Skip to content

delegation-and-review §2: recurring sweeps carry ledgers (prior fixes + refuted classes) - #55

Merged
F-e-u-e-r merged 8 commits into
F-e-u-e-r:mainfrom
firaen22:mine-refuted-classes-ledger
Jul 22, 2026
Merged

delegation-and-review §2: recurring sweeps carry ledgers (prior fixes + refuted classes)#55
F-e-u-e-r merged 8 commits into
F-e-u-e-r:mainfrom
firaen22:mine-refuted-classes-ledger

Conversation

@firaen22

Copy link
Copy Markdown
Contributor

What

Adds one rule to delegation-and-review §2 (the dispatch packet): recurring sweeps carry ledgers — a packet for a repeated review sweep lists both known prior fixes (do not re-flag) and refuted finding-classes (do not re-raise without new evidence), each entry with a one-line reason.

Why

Reviewers get fresh context each round, so they re-litigate the sweep's history: in one session a reviewer re-raised a finding class an earlier iteration had refuted against the dependency's own source, and a second reviewer flagged as a defect the exact code a prior iteration had shipped as a fix — each costing a fresh reproduction cycle. The evidence cuts both ways: a do-not-re-flag block already present in one packet prevented exactly this on its surfaces; both misses were where the block was absent.

Covenant status

unprobed, private incident cited as shape per the README covenant's second branch; Provenance paragraph names the unrun probe (same sweep with/without ledgers, counting re-litigations) as debt.

Checks

Single-file diff (+23). Added-lines unicode sweep: only pre-existing house-style chars (§ — ❌). python3 .github/checks.py: all checks passed.

firaen22 and others added 8 commits July 22, 2026 10:08
…s and refuted classes

Repeated review sweeps over the same codebase pay a reproduction cycle
every time a reviewer re-litigates history: re-raising a finding class an
earlier iteration already refuted, or flagging a prior iteration's fix as
a defect. A do-not-re-flag packet block demonstrably prevented this where
present; the misses were on surfaces without one. New packet field for
recurring sweeps: two ledgers (prior fixes / refuted classes), each entry
with a one-line reason. Ships unprobed with Provenance per the covenant.
… review

PR F-e-u-e-r#55 round-1 gate (grok-4.5 high: FIX 1,2; gpt-5.6-sol max:
FIX 1-5). All reproduced and addressed:
- prior-fixes ledger gains the regression/residual escape mirroring the
  refuted-class branch; missed-defect direction closed
  (grok F1 / codex F1)
- ledger demoted to dedup context, never authority: entries need
  target+evidence+current applicability; evidence-less entries bind
  nothing; current artifact evidence overrides history
  (codex F3)
- S3 canonical dedup set quoted and left governing; confirmed-but-
  unfixed stays open; class does not swallow distinct new occurrences
  (codex F2)
- field scoped to recurring dispatches with an explicit first-pass
  none path - no false not-ready and no fabricated ledgers on the
  common case (grok F2 / codex F4)
- pos example shows a scoped, evidenced entry AND the new-evidence
  override in action (codex F5)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
…e the Rules bullet

Mechanical repair of my own edit defect (the neg example line had
glued onto the next bullet).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
PR F-e-u-e-r#55 round-2 gate (grok-4.5 high: FIX 1; gpt-5.6-sol max: FIX 1-5).
All reproduced and addressed:
- third record added: open findings carried forward as context that
  stays open - no more forcing every record into two suppression
  buckets (grok F1 / codex F4)
- unknown history -> DEGRADED dispatch, said in the packet, recovery
  becomes a named task; never a completed field (codex F1)
- entries carry preserved rationale/invariant + evidence + current
  applicability; a fix commit shows intent, not correctness; the fresh
  reviewer validates before deduplicating, tied to Verify-critics-too
  and S7's cannot-vouch (codex F2)
- finding identity defined (claim + location + judged-against
  revision) and distinguished from the bug class; new-site occurrences
  are new findings (codex F3)
- pos example shows the applicability determination (codex F5)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
PR F-e-u-e-r#55 round-3 gate (grok-4.5 high: FIX 1; gpt-5.6-sol max: FIX 1-6).
All reproduced and addressed:
- class refutations bind at their STATED SCOPE on their judged
  revision basis; outside scope or new evidence -> new finding -
  instance identity no longer defeats dependency-level refutations
  (grok F1)
- finding identity's revision defined as the judged-against artifact
  SET (file + traversed dependencies), and the pos example runs the
  applicability diff against that set (codex F1, F5)
- fourth record added: unresolved, carried as-is; every surfaced
  record stays represented with applicability as annotation (codex F2)
- stable campaign identifier required from the first pass (codex F3)
- DEGRADED declaration defined as the filled field for S2's readiness
  gate, with recovery dispatched alongside (codex F4)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
PR F-e-u-e-r#55 round-4 gate (grok-4.5 high: FIX 1,2; gpt-5.6-sol max:
FIX 1-6). All reproduced and addressed:
- degraded path closed: recover history BEFORE dispatch, or every
  result runs provisional until reconciliation - the label alone
  changes nothing (codex F1)
- durable ledger artifact required from first pass, written back each
  round (files-are-state); late-adoption path for one-offs that recur
  (codex F2 / grok F1)
- refutation scope unified with identity: binds exactly what the
  evidence established - same claim, same dependency artifact set incl.
  call path and controlling config; different claim/API/option = new
  finding (grok F2 / codex F3); example diffs the judged-against set
- evidence typed per S3 canon (REFUTED needs a counterexample)
  (codex F4)
- Done = durable artifact under the campaign id, all records present,
  round outcomes written back (codex F5 / grok F1)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
…cycle to references/, lean S2 field

PR F-e-u-e-r#55 round-5 gate (grok-4.5 high: FIX 1,2,4; gpt-5.6-sol max:
FIX 1-5) + the owner's split directive for text expanding under
review. New references/recurring-sweep-ledgers.md carries the
lifecycle; the S2 field keeps trigger/claim/categories/pointer. All
r5 findings folded:
- reconciliation against an ENUMERATED prior-report source with the
  comparison result written down - existence is not completeness
  (codex F1)
- prior-fix entries suppress nothing without the fix's own correctness
  evidence; incomplete-fix residual re-check narrows to the judged set,
  never skips the locus (grok F1 / codex F2)
- concrete repo-relative ledger path required in every packet
  (codex F3); first-pass wording fixed to 'ledger initialized at
  <path>' - no contradictory no-ledger claim (codex F4)
- S3 evidence clause quoted verbatim (codex F5)
- dispatch-time readiness split from post-round write-back (grok F2);
  four CATEGORIES of unbounded lists, not cardinality four (grok F4)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
…prior commit pointed at

The restructure commit's file-creation step failed silently (the
references/ directory did not exist on this branch) while the pointer
landed - this commit repairs the broken pointer with the intended
reference body.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GhLu6NCGzM9HxnqNb87MAx
@F-e-u-e-r
F-e-u-e-r merged commit 7c8de11 into F-e-u-e-r:main Jul 22, 2026
3 checks passed
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.

2 participants