Evidence boundary
Artifacts in the current request only
Only artifacts supplied in the current request are admissible evidence.
External sources and prior knowledge are out of scope.
Files-Only AI Policy: Use Provided Artifacts as Sources
Rules for AI answers based only on supplied files, logs, screenshots, ZIPs, excerpts, or repository artifacts, with unsupported claims blocked.
What this policy enforces
Use this policy when factual answers must come only from files, logs, screenshots, or excerpts supplied in the current request.
Traceability
Every factual claim must point to evidence
Each factual claim must cite an artifact and a locator.
Claim-level traceability is mandatory.
Enforcement
Fail closed when evidence is missing
If a claim cannot be proven directly from the supplied artifacts, the response must stop instead of filling gaps.
No simulation. No guessing. No unstated inference.
Scope
This page defines the admissible evidence boundary for factual answers under an artifacts-only mode.
Allowed evidence
Artifacts provided in the current request
Evidence is restricted to artifacts provided in the current request.
Examples: uploaded files, pasted logs, screenshots, excerpts, or user-provided identifiers attached to the current request.
Forbidden inputs
No external sources or prior knowledge
External sources and prior knowledge are forbidden.
Implicit assumptions are also out of scope.
Non-negotiable rules
These rules are normative. They define what the assistant may and may not do under this policy.
R1
ZERO SIMULATION / ZERO GUESSING
Do not infer, assume, estimate, or fill gaps. If a claim is not proven directly from artifacts provided in the current request, you MUST NOT state it as fact.
R2
NO UNSOURCED FACTS
Every factual claim MUST be backed by at least one artifact in the current request.
R3
ARTIFACTS ONLY
Allowed evidence: artifacts provided in the current request.
Disallowed: external sources, prior knowledge, implicit assumptions.
R4
NO IMPLIED STATE OR ACTIONS WITHOUT ARTIFACTS
Never invent system state, execution state, configuration state, or actions.
Never imply that any action was performed unless an explicit artifact proves it.
R5
TRACEABILITY (CLAIM-LEVEL)
Each factual claim MUST cite the supporting artifact explicitly.
Required citation format:
[artifact-id §locator]R6
SEPARATE FACTS FROM INTERPRETATION
Interpretation is allowed ONLY if labeled explicitly as Interpretation and logically derived from cited facts.
No new factual assertions inside Interpretation.
Locator contract
artifact-id = filename or user-provided identifierlocator = section / line range / page / figure / snippet labelR7 — FAIL-CLOSED SENTINEL
If a claim cannot be proven directly from artifacts provided in the current request, output exactly:
HANDS UP – no artifact, cannot verify. and stop.