feat(ecology): record and replay joint environment intervals
ADR-0173 adds joint batch AEVOEBT1: the unchanged population batch plus each tick's recorded infall import and radiated heat. Unattended recording applies each tick's exchange before the shared wildlife controller chooses, so a recorded interval equals the same live joint ticks; replay applies the same exchange and the recorded intentions and must re-derive every flow. Forged flows or boundaries, missing ticks, wrong cuts, truncated input and future formats are refused. Whole-group conversion replays exactly and cohort intervals record identical biology and flows. A full habitat counts infall bubbles skipped and imports nothing for them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -12,6 +12,11 @@ tick or not at all. Closed and source-off regions keep their biology byte for by
|
||||
Raw and cohort representations agree under infall. Old formats and goldens keep
|
||||
their meaning; no default/live version changes.
|
||||
|
||||
[ADR-0173](adr/0173-cellular-environment-batch.md) adds joint batch `AEVOEBT1`:
|
||||
the unchanged population batch plus each tick's recorded import and radiated heat.
|
||||
Recorded unattended intervals equal the same live joint ticks; replay re-derives
|
||||
every flow and refuses forged flows, boundaries and cuts.
|
||||
|
||||
Both full Release suites pass **94/94**; see [Validation](engineering/VALIDATION.md).
|
||||
Producers (ADR-0136 step C), durable admission (12.5c), worker/census/migration
|
||||
integration (12.5d) and target-host qualification (12.5e) remain open.
|
||||
|
||||
@@ -0,0 +1,48 @@
|
||||
# ADR-0173: Cellular environment joint batch
|
||||
|
||||
## Status and scope
|
||||
|
||||
Accepted portable M12.5b evidence, 2026-10-19. Extends ADR-0172 and the population
|
||||
batches of ADR-0142/0144/0171. No durable admission, SQL schema, checkpoint profile,
|
||||
automatic scheduling or live/default rule change.
|
||||
|
||||
## Decision
|
||||
|
||||
Joint batch 1 (`AEVOEBT1`) carries one complete, unchanged population batch (format
|
||||
2 for envelope 2/owner 7, format 3 for envelope 3/owner 8) plus one recorded
|
||||
environmental flow per biological tick: imported energy/C/N/P/other and radiated
|
||||
heat. A whole-group conversion carries no flows. The joint owner's configuration,
|
||||
source progress and nested population are all in the predecessor envelope, so the
|
||||
batch adds no environmental input: replay re-derives every infall and export from
|
||||
configuration, tick and seed and must reproduce each recorded flow exactly.
|
||||
|
||||
**Unattended intervals.** Every controlled cell must be fully suspended at the cut
|
||||
and stays frozen. For each tick the recorder applies the joint exchange, then runs
|
||||
the unchanged wildlife controller on the exchanged state, so new deposits are
|
||||
visible to that tick's choices exactly as in a live joint tick. Recording equals the
|
||||
same ticks advanced live. Replay applies the same exchange, then the recorded
|
||||
intentions; the controller is not rerun. Cohort boundaries (sequence and members)
|
||||
are checked after each tick, as in the population batch. A whole-group conversion
|
||||
changes membership only.
|
||||
|
||||
The nested population batch keeps its exact format and meaning. It is never valid
|
||||
evidence for a joint result on its own: replaying it without the exchange spends
|
||||
different stock. Flows are evidence, not authority: a forged flow, boundary,
|
||||
missing tick, wrong cut, future format or truncated input is refused before any
|
||||
result is returned, and the base is never modified.
|
||||
|
||||
## Consequences
|
||||
|
||||
Durable joint publication (12.5c) can bind a manifest to the predecessor joint
|
||||
envelope and this complete trace, and recovery can replay it exactly without
|
||||
re-deriving controller choices. Recorded flows let SQL-side evidence report
|
||||
environmental crossings separately from migration flows without replay.
|
||||
Representation conversion is exact: cohort and individual intervals record
|
||||
identical biology and flows. No performance, compression or target-host claim.
|
||||
|
||||
## Alternatives
|
||||
|
||||
Recording the population batch alone and re-deriving flows was rejected: an
|
||||
auditor could not distinguish environmental crossings from migration without
|
||||
replay. Running the controller before the exchange was rejected: it would make
|
||||
unattended intervals diverge from live joint ticks.
|
||||
@@ -1,5 +1,7 @@
|
||||
# Architecture decision records
|
||||
|
||||
- [ADR-0173: Cellular environment joint batch](0173-cellular-environment-batch.md) - portable `AEVOEBT1` wraps the unchanged population batch with per-tick recorded environmental flows; unattended recording runs the controller after each exchange and replay re-derives every flow exactly.
|
||||
|
||||
- [ADR-0172: Cellular environment ownership](0172-cellular-environment-ownership.md) - one owner per patch of space for cells and their local environment; reserves stay individual; open planet boundary with recorded imports/exports; portable envelope `AEVOCEN1` with finite infall entitlements and radiative heat export.
|
||||
|
||||
- [ADR-0171: Retained-plan exact populations](0171-retained-plan-exact-populations.md) - portable envelope/batch 3 preserve every member's expressed/heritable state, exact conversion/policy replay and projection-2 plan counts; durable admission remains separate.
|
||||
|
||||
@@ -66,7 +66,7 @@ or authorize production enablement.
