Files
aevolution/docs/engineering/M12_UNSCRIPTED_GENERATIONS.md
T
dingo 3a1b8c7923
Foundation / checks (ecs) (push) Successful in 2m22s
Foundation / checks (deploy) (push) Successful in 4m16s
Foundation / checks (network) (push) Successful in 20m53s
Foundation / checks (client) (push) Successful in 22m3s
Foundation / postgres-tools (push) Canceled after 0s
Foundation / server-image (push) Canceled after 0s
Foundation / image (push) Canceled after 0s
Foundation / checks (tsan) (push) Canceled after 22m4s
Foundation / checks (postgres) (push) Canceled after 24m24s
Foundation / checks (checks) (push) Canceled after 24m33s
test(ecology): qualify unscripted maturing wildlife generations
2026-10-19 12:35:18 -04:00

110 lines
5.8 KiB
Markdown

# Bounded unscripted wildlife generations
Authority: [ADR-0162](../adr/0162-bounded-unscripted-wildlife-generations.md).
This offline instrument measures real funded player-origin organisms under the
existing controller and rules 9. No controller, biology, persistence, default
opening or live region behavior changes.
## Protocol
The shared finite ecology fixture supplies the entire opening. Player founders
request a bud every 100 ticks until a real funded bud succeeds. All players then
freeze for the rest of the run; their biology is checked unchanged. Wildlife uses
only the production controller. There are no scripted wildlife actions, mutation
selections, injected adults, reserve grants or replenishment. Ancestry follows
actual retained parent IDs, including actual twins; it is not genotype diversity.
The predeclared matrix uses four founders, rich finite resources and 80,000 ticks
on development seeds 0/1 and previously unmeasured hold-outs 2/3. Each case must
reach generation 3, retain living wildlife and produce births in its last quarter.
Raw and cohort execution compare complete canonical biology and bud, mutation and
growth reports every tick, with zero error permitted. Every tick checks complete
energy and C/N/P/other closure in both representations. Whole-genome conversion
alternates every 257 ticks. Both owners restore every 4,096 ticks, preserving
canonical bytes and the controller's next intentions. Controlled cells never enter
cohorts. The instrument reports conversion counts, member-ticks and ancestry by
founder. Rules-8 and resource-scarce controls run separately.
## Exploratory baseline
Before declaring the hold-outs, 80,000-tick raw rules-9 runs measured:
| Seed | Founders | Wildlife births | Deepest generation | Peak wildlife | Final wildlife |
|---|---:|---:|---:|---:|---:|
| 0 | 1 | 12 | 2 | 7 | 5 |
| 1 | 1 | 21 | 3 | 9 | 8 |
| 7 | 1 | 30 | 4 | 10 | 9 |
| 19 | 1 | 28 | 3 | 11 | 8 |
| 0 | 4 | 102 | 4 | 37 | 32 |
| 1 | 4 | 97 | 4 | 39 | 31 |
| 7 | 4 | 94 | 4 | 35 | 28 |
| 19 | 4 | 124 | 4 | 39 | 36 |
The first funded bud was tick 6,100 in every case. A rules-8 seed-7/one-founder
control had one birth, generation 1 and zero growth. The weaker single-founder
seed 0 is retained; no general all-opening or minimum-survival claim follows.
## Declared matrix results
The complete GCC 14.2 and MSVC 19.44 Release matrices pass with exact raw/cohort
biology, complete step/bud/mutation/growth reports and closure on every tick.
Their complete JSONL outputs match after newline normalization. Retained outputs:
[GCC](results/m12-unscripted-generations-gcc14.jsonl) and
[MSVC](results/m12-unscripted-generations-msvc1944.jsonl).
| Seed | Role | Births | Generation | Final wildlife | Final-quarter births | Engulfments | Twin events | Funded growth steps |
|---|---|---:|---:|---:|---:|---:|---:|---:|
| 0 | Development | 102 | 4 | 32 | 42 | 70 | 26 | 17,860 |
| 1 | Development | 97 | 4 | 31 | 37 | 66 | 20 | 16,796 |
| 2 | Hold-out | 102 | 4 | 33 | 35 | 69 | 19 | 17,718 |
| 3 | Hold-out | 100 | 4 | 33 | 37 | 67 | 17 | 19,468 |
Every founding ancestry produces wildlife: total births by root 100/101/102/103
are 12/27/36/27, 25/12/26/34, 20/30/27/25 and 23/25/29/23 respectively.
The rich runs contain no starvation deaths; their 272 deaths are actual
engulfments. The 82 twin events count actual same-parent newborn pairs, including
any founder twins. These measurements qualify those interactions in this matrix;
they do not establish balance or all contention states. No failed spatial bud
placement was observed, so the run does not qualify that refusal path.
Including controls, each full matrix covers **480,000 paired ticks**, 23,292 member
conversions, 2,993,603 cohort member-ticks and 114 paired restore checks (228 owner
restores). The rules-8 control retains one generation and zero growth. The scarce
control produces no buds or wildlife; its four active founders starve through the
normal biology. This is not a terminal-extinction acceptance record.
Native MSVC 19.44 Release CTest passes **82/82, 19.29 s**, including the short
12,000-tick rules-8/rules-9 contract. The same short contract passes Clang 19.1
ASan/UBSan. clang-tidy 19 and repository clang-format 19 pass. The first standalone
tidy invocation lacked the new target's compiler module map; building that analysis
target generated it and the final analysis passed. No production implementation
changed, so the already-qualified private persistence matrix was not repeated.
## Reproduction
After configuring the repository's Release preset and compiler environment:
```text
cmake --build --preset release --target evo-cellular-generations
ctest --preset release -R cellular-unscripted-generations
build/release/evo-cellular-generations --qualify
build/release/evo-cellular-generations --seed 7 --founders 1 --ticks 80000 --rules 9 --rich 1 --paired 1
```
On Windows the executable has the `.exe` suffix. `--verify` is the short CTest
contract: rules 8 and 9 each run 12,000 ticks with real funded birth, alternating
conversions and two restores. Rules 9 must grow; rules 8 must not. Failure is a
nonzero exit under NDEBUG as well as debug. Options and work are strictly bounded.
`--qualify` runs all four declared cases and both controls, retaining all case
results before reporting a failed ecological threshold.
## Limits
80,000 ticks equal 4,000 simulated seconds, not elapsed days. Rich finite contact
food does not establish sustainable producers, a complete nutrient cycle or a
joint cellular/environment owner. The inherited genome and production controller
are unchanged. This does not qualify automatic mutation, useful compression,
target-host costs, multi-host recovery, actual-days soak, terminal extinction,
returning-player enjoyment or M12 completion. No live/default upgrade or World0
operation follows from this instrument.