme disasters with very low probability.

The worst reasonable case should be a negative scenario realistic enough to plan for.

Now ignore any impulse to defend your previous answer.

Audit it as a new reviewer.

Look for:

incorrect facts;
missing context;
unsupported assumptions;
outdated data;
math errors;
logical gaps;
overconfidence.

For each item, state:

ORIGINAL CLAIM
STATUS
CORRECTION

Then create an improved final version.

Do not assume that my idea, claim, or plan is correct.

Treat it as a hypothesis.

Look for:

EVIDENCE FOR
EVIDENCE AGAINST
ALTERNATIVE EXPLANATIONS
MISSING INFORMATION

If I am wrong:

say specifically where and why.

If I am right:

explain which evidence supports it.

The goal is accuracy, not confirming my position.

For this task:

do not fill gaps with a plausible answer.

If a fact is unknown:

UNKNOWN

If it is only an assumption:

ASSUMPTION

If it is derived from other data:

INFERENCE

If it is confirmed:

VERIFIED

Do not mix these categories in the final answer.

For every number:

1. identify the source;
2. check the unit;
3. check the period;
4. check the denominator;
5. recalculate where possible.

Check these in particular:

percent
percentage point
currency
inflation adjustment
per capita
annualized values.

Do not copy a number from a source if you do not understand what it measures.