Pokémon Random Battle
- Score
- 0.0
- Wallet spend
- $0.00 of $200
- Model cost
- $4.06
- Agent time
- 2h 32m
- Steps
- 457
- Tokens in / out
- 53.8M / 105k
- Self-checks
- 4 passed of 4 used
- Rank at $200 wallet
- 6th of 19
Score
0.0raw metric 0.01167
The raw metric (mean battle points against Foul Play) is rescaled so that 50% of battle points against Foul Play scores 0 and 100% scores 100, clamped to that range.
Verifier details
- metric
- mean_battle_points
- samples
- 3000
- wins
- 35
- draws
- 0
- losses
- 2965
- capped_games
- 0
- opponent_forfeits
- 0
- move_timeout_forfeits
- 2925
- paired_standard_error
- 0.0019
After grading, a reviewer model reads the submission's files and the agent's trajectory and checks each of the task's rules. A failed check makes the score 0. Step numbers link into the trajectory.
✓implementation provenanceThe submission declares no base model (base_model and base_revision are both "none") and no training data. It reuses Fou
The submission declares no base model (base_model and base_revision are both "none") and no training data. It reuses Foul Play at pinned revision 6c467c08 and poke-engine 0.0.48. The foulplay/ tree was extracted from the organizer's public foul-play tarball and copied into the output (trajectory#20). The only change is a small ZeroDivisionError guard in fp/battle/inference.py. The poke_engine wheel matches a trusted organizer-built audit. The other wheels are the pinned Foul Play requirements, downloaded with pip (trajectory#23). An existing bot modified without training is allowed, and the declared provenance matches the code and the session.
✓actual served policyserve.sh installs the vendored wheels offline into /tmp/bot-deps and runs server.py. server.py starts 8 workers running
serve.sh installs the vendored wheels offline into /tmp/bot-deps and runs server.py. server.py starts 8 workers running foul_play_worker.py, which builds a Battle from the observation, runs Foul Play's forced/revival/find_best_move MCTS search through poke_engine, and maps the result to a legal_actions index. Indices are clamped to the legal range, and errors fall back to action 0. The development servers were deleted before submission (trajectory#372). The served path is the declared Foul Play policy, with no cached or remote actions.
✓permitted informationEach decision uses only the per-request observation (the request and public_log for the acting player) plus the public F
Each decision uses only the per-request observation (the request and public_log for the acting player) plus the public Foul Play set priors JSON, which is a copy of the public task file. Workers block socket connections, and nothing reads opponent-hidden state or validation answers. validation.jsonl was used only for local evaluation during the session, not shipped as action caches.
✓harness credentials used only by the harnessIn the trajectory, API_KEY and BASE_URL appear only in documentation text. No step calls model APIs with harness credent
In the trajectory, API_KEY and BASE_URL appear only in documentation text. No step calls model APIs with harness credentials, copies them, or launches extra harness CLIs. Network activity was limited to pip downloads of the pinned requirements and curl against the local 127.0.0.1 server.
LLM adversarial review is not proof of arbitrary program behavior. Host storage and task-specific deterministic checks remain required.
frontier selfcheck runs the submission on the grading infrastructure against public examples and reports each gate, so the agent can catch problems before grading. Each run gets four. This agent used 4 of 4.
- 1passed2h 16m into the runtook 4m 54s
startup 1.2 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 124 of 124 POST /act responses (32-32 observations, up to 154 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.001 s, p95 1.001 s. Grading forfeits every game in a
6 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.2 s
- ✓battle slice · 124 POST /act batches over 124 decision rounds, 32 games at a time: the first 32 public games (teams from development seed 91827461), 3936 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3968 observations, 32-32 per batch, up to 154 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 124 of 124 POST /act responses (32-32 observations, up to 154 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.001 s, p95 1.001 s. Grading forfeits every game in a late batch (each counts as a loss) and plays on, so this slice lost 3968 games that way. A grade sends 3,722 requests and plays whole games, whose late turns carry longer public logs
- ✓execution budget (projected) · projected from 124 of 3,722 requests at concurrency 1: mean 1.001 s per request x 3,722, as many as one complete 3,000-game grade sent; grading sums every request's in-sandbox time
- ✓healthy after requests
- 2passed2h 22m into the runtook 1m 59s
startup 1.1 s of 300 s; 2/2 gates passed; The speed gates were measured on a sample and the execution budget projected from it: 124 of a complete grade's 3,722 POST /act requests, from the first decision rounds of the public games rather than whole games.
7 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.1 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 0.07 s round trip on this side with the transport: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
- ✓battle slice · 124 POST /act batches over 124 decision rounds, 32 games at a time: the first 32 public games (teams from development seed 91827461), 133 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3936 observations, 30-32 per batch, up to 269 KB per request, every answered action legal; not scored
- ✓move response within 1 s · slowest of 124 POST /act responses (30-32 observations, up to 269 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.007 s, p95 0.008 s. A grade plays whole games, whose late turns carry longer public logs, and sends 3,722 requests; every game in a response over the deadline is forfeited (a loss)
- ✓execution budget (projected) · projected from 124 of 3,722 requests at concurrency 1: mean 0.007 s per request x 3,722, as many as one complete 3,000-game grade sent; grading sums every request's in-sandbox time
- ✓healthy after requests
- 3passed2h 24m into the runtook 4m 19s
startup 2.2 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 116 of 124 POST /act responses (26-32 observations, up to 159 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.001 s, p95 1.001 s. Grading forfeits every game in a
6 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 2.2 s
- ✓battle slice · 124 POST /act batches over 124 decision rounds, 32 games at a time: the first 32 public games (teams from development seed 91827461), 3655 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3931 observations, 26-32 per batch, up to 159 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 116 of 124 POST /act responses (26-32 observations, up to 159 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.001 s, p95 1.001 s. Grading forfeits every game in a late batch (each counts as a loss) and plays on, so this slice lost 3687 games that way. A grade sends 3,722 requests and plays whole games, whose late turns carry longer public logs
- ✓execution budget (projected) · projected from 124 of 3,722 requests at concurrency 1: mean 0.995 s per request x 3,722, as many as one complete 3,000-game grade sent; grading sums every request's in-sandbox time
- ✓healthy after requests
- 4passed2h 29m into the runtook 3m 39s
startup 1.5 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 13 of 124 POST /act responses (22-32 observations, up to 291 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.815 s, p95 1.001 s. Grading forfeits every game in a
7 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.5 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 1.09 s round trip on this side with the transport: [5, 7, 3, 7, 5, 5, 7, 11, 9, 11, 7, 7, 7, 11, 9, 11, 9, 7, 5, 9, 5, 11, 11, 9, 0, 11, 5, 0, 9, 11, 6, 9]
- ✓battle slice · 124 POST /act batches over 124 decision rounds, 32 games at a time: the first 32 public games (teams from development seed 91827461), 395 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3593 observations, 22-32 per batch, up to 291 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 13 of 124 POST /act responses (22-32 observations, up to 291 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.815 s, p95 1.001 s. Grading forfeits every game in a late batch (each counts as a loss) and plays on, so this slice lost 388 games that way. A grade sends 3,722 requests and plays whole games, whose late turns carry longer public logs
- ✓execution budget (projected) · projected from 124 of 3,722 requests at concurrency 1: mean 0.816 s per request x 3,722, as many as one complete 3,000-game grade sent; grading sums every request's in-sandbox time
- ✓healthy after requests
The agent's last message before its session ended.
Submitted.