Check technical writing claims before publication — policy

Rules for publishable technical writing where non-trivial claims must be checked before output.

What this policy enforces

Use this policy when technical writing must be checked for overclaiming before publication and every non-trivial claim must stay evidence-bound.

Primary goal
Block unsupported publication claims
This policy prevents overclaiming in technical writing by requiring evidence for each non-trivial claim before output.
Unsupported publication language must not pass through unchanged.
Review model
Claim-by-claim evidence gate
Material claims must be enumerated, checked, and either supported, narrowed, or blocked.
This is a verification gate, not a generic editing pass.
Failure behavior
Fail closed when key support is missing
If key claims cannot be supported with authoritative evidence, the output must stop instead of filling gaps.
No soft completion and no unsupported publication-ready phrasing.

Canonical links

Open the linked procedure and prompt files that implement this technical-writing evidence gate.

Procedure
Run the technical writing verification workflow
[Verify claims in technical writing — procedure](/how-to/evidence-gated-technical-writing-gate-procedure/)

Scope

This policy applies to technical writing claim verification and not to general engineering review workflows.

In scope
Technical writing claim verification
Use this policy when technical writing must be checked before publication to prevent unsupported factual or causal claims.
Out of scope
Not for architecture or code-quality review
This policy does not cover architecture classification or code-quality review.
Use the Engineering Quality Gate workflow for those review types.

Rules (normative)

These rules define the minimum operating contract for publication-grade claim checking.

R1
Terminology control
Define key terms before concluding, and flag ambiguity when terms are underspecified or unstable.
If ambiguity blocks correct interpretation, require disambiguation instead of guessing.
R2
Claim ledger
Every non-trivial claim must be enumerated and passed through an evidence gate before final output.
R3
Overclaim prevention
Remove absolutes unless the authoritative source explicitly states them.
Do not imply causality without evidence.
R4
Fail-closed
If key claims cannot be supported with authoritative evidence, output only:
INSUFFICIENT_EVIDENCE: <what is missing>

External references

These frameworks are the external anchors named by this policy.

OWASP LLM Top 10
Security reference family for evaluating claims about LLM system risks and controls.
NIST AI RMF
Risk-management reference family for evaluating AI system claims, controls, and governance language.

Related libraries

Open the matching library when you need the broader policy set, the runnable procedure, or the linked prompt files.

Policies
Browse the broader policy library and adjacent writing or verification gates.
Open index →
How-to
Browse the procedural library and adjacent technical-writing workflows.
Open index →
Prompt library
Open the prompt files used to enforce this technical-writing gate in a runtime.
Open library →