Files
aevolution/docs/engineering/M12_POPULATION_QUALIFICATION.md
dingo bee57e5829
Foundation / checks (deploy) (push) Successful in 5s
Foundation / checks (ecs) (push) Successful in 2m47s
Foundation / checks (tsan) (push) Successful in 13m32s
Foundation / checks (client) (push) Failing after 17m9s
Foundation / checks (network) (push) Successful in 19m58s
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 20m34s
Foundation / checks (checks) (push) Canceled after 20m49s
M12.2c: qualify the current cellular reference instrument
2026-10-17 11:44:00 -04:00

87 lines
4.7 KiB
Markdown

# M12.2 population qualification: baseline reproduction
2026-10-17: [M12.2c](M12_CELLULAR_REFERENCE.md) adds the current rules-8 exact
cellular control with encoded replay, funded mutation/birth/death and offline
state evidence. M11.9 is closed for now by the maintainer. The actual cellular
cohort model and target-host budgets remain open.
Follow-up (2026-10-16): [M12.2b](M12_RESERVE_STRATA_EXPERIMENT.md) tests a separate
exact-reserve/age candidate and matches 83 synthetic trajectories on MSVC/GCC.
The baseline rejection below remains valid; no production cellular tier qualifies.
## Result (2026-10-15)
M12.2a reproduces the rejected M4.6 baseline without modifying its model or
historical results. All 15 sensitivity records and five policy records exactly
match the historical non-timing output. Eight scenarios exhibit false extinction;
zero widths qualify. **M12.2 is not complete. Automatic tier selection remains
disabled, and M12.3-12.5 implementation is blocked on qualification.**
Raw run: [m12-2-baseline-msvc1944.jsonl](results/m12-2-baseline-msvc1944.jsonl).
Historical method and full error matrix:
[AGGREGATE_BUDGET](AGGREGATE_BUDGET.md), ADR-0040/0041. Fixture version 2,
100 founders, 10,000 ticks, widths 10/25/50/100/200, energy supplies 240/300/360,
no randomness. Every run checks exact energy/matter conservation at every tick.
The executable's successful exit validates its arithmetic/instrument; it does
not mean that any reported fidelity gate passed.
## Execution and costs
Executed on the local Windows x64 development host with MSVC 19.44 Release
(`/O2`, `NDEBUG`), using the checked-in `release` preset and
`tools/windows-check.ps1 -Preset release -Action check`. Processor identifier:
`AMD64 Family 26 Model 68 Stepping 0, AuthenticAMD`. Hardware model retrieval
through CIM was denied in the sandbox; this is not target-host qualification.
No affinity or frequency lock was applied.
```powershell
./tools/windows-check.ps1 -Preset release -Action check
./build/release/evo-resolution-benchmark.exe |
Set-Content -Encoding utf8 docs/engineering/results/m12-2-baseline-msvc1944.jsonl
```
Warmup plus three fresh measured repetitions; baseline supply 300 median time for
10,000 ticks follows. These are uncontrolled local observations, not capacity bounds.
| Age width | Individual ms | Aggregate ms | False extinction at supply 300 |
|---|---:|---:|---|
| 10 | 30.084 | 0.241 | no |
| 25 | 30.616 | 0.055 | yes |
| 50 | 29.853 | 0.131 | no |
| 100 | 30.407 | 0.045 | yes |
| 200 | 28.648 | 0.046 | yes |
Small elapsed times for empty aggregates cannot count as useful throughput.
All original methodological limitations remain: dead individual rows are retained,
temporary scratch/allocator overhead and RSS are excluded, conversions are outside
timings, and there is one genome with abundant matter and no cellular space,
predation or M11 controller. This deterministic fixture has no hold-out seeds;
it cannot establish multi-genome diversity bounds or M11 demographic fidelity.
## Checks and limits
- Release configure/build passed with warnings-as-errors. The sandboxed CTest run
passed 67/69; `world-local-files-contracts` and `cli-file-integration` failed
where the sandbox denied filesystem hard-link fixtures. Repeating Release CTest
outside the sandbox passed **69/69**, with no source or test changes.
- Parsed every JSONL record; checked 15 sensitivity and five policy records;
compared every non-timing field in those 20 records against the retained
`resolution-budget-msvc1944.jsonl`. Exact match. Eight false-extinction cases,
zero accepted widths. The historical baseline was preserved.
- No C++ changes or candidate model were introduced. clang-format, clang-tidy
and sanitizers were not rerun for documentation/raw measurement output.
- A sandboxed `wsl --list --quiet` returned `Wsl/E_ACCESSDENIED`. No new Linux,
R730xd, PostgreSQL, worker-recovery, migration or World0 validation is claimed.
- M11 repeated-life product acceptance is not recorded. The first sign-in feedback
and engineering suite cannot stand in for that human review.
## Resolution before dependent implementation
Follow the [M12 delivery plan](M12_DELIVERY_PLAN.md): separately version a candidate,
predeclare trajectory/demographic/biomass/diversity bounds and held-out seeds/regimes,
retain within-cohort viable variation, compare against M11 cellular rules, exercise
bin/interval sensitivity and measure complete region/conversion/fallback costs on
the Linux target host. Unsupported states stay active only within a measured budget;
capacity refusal uses explicit backpressure, never culling or silent skipped work.
No numerical production error budget or fallback capacity is accepted by this run.