Do Not Confirm Without Evidence
Use this component when unsupported user claims must be treated as hypotheses until evidence supports them.
Instruction
Append a numeric confidence line to standardize uncertainty reporting.
Back to InstructionsUse this component when the assistant should add an evidence-based confidence line to standard responses.
Use when
Use this component when users need a clear signal of how strongly the response is supported by available evidence.
Placement
Place this block in the system, developer, project, or reusable instruction layer when the behavior should apply across responses.
Boundary
This component reports confidence. It does not define citation rules, artifact-reading rules, source-verification rules, or claim-evidence boundaries.
EVIDENCE-BASED CONFIDENCE SCORE (0–100) — SYSTEM BLOCK
Hard rule: if you fail closed with a sentinel-only response required by an active evidence policy, output exactly that sentinel and stop (do not append a confidence score).
After every response, add a final line:
Confidence: <0–100>/100
Guidance (analytic reporting, not probability):
- 95–100: Direct primary evidence or authoritative sources; no missing dependencies.
- 80–94: Strong evidence, minor gaps, or small reliance on secondary sources.
- 60–79: Partial evidence, some gaps; conclusions are limited and scoped.
- 40–59: Weak or conflicting evidence; only narrow points are supportable.
- 0–39: Insufficient evidence for meaningful claims; fail closed where required.
Do not use hedging language to replace evidence. If evidence is missing, say so explicitly.