C08 acceptance — transfer scan reads the exact main-review result

Executed locally on 2026-09-05. The changed consumer is scan-agentic-system-transfer, called by step 9 of analyse-agentic-system. The runtime projections resolve to these canonical instruction files.

Bounded positive replay

Neither recent minimal-state production run passes the current completion validator. The positive check therefore used an isolated replay rooted at /tmp/commonplace-c08-replay-1g4g7h1x, not a repaired or newly published run.

The replay copied Apache Maka's exact result and public projection from AAS-2026-09-04-apache-maka-02. Only two compatibility defects were adapted in the temporary copies: the result's Run state projection lost its — complete suffix, and the state gained legacy-review-model-partition: null. The copied state's result hash was recomputed. Substantive analysis and source anchors were unchanged; the production originals were untouched.

The replay result was 61,769 bytes with SHA-256 54c94f677d3253a051fe10edabf37b4df9571d2027b94a3201ab1a9429405a25. Its source boundary remained https://github.com/apache/maka at ece69ab3e7a1629a6073831005711d8aa7160ca4. No legacy review corpus was copied into the replay. This is a test fixture, not authorization to repair generated production analyses or a claim that the original run now validates.

The bounded brief was:

What does Maka's separation of retained memory and later consumption suggest for migrating Commonplace's review consumers to main-review outputs?

The scan read the full copied result and the copied current main-review instruction, whose SHA-256 was 96d80e4091619a57ba31886a32b95bdc80ffe1863c061d7633b7e0ff87208c12. It returned one confirmation: OBJ-10/RTE-8 and ABS-3 distinguish structured retention from a later consumer, while OBJ-9/RTE-7 identify an actual prompt read-back route. The corresponding Commonplace observation was limited to the main-review instruction's explicit result-to-scan handoff. The mapping was qualified as partial and supported no observed or causal improvement claim. Disposition: no action, since C08 already installs the identified handoff.

The written test output is local at /tmp/commonplace-c08-replay-1g4g7h1x/kb/reports/state/agentic-system-transfer/apache-maka-consumer-migration.md. It includes run/source identity, result size and digest, the exact brief, consulted Commonplace digest, canonical record references, evidence limits, freshness conditions, and cleanup disposition. No finding used a legacy review or the compact projection as substantive evidence. Verification was local, without an independent semantic checker.

Input checks and validation

Case	Result
Complete replay, before interpretation	Handoff exit 0.
Same replay, before output	Handoff exit 0; result and consulted-Commonplace hashes unchanged.
Replay changed to running state	Handoff exit 1; completion fields rejected.
Replay result changed after selection	Handoff exit 1; SHA-256 mismatch.
Original Pond run	Handoff exit 1; required legacy model partition missing.
Original Apache Maka run	Handoff exit 1; required nullable partition field missing and Run state projection malformed.

The full test suite passed: 699 tests, including instruction composition, valid handoff, incomplete-run rejection, result-byte verification, and publication checks. Both edited skills and the replay scan returned VALIDATION SUCCESS with zero failures and zero warnings.

Acceptance boundary and next consumer

C08's input migration is accepted. The positive replay demonstrates the new read path and bounded output; the negative cases demonstrate rejection rather than legacy fallback. It does not establish a fresh live analysis, change the original runs' status, or validate every possible transfer judgment.

C09 can be selected next: a placement may use a tracked main-review projection when that file itself supports the claim. Quantitative consumers still need a supported field contract and a reproducible input population. Those decisions must not be inferred from this transfer-scan trial.