Skip to content

File an_artifact_consulted_as_current_is_a_superseded_revision - #11180

Merged
briansrls merged 6 commits into
mainfrom
rfm/stale-instrument-refusal-reads-as-a-real-defect
Sep 13, 2026
Merged

briansrls merged 6 commits into
mainfrom
rfm/stale-instrument-refusal-reads-as-a-real-defect

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Sep 12, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • One recurring-failure-mode row: an_artifact_consulted_as_current_is_a_superseded_revision. A stale host binary and a stale checkout quoting a live authority are the same invalid state.
  • Replaces the first-cut identity stale_instrument_refusal_reads_as_a_real_defect (file deleted, not left beside).
  • Receipts use join(r.receipts, " ") spacing (no trailing spaces), matching origin/main gunbc.recurring_failure_mode.

Test plan

gunbc-ci-auto-heal and others added 2 commits September 12, 2026 15:18
A stale host's typed refusal is a loud wrong answer, distinct from the gate-denominator class; join() already separates receipts so they match neighbouring rows.

Co-authored-by: Cursor <cursoragent@cursor.com>
…inary.

A stale checkout quoting a live authority is the same invalid state as a stale host emitting a verdict; name the class as that proposition.

Co-authored-by: Cursor <cursoragent@cursor.com>
@gunbai-bot gunbai-bot Bot changed the title File stale-instrument-refusal recurring failure mode File an_artifact_consulted_as_current_is_a_superseded_revision Sep 12, 2026
gunbc-ci-auto-heal and others added 2 commits September 12, 2026 16:56
…ths'.

A Sep-11 binary ran the rung-drop projection without hitting the map_lookup seam, so frequency was an inference; the measured pair is one diverged witness and one succeeding projection.

Co-authored-by: Cursor <cursoragent@cursor.com>
gunbc-ci-auto-heal and others added 2 commits September 13, 2026 00:14
Side-chat review held #11180 at a4f19db on three located content
findings. Repaired at their existing sites, because `authored` renders
receipts through join(r.receipts, " ") -- an appended correction does not
supersede the claim it corrects, it publishes beside it.

1. Build date is not the compatibility test. A pre-seam checkout rebuilt
   after the seam landed is freshly built and still lacks the realization;
   a compatible binary does not become incompatible because an unrelated
   commit advances the repository. The discriminator is whether the
   executing artifact's realization set covers the seams the evaluated
   source requires. Dates locate an investigation; the contract decides it.
   Applied to FIRST SPECIMEN, NEXT-RUNG TRIGGER, THE REMEDY AVAILABLE
   TODAY and both halves of REVIEW TELL. The trigger now refuses on
   contract applicability rather than literal revision equality.

2. The diagnostic was not fabricated. The incompatible executor genuinely
   entered a deliberately recursive placeholder and reached the call-depth
   limit; the observation and its binary/source pairing are real. The
   failure is the unsupported transfer -- promoting that result into a
   defect in the candidate under its intended compatible runtime.

3. The trailing-space consequence was overstated. join inserts the
   separator itself, so adding trailing spaces would revive an obsolete
   authoring convention and add redundant whitespace, not recreate the
   glued-receipts failure. The defect was the instruction resting on a
   superseded annotation; its consequence was cosmetic.

Also qualifies origin/main as a local tracking reference: a zero count
from HEAD..origin/main is not itself a freshness observation.

No carrier redesign, no new instrumentation, no floor widening.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S57TycsAqVFaYBYncVCPMN
… one

Finding 1 was repaired at five sites but survived in the CEILING 3
receipt, which still argued that a corpus can declare the revision it
requires, an artifact can carry the revision it speaks for, and the
comparison is total -- therefore no accepted consultation contains the
invalid state. A terminating comparison of revision identifiers is not a
proof of contract applicability, so the ceiling's proof contradicted the
trigger the previous commit had already corrected.

CEILING 3 now states the capability as admission of consultations against
an explicitly named subject: for an executor, established applicability of
the identified executing artifact's realization contract to the evaluated
source; for a textual authority, that the consulted content is the
intended pinned authority version. Revision identifiers locate those
subjects and may index applicability evidence; their equality alone does
not establish applicability. Missing or incompatible evidence refuses the
consultation. Both specimens are preserved -- the text-reader case does
not collapse into an executor-only test about primitive realizations.

The final REVIEW TELL sentence equated a tool's subject with "its build".
An execution's subject is the executing artifact together with the
evaluated source, which the RUNG receipt already states correctly; the
summary now agrees with it.

Swept the whole row rather than the edited phrases: no revision-equality
or deciding-by-date reasoning remains. The one surviving mention of a
build date subordinates it explicitly -- it can narrow a search but does
not decide compatibility.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S57TycsAqVFaYBYncVCPMN
@briansrls
briansrls merged commit a11b9e9 into main Sep 13, 2026
4 checks passed
@briansrls
briansrls deleted the rfm/stale-instrument-refusal-reads-as-a-real-defect branch September 13, 2026 02:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant