feat(persistence): admit durable retained offspring plans
Foundation / checks (tsan) (push) Canceled after 0s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (client) (push) Canceled after 1m11s
Foundation / checks (checks) (push) Canceled after 1m13s
Foundation / checks (postgres) (push) Canceled after 1m8s
Foundation / checks (ecs) (push) Canceled after 1m13s
Foundation / checks (network) (push) Canceled after 1m14s
Foundation / checks (deploy) (push) Canceled after 1m18s

This commit is contained in:
dingo
2026-10-19 15:04:30 -04:00
parent 39509ecd40
commit c60cdf8e19
20 changed files with 850 additions and 46 deletions
+56
View File
@@ -1,5 +1,61 @@
# Validation report
## 2026-10-19 - Durable raw retained-plan admission
[ADR-0168](../adr/0168-durable-retained-plan-admission.md)/schema 55 adds the
explicit raw bundle-11 -> 12 upgrade and trace-4 durable publisher/recovery.
Complete retained source and target bytes are independently bound, restored and
compared. A valid equal-stock but different-position upgrade fails recovery;
metadata checks alone never establish the biological transformation. SQL
parser-local normalization first verifies the actual trace length/SHA-256 and
version; retained evidence and retry identity use the original bytes.
Executed locally before the following census segment:
- MSVC 19.44 Release full CTest **89/89, 22.93 s**; GCC 14.2 Release **89/89,
17.58 s**, with NDEBUG.
- Clang 19.1/clang-tidy 19 persistence build and private population/fault suite
passed **56 s total**, including retained plans and existing raw/cohort,
interval, migration and registered-census controls.
- Clang 19.1 ASan/UBSan persistence build and the same private suite passed
**111 s total**, without sanitizer reports.
- GCC 14.2 full private schema-55 matrix passed **624 s total**: complete
integration suite (454 s), network partitions (82 s), reservation concurrency,
backup/history-domain/restore, disaster recovery, upgrade rehearsal, migration
history, runtime privileges and checksum drift. No shared database was used.
- Repository clang-format 19 and staged/working diff checks passed. No golden
was replaced. The exact staged deployment policy passes: 75 objects in 74
files, 138 rendered objects in 17 overlays, zero failures.
The refreshed DEV session window expires **2026-10-19 23:04:00 UTC**. Pipeline
build/deployment success and live reachability are not inferred from this push.
The real finite-food fixture follows the required earlier version upgrades,
preserves suspended lives, selects an earned plan without changing the parent's
expressed genome, publishes result 4 and produces at least two funded children
with that plan. Complete recovery under a replacement lease preserves the
family. Tests refuse skipped/repeated upgrades, population conversion, pending
mutation work, stale/current conflicting retries, wrong result meanings,
trace-3/owner-8 pairing and altered traces with the original digest. A triggered
result-insert failure rolls back journal/head; process loss before publication,
an ambiguous timed request, exact duplicate retry, source loss/corruption and
biologically false source transformation retain their explicit failure behavior.
An earlier full run used an old test binary missing the new migration-checksum
argument and failed its upgrade rehearsal. The test was corrected and rebuilt.
The next full run was canceled at the maintainer's stop request; its shell trap
printed an exit-0 timing summary before reporting the child hangup. That summary
is **not** a completed test pass. Only the subsequent uninterrupted run counts.
All database tests use disposable local clusters, not the shared DEV server.
The maintainer confirms delivery through Git Actions, Harbor, ArgoCD and
Kubernetes, with workloads in `evolution-dev` on workers and the existing
PostgreSQL/network/URLs. This workstation has no configured Kubernetes context;
direct access is not required for that delivery flow. No target-host performance,
multi-host failure, deployed reachability, actual-days ecology, product/UI,
TSan or World0 evidence is inferred from local tests or a Git push. Default
biology and player control behavior remain unchanged; M12 is still in progress.
## 2026-10-19 - Portable retained-plan checkpoint upgrade
[ADR-0167](../adr/0167-retained-plan-checkpoint-upgrade.md) adds raw-only bundle