feat(persistence): admit durable funded maturation upgrades
Foundation / checks (deploy) (push) Successful in 6s
Foundation / checks (ecs) (push) Successful in 2m24s
Foundation / checks (network) (push) Successful in 20m40s
Foundation / checks (client) (push) Successful in 21m57s
Foundation / checks (tsan) (push) Successful in 26m32s
Foundation / checks (postgres) (push) Successful in 1h3m17s
Foundation / checks (checks) (push) Successful in 1h14m29s
Foundation / postgres-tools (push) Successful in 34s
Foundation / server-image (push) Successful in 4m6s
Foundation / image (push) Successful in 4m52s

This commit is contained in:
dingo
2026-10-19 06:23:51 -04:00
parent a5307c768b
commit 463b091056
25 changed files with 799 additions and 74 deletions
@@ -63,3 +63,7 @@ Rules-9 durable checkpoint admission, migration, registered census, environmenta
coupling and client presentation require their own contract/qualification. Current
durable bundle allowlists continue to reject this rule. No actual-days soak,
deployment, World0 operation, extinction or M12 completion is inferred.
ADR-0158/schema 53 subsequently admits explicit durable bundle-10 -> 11 upgrades.
The maturation rule above is unchanged; migration/census and ecological/product
qualification remain separate gates.
@@ -0,0 +1,52 @@
# ADR-0158: Durable admission of funded cellular maturation
## Status
Accepted for the scoped M12 maturation durability gate, 2026-10-19. Extends
ADR-0146/0157. No deployment, automatic upgrade, balance acceptance or World0 change.
## Decision
Admit cellular bundle 11 `{9,7,2,1,1,4,3,2,1}` through schema 53. Reuse the
immutable `cellular_profile_migrations` journal and current lease/head fence.
Only explicit bundle 10 -> 11, same-profile (4 or 5), same-tick upgrades are
allowed. Preserve all stocks, organisms, genomes, growth, presence, boundary
flows, archive, policy, cohort membership and conversion sequence. No new genesis,
skipped prerequisite, downgrade or independently published upgrade record.
The adapter verifies both predecessor bindings, derives and reads back the full
candidate object, then atomically records evidence/result/head. Pending intervals
and biological handoffs prevent a new upgrade. Exact retries retain original
bytes/receipt under current authority; changed bytes conflict. An uncertain
transaction retains the exact prepared request. Recovery verifies both complete
manifests/objects and the journal and rederives the recorded operation, selected
from the exact source/target tuple. No fallback to older biology or fresh genesis.
Both raw trace 3 and population envelope/batch 2 keep their existing layouts.
The bound predecessor's rules select funded maturation. Representation changes
preserve the tuple. Parser-local SQL normalization can reuse the existing shape
checks, after checking real trace binding; stored bytes, strong digests and retry
identity always retain bundle 11. Legacy parser/publisher helpers remain private.
Keep all original version meanings and migration checksums. Existing freeze,
interval, player-choice, permanent-death and dedicated-publication guards remain.
No worker upgrades automatically, and DEV/default provisioning stays rules 8.
An explicitly upgraded host can recover and continue either representation.
Biological migration still accepts bundle 10 only. Rules-9 census queries return
unsupported until separately qualified; a missing projection is never zero life
or evidence of extinction.
## Alternatives and qualification
A fresh opening loses funded ancestry and stock history. Reusing profile-change
evidence obscures a biological-rules change. A separate journal duplicates the
existing atomic authority. Extend its exact pair and replay dispatch instead.
Require both profiles, real juvenile growth after upgrade, exact continuation and
profile conversion, stale/current leases, retries/conflicts, pending-work refusal,
ambiguous outcomes/process loss, missing/corrupt/false source or result objects,
rollback, retained history and runtime privilege audits on private PostgreSQL.
Run portable Release, sanitizers, analysis and the full persistence/backup/DR
matrix. This qualifies the retained biological rule and its storage path, not
autonomous multi-generation ecology, linked environmental budgets, actual days,
deployment-host costs, product readiness or M12 completion.