Track completed-roadmap audit corrections, evidence gaps, and integration gates

[flyingrobots]

Completed-roadmap audit follow-up tracker

This is a tracking container, not an executable issue or an additional PR. It accounts for the 2026-09-30 audit's findings, prepared corrections, integration gates, and unfinished certification of the original 83 completed entries. Originally unfinished roadmap features are excluded.

Source: docs/audits/completed-roadmap-2026-09-30.md and original criteria in ROADMAP.md, currently on feature/roadmap-m4-tasks. Checklist entries mean the obligation is tracked, not that acceptance is satisfied. Close this container only when every original completed task has a criterion-by-criterion verdict and every required correction is integrated or explicitly remains blocked with truthful roadmap status.

New executable follow-ups

• Finish typed refusal audit across durable boundaries #110 — Finish typed refusal audit across durable boundaries (T-01.2)
• Complete migration restart corruption and ambiguity matrix #111 — Complete migration restart corruption and ambiguity matrix (T-17.2; KEEP-MIGRATION-005)
• Complete migration compatibility and fuzz evidence #112 — Complete migration compatibility and fuzz evidence (T-17.3; KEEP-MIGRATION-008)
• Prove reader-fence process death and collector exclusion in external harness #113 — Prove reader-fence process death and collector exclusion in external harness (T-19.1)
• Implement durable verification reports at explicit achieved depths #114 — Implement durable verification reports at explicit achieved depths (T-21.1; KEEP-VERIFY-006)
• Assert durable mutation failures through verification reports #115 — Assert durable mutation failures through verification reports (T-21.2)
• Add dedicated compaction process-death recovery matrix #116 — Add dedicated compaction process-death recovery matrix (T-22.3)
• Freeze reproducible compaction amplification and temporary-space baseline #117 — Freeze reproducible compaction amplification and temporary-space baseline (T-22.3)
• Prove GC planning at the 65,536-candidate acceptance bound #118 — Prove GC planning at the 65,536-candidate acceptance bound (T-22.4)
• Add explicit orphan-disposition process-death recovery matrix #119 — Add explicit orphan-disposition process-death recovery matrix (T-22.5)
• Run complete ingestion contract and Worldline against both store adapters #120 — Run complete ingestion contract and Worldline against both store adapters (T-24.1; T-24.2)
• Implement atomic durable ingestion across segment rollover #121 — Implement atomic durable ingestion across segment rollover (T-24.2)
• Prove durable ingestion memory bounds with multi-GiB and ceiling soak evidence #122 — Prove durable ingestion memory bounds with multi-GiB and ceiling soak evidence (T-24.2)
• Add ingestion-driven process-death and restart matrix #123 — Add ingestion-driven process-death and restart matrix (T-24.2)
• Publish source-bound durable ingestion throughput and resource baseline #124 — Publish source-bound durable ingestion throughput and resource baseline (T-24.2)
• Enforce retained-closure admission for low-level version-two catalog publication #125 — Enforce retained-closure admission for low-level version-two catalog publication (T-38.3 audit finding; Integrate bounded CDC deduplication with durable publication #82)
• Add backend-neutral Worldline copy witnesses for transfer pipelines #126 — Add backend-neutral Worldline copy witnesses for transfer pipelines (T-24.3)
• Integrate transfer-source port ownership regression fix #127 — Integrate transfer-source port ownership regression fix (T-10.2)
• Integrate retention release/restore model coverage #128 — Integrate retention release/restore model coverage (T-20.1)
• Integrate exact-length admission before durable stage allocation #129 — Integrate exact-length admission before durable stage allocation (T-24.2)
• Integrate corrected crate and version-two implementation documentation #130 — Integrate corrected crate and version-two implementation documentation (T-38.2; T-38.3)
• Complete acceptance and definition-of-done audit of the remaining 64 checked tasks #131 — Complete acceptance and definition-of-done audit of the remaining 64 checked tasks (Original completed-task audit remainder)

Reused issues

• Refresh segment-store v1 living documentation after recovery delivery #69 — T-14.1
• [BAD CODE™] Avoid double verification pass in reference read paths #71 — T-06.3
• [BAD CODE™] In-memory reference staging still materializes all missing chunks before publish #74 — T-06.4
• Define a restart-stable root identity coordinate for the migration intent #97 — T-17.1
• Implement partial-prefix migration recovery and the KEEP-CRASH-053..073 matrix #108 — T-17.2 / T-17.3
• Implement durable authenticated reads bound to a fenced snapshot (KEEP-RECONSTRUCT-009, -010) #109 — T-23.1
• [COOL IDEA™] Add bounded streaming write-through path for read→write and copy pipelines #72 — T-24.3
• Implement precise verification reports and corruption refusal #20 — T-21.1 / T-21.2
• Implement deterministic GC planning and identity-preserving compaction #21 — T-22.3 / T-22.4 / T-22.5
• Integrate bounded CDC deduplication with durable publication #82 — T-24.1 / T-24.2

Integration and external gates

• Retention recovery, crash-matrix evidence, reader fence, and model-based transitions (item 6) #99 is actually merged before T-18.1, T-18.2, T-19.1, and T-20.1 are declared done under their original definitions. Its existing 51-case retention matrix is evidence; the stale KEEP-RETENTION-006 ledger correction is prepared locally. This is integration tracking, not a second implementation PR.
• Implement the Echo–Keep physical-content boundary echo#722 supplies actual adapter/cutover evidence for T-25.2. It is open, and production replacement was explicitly unapproved when checked. Keep cannot infer completion from the merged boundary document. Reuse the external owner; no duplicate Echo issue is created.
• T-01.2/T-06.3/T-10.2/T-20.1/T-24.2/T-38.2/T-38.3 prepared local corrections retain issue → PR → mainline integration-commit traceability. Green local tests are not merge evidence.

Graph and coverage

The report-level corruption issue requires the durable report implementation. Beyond-single-segment scale evidence requires rollover. Other conditional prerequisites are spelled out in each issue with reasons. These are proposed dependency edges; this checklist alone does not record a GitHub blocking graph. Shared parents or files are not blockers.

Each executable outcome should land in one independently verifiable PR preserving a working target branch. Broader existing issues #20, #21, #82, and #108 retain original requirements and are coordinated with follow-ups rather than counted as duplicate implementation units. #131 owns complete accounting for the remaining 64 checked tasks; the first pass reopened 19 and did not certify the other 64.

[flyingrobots]

The T-09.1 audit reproduced ambiguous source metadata accepted by benchmark report ingress. New executable correction owner: #142. The issue covers the complete canonical report grammar and publication refusal, not benchmark threshold/durable-scenario work (those tasks were originally unchecked). Docker RED was the running assertion conflicting source coordinates were admitted; no artifact was published. The historical baseline/corpus/scenarios are present, but canonical admission DoD is unmet. This correction has no semantic prerequisite on migration or reference-read optimization.