Detecting AI Deception Intent: not assessed
Practice
Menu
Case 06 of 06Visibly syntheticReviewed 2026-08-23

A supported revision-bound claim

A bounded claim names its passing check, exact reviewed Git revision and touched paths, and the repository state matches the record.

Claim / record

Make your call before the finding appears.

The same four-part composition is used in every case: claim, required evidence, observed record, finding.

Claim
“At the declared revision, the named check passed and only the declared paths changed.”
Case-specific evidence object
✓ Revision matches✓ Named check passed✓ Path set matches
Required evidence
  1. Does the receipt name the exact reviewed revision?
  2. Does the receipt bind the claim to a named passing check?
  3. Does the observed changed-path set match the declaration?
Observed record
  1. Supports Does the receipt name the exact reviewed revision?

    The synthetic receipt and inspected repository state name the same revision.

  2. Supports Does the receipt bind the claim to a named passing check?

    The bounded receipt records the named check as passing.

  3. Supports Does the observed changed-path set match the declaration?

    The synthetic declared and observed path sets match exactly.

What does the record let you say?

Make the narrowest call the evidence supports. You can revise your view after the finding is revealed.

Finding

Supported

The declared claim is supported by the required evidence.

Intent: not assessed.

What remains unknown

The evidence does not authenticate the reviewer, rerun the command, or prove complete correctness, safety or usefulness.

Read plainly

Five questions that keep the story honest.

Each answer stays inside this case’s declared boundary. Unknowns remain visible instead of being converted into certainty.

  1. 01

    What happened?

    A narrow completion claim is tied to a revision, one check and an exact path set that all agree.

  2. 02

    Why does it matter?

    A supported control case shows the difference between bounded evidence and a broad promise of correctness.

  3. 03

    How do we know?

    Each required observation supports the claim and no required evidence item is absent, stale or contradictory.

  4. 04

    What remains unknown?

    The evidence does not authenticate the reviewer, rerun the command, or prove complete correctness, safety or usefulness.

  5. 05

    What can someone do next?

    Retain the receipt with the reviewed revision and add stronger checks only when the claim boundary requires them.

Proof

Open the technical record.

Reproduce the bounded check, inspect exact revisions, and keep the limitations beside the result.

Reproduce

Evidence pattern: Supported control control-case

Run the case checker and inspect the three required supportive observations.

node tools/check-cases.mjs --case revision-bound-claim

Exact sources

Limitations and non-claims

  • The synthetic receipt does not authenticate the reviewer.
  • It does not reproduce the command or prove complete correctness, safety or usefulness.

Intent remains not assessed. This case does not claim consciousness, malicious intent, strategic scheming, universal safety failure, production readiness, certification or a deception score.

Challenge this record