SOURCE ID: S11
ORIGINAL PATH: C:/Users/thisi/Desktop/PK-squads/submission-polish/workflow/knowledge-register/audits/2026-09-05-memory-reader/REVIEW.md
FULL SOURCE SHA256: 177ae5b0e671e405ebf720fee3f49a48ee40042ca450250f1d753ed2454663c5
CONTEXT: The excerpt supports the introductory-retraction defect and repair. It retains receipt-custody and coverage limits; unrelated retrieval counts are not exported as an effectiveness claim. This later reader did not produce the earlier manual audit recovery.
EXTRACTION: Original decoded lines, preserved without textual edits; UTF-8 with LF newlines. Gaps are explicit.

--- Original source lines 3-7 ---
The new entry point connects existing experiment reports and compact tried-result rows
to their original method, result and caveats. Eight explanatory records add question,
why, lesson, reopening condition and explicit limits. It reads Git objects from one
resolved revision and leaves source documents in place. It is linked from START-HERE
and the master index.

--- Original source lines 20-23 ---
- Independent code review reproduced an introductory retraction disappearing from both
  `show` and `search` while source verification passed. The reader now retains/searches
  the preamble, with dedicated regression cases. Source integrity alone did not catch
  this semantic extraction defect.

--- Original source lines 36-57 ---
The original shared scratch receipt path was overwritten after the first read by an
unidentified writer. The consumer denied authorship of that replacement and preserved
its original from retained state at a unique path. `consumer.json` is that reconstruction
(original and trailing-newline-normalized banked SHA-256 values in assessment.json);
no pre-overwrite hash was captured. This weakens session
custody and timing claims. It does not replace the source-locator checks or justify using
the unrelated replacement's metrics. Future acceptance receipts need unique paths and
an immediate content hash at return.

## Remaining gaps

This is not full-goal completion. The master-index census names 54 existing surfaces,
all present at the selected revision, with explicit reader dispositions. Most are still
unintegrated and semantically unreviewed. Important attempt history, retractions as
individual records, math, tools, datasets/models, other branches/hosts and transcripts
need coverage. The first consumer followed README's known chains; it is not a held-out
discovery benchmark and there is no measured comparison with prior onboarding.

Next: reconcile these source families and recover the high-value items they contain,
then extend acceptance to unseen questions and session turnover. Compact structured
retraction lookup is a concrete context opportunity: the first consumer opened the
whole register. Preserve qualifiers and contradictions when adding that view.
