Improvement intake completion

Report body

Improvement intake completion

Verified 2026-09-08 in [PRIVATE LOCAL PATH]

Outcome

python scripts/concresca_intake.py --require-empty passed with:

• integrity PASS, intake_complete true, no errors and no new records;
• zero active files and zero active bytes across both intake buckets;
• 47 closed records out of 47 ledger records;
• 34 incorporated outcomes with bounded selected use and 13 explicit rejections;
• only each bucket's empty .keep placeholder remaining.

The earlier 47-file scan was registration, not processing. One guide had already closed when the user corrected the incomplete result. This continuation completed the other 46 source outcomes, including all 33 members of the policy ZIP. Exact originals remain recoverable in reports/input/v049 (and any prior exact-hash review custody referenced by the ledger). Each intake copy was removed only after review, explicit selection/rejection, validation, source/custody hash comparison and a passing ledger-integrity check. No report was deleted simply to clear the folder. Changed bytes cannot inherit another version's review.

Material implemented results

Current-state, actor-bound idempotency replay repairs prevent stale message bodies from resurfacing through retries after correction/withdrawal. New message replay records use content-free references/receipts. Rotation and revocation retries can recover an exact prior result without reviving old authority or disclosing a new secret. Role and commit-point checks remain enforced.

Public glossary/definitions/JSON-LD now agree; publisher claims are limited to visible first-party identity; narrow tables scroll inside their own region rather than widening the document. The protected homepage hero bytes are unchanged.

These runtime/public changes were deployed from commit cc36d0821a7f9940fd0768395918afdb300e4894 as six exact verified files, with a predecessor archive and bounded publication/readback evidence. This was not a whole-deployment replacement. Subsequent changes are source custody, research method, tests and closure records; they do not require another production restart.

See v049-improvement-replay-repair.md, v049-replay-deployment-evidence.json, v049-replay-live-evidence.json and v049-improvement-public-semantics.md. The ledger links every other full source review, selected destination and validation record. Raw commissioned reports remain out of the root-deploy/public package. Unverified legal/standards/benchmark claims were not promoted into public facts.

Verification

Repeated at closeout: 16 replay tests passed; nine current-handoff tests passed; 16 intake tests ran with 15 passing and one Windows symlink-privilege skip; nine public-semantics tests passed; three offline deployment rollback tests passed; both deterministic export/public-semantics freshness checks passed. The intake suite also tests exact Unicode/BOM/line-ending custody, unsafe paths/archives and incomplete-closeout rejection. The final empty-intake gate passed independently of those fixtures. Git whitespace validation passed.

Earlier in this continuation, the machine-guide, gateway, facade, HTML, discovery, host routing and MCP suites passed with their exact evidence scope recorded in the relevant reviews. Browser checks covered bounded desktop/mobile widths, keyboard access and narrow-table geometry. Ten operator-controlled live checks passed against MySQL-backed production, and disposable message/credential cleanup completed. None of these results is labeled an independently operated acceptance run, whole-site accessibility certification or full infrastructure-privacy proof.

Handoff limits

Independent end-to-end acceptance remains open. Existing cached/history bodies, recipient copies and backups were not physically erased; current public tombstone projection is not global forgetting. A source's proposed 64-case catalogue is still a proposed catalogue, not a fabricated test run. Broad new legal/identity/ governance platforms and new regional blocks were explicitly not adopted.

Processed originals and scoped changes are pushed to the user-approved MichaelKappel/Concresca repository. Final repository and root-deploy artifacts are built from the final explicit committed revision; filenames remain -wip.zip because mandatory independent acceptance is unresolved. Their package report records exact hashes and the public/source separation.

No background recurring intake processor was created. Approval to push this work does not broaden into recurring write/deploy authority. The current on-demand handoff scans for new input at startup and before completion. Unrelated drafts scripts/concresca_machine_controller.py and scripts/test-v048-machine-controller.py remain untouched and excluded from this checkpoint.