02 — Run · Round 2 · test · r2-smoke-gpt6luna
GPT-6 Luna vs Jev Router
Jev Router destroyed the enemy base after 10.1 game-minutes · 16.1 min wall clock
LOSS russia · west
GPT-6 Luna
openai/gpt-6-luna
Value destroyed$2,100Value lost$22,000
Units killed / lost19 / 37Buildings killed / lost0 / 11
Peak army$2,000Orders issued86
Decision turns (failed)76 (0)Mean latency3.1sModel cost$0.027
WIN ukraine · east
Jev Router
typesafe/jev-router
Value destroyed$22,000Value lost$2,100
Units killed / lost37 / 19Buildings killed / lost11 / 0
Peak army$10,700Orders issued99
Decision turns (failed)76 (13)Mean latency5.1sModel cost$0.162
Evaluation
passchecks 10/10 verdict computed from the gating checks
| Status | Kind | Check | Detail |
|---|---|---|---|
| pass | step | environment | Environment check (snapshot, engine, content, key) |
| pass | step | launch | Launch run-match as a detached process |
| pass | step | supervise | Supervise in the foreground until run-match exits |
| pass | step | render | Render the 1080p recording |
| pass | step | notes | Append Notes to summary.md |
| pass | code_check | outputs-exist | all present |
| pass | code_check | game-has-winner | jev won: base destroyed |
| pass | code_check | decisions-logged | turns per slot: {'Multi0': 76, 'Multi1': 76} |
| pass | code_check | llm-turns-valid | failed/total per slot: {'Multi0': '0/76', 'Multi1': '8/76'} |
| pass | code_check | video-decodes | duration 102.5s |
| pass | code_check | video-complete | video ends at tick 15078 of 15078 |
| pass | code_check | limits-respected | ended after 16.1 wall-minutes (budget 30.0 min): base destroyed |
| pass | checklist | notes-explain-win | summary.md Notes section cites both models final thoughts explaining how Jev Router won by destroying all enemy buildings |
| pass | checklist | players-match-request | result.json records Multi0=gpt6luna (openai/gpt-6-luna) and Multi1=jev (typesafe/jev-router) matching the requested players |
| pass | checklist | no-silent-outage | run.log shows continuous decision ticks every 200 ticks with no long gaps; Jev Router had 8 failures but they are within the 20% threshold and no silent stretches |
Decision timeline
What each model saw as its situation and what it ordered, every 8 game-seconds. Click a turn to jump the video there.