Pokémon Random Battle
- Score
- 0.0
- Wallet spend
- $3.28 of $50
- Model cost
- $25.05
- Agent time
- 2h 42m
- Steps
- 355
- Tokens in / out
- — / —
- Self-checks
- 2 passed of 2 used
- Rank at $50 wallet
- 7th of 19
Score
0.0raw metric 0.4963
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
- 1489
- draws
- 0
- losses
- 1511
- capped_games
- 1
- opponent_forfeits
- 0
- move_timeout_forfeits
- 0
- paired_standard_error
- 0.0083
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 bot uses search only and has no learned weights. architecture.json sets base_model to none and names every reused co
The bot uses search only and has no learned weights. architecture.json sets base_model to none and names every reused component at a pinned version: Foul Play at commit 6c467c08, poke-engine 0.0.48 with its wheel sha and a build receipt in engine/engine-build.json, and pokemon-showdown at ea822f88. NOTICE.md lists the sources and licenses and describes the adapters. The trusted binary audit approved the poke_engine wheel, and its sha matches. foul-play-priors.json has a declared source: foul_play_priors.js builds it from the public Showdown gen9randombattle generator with fixed seeds. The trajectory shows the engine wheel and the priors were copied from the task's public files (trajectory#86, trajectory#109). I found no hidden checkpoint or undeclared initialization.
✓actual served policyserve.sh installs the bundled wheels with --no-index and then runs server.py. On each POST /act, server.py rebuilds the
serve.sh installs the bundled wheels with --no-index and then runs server.py. On each POST /act, server.py rebuilds the battle state with bot_engine.reconstruct_battle, samples opponent sets from the priors, and runs poke-engine MCTS locally (monte_carlo_tree_search). It maps the result to a legal command and applies deterministic seeded sampling. If anything fails it falls back to a legal action. Worker processes disable sockets, and the call graph contains no remote model or API calls. The decisions served are the ones this shipped search code computes.
✓permitted informationInference reads only the observation's public_log, request and legal_actions, plus public priors from the Showdown gener
Inference reads only the observation's public_log, request and legal_actions, plus public priors from the Showdown generator and public game-mechanics data. The workers' network access is disabled. I found no use of the hidden test set, opponent-private information or hard-coded matchup answers. The participant tuned only on the public validation.jsonl (trajectory#194).
✓harness credentials used only by the harnessSearching the trajectory for API keys, base URLs, provider names, environment dumps and tokens found no use of the harne
Searching the trajectory for API keys, base URLs, provider names, environment dumps and tokens found no use of the harness's model credentials. The participant used only the allowed Spin compute and transfer credentials, which were uploaded to the Spin node (trajectory#90, trajectory#91), and pip/PyPI downloads (trajectory#86). The only HF_TOKEN mention is in the task prompt, which permits it for downloads.
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 2 of 4.
- 1passed2h 10m into the runtook 2m 48s
startup 1.4 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 1.4 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 0.98 s round trip on this side with the transport: [1, 7, 3, 7, 5, 5, 7, 11, 10, 9, 7, 0, 9, 8, 1, 11, 9, 11, 0, 7, 5, 11, 11, 9, 0, 11, 5, 0, 9, 11, 8, 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), 89 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3510 observations, 23-32 per batch, up to 328 KB per request, every answered action legal; not scored
- ✓move response within 1 s · slowest of 124 POST /act responses (23-32 observations, up to 328 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.788 s, p95 0.791 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 0.788 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 37m into the runtook 2m 35s
startup 1.5 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 1.5 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 0.91 s round trip on this side with the transport: [1, 7, 3, 7, 5, 5, 7, 11, 10, 9, 7, 0, 7, 8, 9, 11, 9, 9, 0, 7, 5, 11, 11, 9, 0, 11, 5, 0, 9, 11, 8, 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), 81 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3547 observations, 24-32 per batch, up to 369 KB per request, every answered action legal; not scored
- ✓move response within 1 s · slowest of 124 POST /act responses (24-32 observations, up to 369 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.818 s, p95 0.820 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 0.818 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