feat(lifecycle): preserve inherited plans in exact populations
Foundation / checks (ecs) (push) Successful in 2m26s
Foundation / checks (deploy) (push) Successful in 4m57s
Foundation / checks (network) (push) Successful in 20m53s
Foundation / checks (client) (push) Successful in 22m12s
Foundation / checks (tsan) (push) Successful in 39m45s
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 45m41s
Foundation / checks (checks) (push) Canceled after 45m43s

This commit is contained in:
dingo
2026-10-19 15:19:43 -04:00
parent fad79973e2
commit a898a35e1b
22 changed files with 582 additions and 36 deletions
+52
View File
@@ -1,5 +1,57 @@
# Validation report
## 2026-10-19 - Portable retained-plan exact populations
[ADR-0171](../adr/0171-retained-plan-exact-populations.md) adds explicit envelope
3 and population batch 3/nested raw trace 4. Every member retains its complete
expressed/heritable state; the no-tick inheritance candidate preserves policy,
sequence and tags. Projection 2 separately counts cohort plan references and
expressed stock. Existing checkpoint/SQL population allowlists are unchanged.
Executed locally:
- MSVC 19.44 Release full CTest **92/92, 29.63 s** and GCC 14.2 Release full
CTest **92/92, 19.99 s**, with NDEBUG. The final additional conversion and
automatic-policy proofs passed **2/2, 2.82 s** on MSVC and **2/2, 1.80 s**
on GCC. Only these tests changed after the full runs; production sources did not.
- Clang 19.1/clang-tidy 19 new population/restart and census **3/3, 21.17 s**.
Final expanded population/restart tests **2/2, 15.22 s**; format check passed.
- Clang 19.1 ASan/UBSan population/restart, inherited owner and census
**4/4, 108.71 s**, without reports. Final expanded population/restart proofs
**2/2, 48.62 s**, without reports.
- GCC 14.2 persistence consumers rebuilt and the private population/fault suite
passed **37 s total**. Existing raw/cohort, interval, migration and registered
history behavior passes; schema 55 still refuses retained-plan population
publication/profile conversion. No new SQL admission was inferred from readers.
The fixture creates a real funded legacy child and whole-group cohort before
upgrading. It compares the complete nested state and outcomes with an independent
raw continuation every tick through two earned structural plans, funded children,
offline suspension and 64-tick intervals. Manual conversions encode/replay exact
membership without changing biology; automatic policy with a fully suspended
observer collapses every actual wildlife member and replays the same boundaries.
Census equality after normalizing only the representation category preserves all
per-genome stock and independent plan/dead-history measures. A suspended selected
plan still has one holder and zero expressed stock before birth.
Tests refuse old/unknown outer versions resealed with a valid checksum, truncated
envelopes, old raw trace pairings, changed membership boundaries, repeated upgrade
and active-controlled unattended admission. Source bytes remain unchanged on
detached preparation/refusal. Fresh-process full-population continuation is exact;
MSVC/GCC expected output shares SHA-256
`818824b44bb4475177c04492ff07df4b657d3370c90d96cf7ab6ac84f9b1d297`.
An initial test incorrectly used an already populated opening and failed its
empty-input genesis tick. The fixture now explicitly requests an empty opening;
no biological rule or golden was changed to make it pass.
This retains exact individuals and makes no compression, throughput or target-host
cost claim. Database admission, cross-region migration, producer/resource coupling,
control succession/NPC behavior/UI and actual-days/product qualification remain
separate. No new service, default/DEV opening, live upgrade or World0 operation.
The refreshed population-segment DEV window expires **2026-10-19 23:19:02 UTC**.
Rendered deployment policy passes with 138 objects across 17 overlays and zero
failures. A successful push does not establish deployed reachability.
## 2026-10-19 - Explicit worker-only DEV scheduling
[ADR-0170](../adr/0170-dev-worker-placement.md) applies required negative