Observed
PR #27's anti-constant guard varies the bytes on a fixed path:
complete_success_case stale-basis-actual 81 'changed after request' stale.txt 65536
complete_success_case stale-basis-other 82 'a wholly different content' stale.txt 65536
test "$stale_other_basis" != "$stale_actual_basis"
bounded_workspace_observation_basis_v1 takes (path, bytes) pairs. Nothing in the observation witness varies the path while holding the bytes constant. A regression that dropped the path from the derivation — hashing only content — would leave the entire suite green.
Why it matters
The witness claims the basis describes what was read. "What was read" is a (path, bytes) pair, not a byte string. Half the input is unpinned.
The patch witness is not obviously better here, so this may be a shared gap rather than an observation-specific one; that is worth checking before fixing only one side.
Scope note
This was identified while writing #27 and deliberately not added there. The PR's claim was already scoped to "varies with the observed bytes on a fixed path", so there was no gap between claim and evidence to repair — only a strengthening available. Filing it as the strengthening it is.
Acceptance
- A successful observation of identical bytes at a different path derives a different basis.
- The same check exists for the patch witness, or an explicit note records why it does not apply there.
Observed
PR #27's anti-constant guard varies the bytes on a fixed path:
bounded_workspace_observation_basis_v1takes(path, bytes)pairs. Nothing in the observation witness varies the path while holding the bytes constant. A regression that dropped the path from the derivation — hashing only content — would leave the entire suite green.Why it matters
The witness claims the basis describes what was read. "What was read" is a
(path, bytes)pair, not a byte string. Half the input is unpinned.The patch witness is not obviously better here, so this may be a shared gap rather than an observation-specific one; that is worth checking before fixing only one side.
Scope note
This was identified while writing #27 and deliberately not added there. The PR's claim was already scoped to "varies with the observed bytes on a fixed path", so there was no gap between claim and evidence to repair — only a strengthening available. Filing it as the strengthening it is.
Acceptance