|
||||
| 12.2 | Exact cellular model and paired matrix; local samples through 256 lives; bounded four-generation rules-9 predation/twin matrix | Deployment-host tail costs including publication; larger wildlife cohorts and fuller contention coverage; useful compression remains unimplemented. |
|
||||
| 12.3 | Dedicated population profile/trace/store/owning host; exact recovery and private PostgreSQL fault tests | Population-specific executable/network fault drills and target-host publication/capacity qualification. |
|
||||
| 12.4 | Owned intervals; exact transfer/accounting; durable atomic pair/recovery; independent loopback TLS path loss and owner expiry/replacement/heal; local object sync/read-failure matrix | Multi-host/kernel-network, storage-host and orchestrator drills, linked-boundary contracts and integrated ownership/stock audits at scale. |
|
||||
| 12.5 | ADR-0172 portable joint owner: one budget for cells and local environment, effective depth, tagged water/dissolved compartments, refused atmosphere, finite infall entitlements, radiative export; closed/source-off/open/exhaustion and raw/cohort comparisons | Producer process and coefficients (ADR-0136 step C), durable profile/trace/recovery (12.5c), worker/census/migration integration and escrowed cross-worker sources (12.5d), target-host qualification (12.5e). |
|
||||
| 12.5 | ADR-0172 portable joint owner: one budget for cells and local environment, effective depth, tagged water/dissolved compartments, refused atmosphere, finite infall entitlements, radiative export; closed/source-off/open/exhaustion and raw/cohort comparisons; ADR-0173 joint batch with exact unattended replay | Producer process and coefficients (ADR-0136 step C), durable profile/trace/recovery (12.5c), worker/census/migration integration and escrowed cross-worker sources (12.5d), target-host qualification (12.5e). |
|
||||
| 12.6 | Retained archive, regional census/lineage queries, replayed history, scoped multi-region queries and consistent database head snapshots (12.6a-d); registered coverage/history domains follow in 12.7a-b | Cross-domain clone provenance, retained historical coverage/other viable stores, target-host query/index qualification and client permission/presentation; no account power or invented precise statistical ancestry. |
|
||||
| 12.7 | Viability/unknown policy; consistent cuts, registered coverage, history domains/restore forks and verified pending handoff evidence (12.7a-c/schema 52) | Cross-domain clone provenance, other viable-store coverage and durable accepted global evidence records/freshness. Missing/stale/unsupported owners must prevent terminal extinction. |
|
||||
| 12.8 | Deterministic seed/regime drivers, separate-process exact continuation and bounded unscripted generations | Reproducible integrated worlds running for actual days, real restarts/clone comparisons and retained accounting/diversity/cost audits. Accelerated tests cannot supply elapsed days. |
|
||||
|
||||
@@ -1,5 +1,38 @@
|
||||
# Validation report
|
||||
|
||||
## 2026-10-19 - Cellular environment joint batch (M12.5b)
|
||||
|
||||
[ADR-0173](../adr/0173-cellular-environment-batch.md) adds joint batch
|
||||
`AEVOEBT1`: the unchanged population batch plus each tick's recorded import and
|
||||
radiated heat. Recording runs the shared wildlife controller after each tick's
|
||||
exchange; replay applies the same exchange and the recorded intentions.
|
||||
|
||||
Executed locally:
|
||||
|
||||
- GCC 14.2 Release full CTest **94/94, 4.97 s** and MSVC 19.44 Release full CTest
|
||||
**94/94, 6.64 s**, with NDEBUG.
|
||||
- Clang 19.1 ASan/UBSan cellular subset **28/28, 107.92 s** (environment
|
||||
contracts 81.52 s), with no sanitizer reports.
|
||||
- clang-tidy 19 (warnings as errors) and clang-format 19 on the joint owner, batch
|
||||
and tests. GCC 14 -O3 reported `free-nonheap-object` for `reserve` followed by
|
||||
`push_back` in the encoder, a known false positive on this toolchain; the encoder
|
||||
no longer reserves. No warning was suppressed.
|
||||
|
||||
Evidence in the suite: a 64-tick unattended interval (20,760 bytes, 11,200,000 pJ
|
||||
imported, 832 pJ radiated) equals the same ticks advanced live, round-trips exactly
|
||||
and replays to identical bytes; its flows are the interval's complete crossing.
|
||||
Replay refuses a forged radiated amount, a forged cohort boundary, a missing flow,
|
||||
the wrong predecessor cut, a truncated input and a future format; active controlled
|
||||
cells cannot start an interval. A whole-group conversion changes membership only and
|
||||
replays exactly; the subsequent cohort interval records identical biology and flows
|
||||
to the individual one. A full habitat (4,096 deposits) counts infall bubbles skipped
|
||||
and imports nothing for them. An initial fixture ran an every-15-tick source through
|
||||
about 6,000 budding ticks, filled the habitat and correctly imported nothing in the
|
||||
interval; the source now starts after a closed probe finds the first wildlife, and the
|
||||
test checks that the not-yet-active source left that history unchanged.
|
||||
|
||||
No durable admission, SQL schema or target-host claim.
|
||||
|
||||
## 2026-10-19 - Cellular environment joint owner (M12.5a-b)
|
||||
|
||||
[ADR-0172](../adr/0172-cellular-environment-ownership.md) records the maintainer's
|
||||
|
||||
Reference in New Issue
Block a user