M12.7: reconcile pending migration evidence in census snapshots
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 2m26s
Foundation / checks (network) (push) Successful in 21m1s
Foundation / checks (client) (push) Successful in 22m9s
Foundation / checks (tsan) (push) Successful in 24m36s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (postgres) (push) Canceled after 33m38s
Foundation / checks (checks) (push) Canceled after 33m49s

This commit is contained in:
dingo
2026-10-18 21:14:53 -04:00
parent 709dfb9557
commit 27ec5729bf
23 changed files with 592 additions and 23 deletions
+79
View File
@@ -1,5 +1,84 @@
# Validation report
## 2026-10-18 - M12.7c registered census pending handoff evidence
[ADR-0156](../adr/0156-census-pending-migration-evidence.md)/schema 52 captures
the union of active source/destination migration pointers in the same read-only
repeatable-read view as domain, complete registry and current heads. The source
predecessor must match its selected head; a staged destination must also match.
Both predecessor builds must be allowlisted, including an older optimistic
destination cut. After capture, full source capsule and destination ghost replay
verifies exact members/material/bytes, without adding any ghost stock to census.
Executed locally:
- MSVC default Release full CTest **79/79 passed, 14.14 s**.
- GCC 14 Release final semantic-failure population/history/migration/census matrix
passed **exit 0, 18 s** (16 s population). The final predecessor-build allowlist
guard and the complete affected cases passed **exit 0, 17 s** (16 s population).
- Clang 19 **ASan/UBSan** complete affected raw/population matrix, real loopback
TLS request/reply loss and local storage faults passed **exit 0, 77 s**
(73 s population), without sanitizer reports, before the final allowlist guard.
The final rerun passed **exit 0, 71 s** (69 s population), without reports.
- clang-tidy 19 passed on both changed persistence adapters and integration test
files; final semantic-failure assertions passed a separate recheck. Final
allowlist analysis/format also passed. Changed shell scripts pass `bash -n`.
- Initial full private persistence/process/backup/DR and SQL qualification passed
**exit 0, 539 s** before the final allowlist guard. The final frozen-binary rerun
passed **exit 0, 558 s**: C++ integration/process/load/history **435 s**, migration
TLS/storage faults **42 s**, life reservations **4 s**, domain/legacy/backup/
restore **31 s**, host-loss DR **8 s**, upgrade **10 s**, migration history
**5 s**, runtime permission audit **13 s**, checksum/future rejection **8 s**,
plus setup/migration. The final completion sentinel records zero and the log
contains the all-phase success summary. The new getter's missing-grant case
correctly fails the effective-privilege audit before restoring the grant.
- Final `git diff --check` passed; private test clusters stop on exit and preserve
forensic directories. No biological golden or prior numbered migration changed.
Both representations, in both directions, use real player-origin wildlife and
independent connections. Capture before preparation sees no pending request;
the next view sees source-only evidence. A concurrent stage leaves an earlier
view prepared and the next staged. A concurrent abort or paired commit preserves
the captured staged evidence; the next view sees no pointers and, for commit,
both new heads. All pending views retain the original source-owned census stock.
No receipt, ghost or transfer amount is added as another resident population.
Actual read denial of the source's destination capsule and the destination ghost
fails the registered result without fallback to live peer state. Metadata-only
false transferred material and a digest-valid object with the wrong complete
ghost biology are rejected by full replay. Trusted owner fault seams move a
source pointer to the wrong region and inject 65 backend rows; these fail without
partial evidence. The overflow fixture is a response-bound test, not 65 valid
simultaneous migrations or a qualified world size. The original SQL function and
pointer are restored before continuing the real workflow.
Schema 52 adds one least-privilege read function and no state table/index,
registration/publication writes, biological format/rule/golden or dependency.
Prior numbered migrations 1..51 are unchanged. Latest-prefix assertions include
52; unknown-future rejection uses 53. Runtime audit requires the new read grant.
The pre-context backup fixture removes both schema-51 metadata and schema-52
getter to reconstruct the exact old prefix; old external ancestry is not invented.
Limits: active-pointer evidence is not an audit of every historical preparation,
coverage of every viable store, accepted durable global freshness or terminal
extinction. No target-host capacity, new TSan, remote CI, multi-host/storage-host
failure, cellular clone/pending-migration backup restore, actual multi-day ecology
or product acceptance is claimed. Generic backup/DR regression remains separate
from cellular restore qualification. No live database, deployment or World0 action.
Logs: `build/m127c-build.log`, `build/m127c-rebuild.log`,
`build/m127c-final-build.log`, `build/m127c-native.log`,
`build/m127c-population.log`, `build/m127c-final-population.log`,
`build/m127c-asan-build.log`, `build/m127c-asan.log`,
`build/m127c-tidy.log`, `build/m127c-format.log`,
`build/m127c-final-tidy.log`, `build/m127c-final-format.log`,
`build/m127c-allowlist-build.log`, `build/m127c-allowlist.log`,
`build/m127c-final-asan-build.log`, `build/m127c-final-asan.log`,
`build/m127c-allowlist-tidy.log`, `build/m127c-allowlist-format.log`,
`build/m127c-shell.log`, `build/m127c-full.log`,
`build/m127c-full.exit`, `build/m127c-final-full.log`,
`build/m127c-final-full.exit` (completion only).
## 2026-10-18 - M12.7b database history domains and restore provenance
[ADR-0155](../adr/0155-database-history-domain.md)/schema 51 adds a database