# Layatown study, level C (village-level runs: the 80-person world `jevtown`), primary comparison: seeds 6-10, out of sample for every system

World jevtown (hash a635a002c79cd306), scenarios hard-week / siege / strange-week, 240 ticks, `--fast`. Jev runs
`eval8-*-jev-s6..10` (fresh, jev-1.13.0, $2.70 in total), Layatown runs `lt1-*-laya-s6..10` (checkpoint v1, sha256
27f210cb…, served by layatown/serve.py on the RTX 3080 Ti, $0), rules runs `eval8-*-rules-s6..10`. Tables:
`levelC-primary.txt` (`bun compare`, mean ± 95% CI over seeds); character effects: `levelC-character.txt` (`bun report`,
villager-level permutation tests). Criteria C1-C4: layatown/docs/EVALUATION.md section 7.3.

| criterion | hard-week | siege | strange-week |
|---|---|---|---|
| C1 viability: population 80/80 every seed; threats overcome (Jev / Layatown / rules) | pass (1.00 / 0.96 / 0.80) | pass (1.00 / 1.00 / 0.76) | pass (1.00 / 1.00 / 1.00) |
| C2 character: 5 contrasts p < 0.01 in >= 4 of 5 seeds | pass (5/5 each) | pass (5/5 each) | **fail on "brave fights"**: 0/5 seeds for Layatown, and 0/5 for Jev too (only 2-6 brave villagers ever threatened per run; effect positive in all 10 runs, underpowered); other four 5/5 |
| C3 closer to Jev than rules on >= 6 of 8 metrics | pass (7/8; misses share changing action) | pass (8/8) | pass (6/8; misses exhaustion collapses and share changing action) |
| C4 individuality >= 0.8x Jev and above rules | pass (0.162 vs Jev 0.160, rules 0.055) | pass (0.150 vs 0.143, 0.048) | pass (0.177 vs 0.169, 0.064) |

Strictly read, C2 fails in strange-week, for the reference system as well: the scenario does not produce enough
threatened brave villagers to test that contrast. Reported as a failure of the criterion as written.

Serving: every Layatown run completed all 240 ticks with 0 skipped (30 of 30 across seeds 1-10); request latency p50
125-385 ms, p95 338-691 ms, against Jev's 302-360 ms p50 in the same scenarios. One fresh Jev run skipped 3 ticks.
Cost per hour at one tick per 8 s: Jev $0.31-0.38, Layatown $0 marginal (electricity not measured yet).

Note: these are whole-village runs of the 80-person world the study's scenarios were written for, not the
1,000-person production city; the production city's serving is in bench/results/serving/.