Pokémon Random Battle
- Score
- 0.0
- Wallet spend
- $2.47 of $50
- Model cost
- $5.77
- Agent time
- 11h 22m
- Steps
- 2,467
- Tokens in / out
- 248.4M / 2.2M
- Self-checks
- 4 passed of 4 used
- Rank at $50 wallet
- 7th of 19
Score
0.0raw metric 0.49
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
- 1470
- draws
- 0
- losses
- 1530
- capped_games
- 0
- opponent_forfeits
- 0
- move_timeout_forfeits
- 0
- paired_standard_error
- 0.0086
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 provenancearchitecture.json sets base_model/base_revision to none and checkpoint to null, and says no training was done. Policy al
architecture.json sets base_model/base_revision to none and checkpoint to null, and says no training was done. Policy allows a search-only bot with no checkpoint. Reused code is declared: Foul Play (pmariglia/foul-play at a pinned revision, vendored unmodified with runtime monkeypatch adapters in bot/fp_env.py), poke-engine 0.0.48 and the PyPI wheels. The poke-engine wheel is the organiser wheel; the trusted audit approves it and its sha matches the declared value. The only data source is the task's data/foul-play-priors.json, and its copy command is declared. Policy allows existing bots, including modified Foul Play. The weight inventory and uninspected-files list are both empty.
✓actual served policyI traced the full call graph. serve.sh installs the local wheels offline (pip --no-index), checks the imports, then runs
I traced the full call graph. serve.sh installs the local wheels offline (pip --no-index), checks the imports, then runs server.py. server.py forks worker processes (bot/worker.py). Each worker runs fp_env.setup, then decision.py rebuilds the battle from the request and public log with Foul Play's protocol parser. belief.py samples hidden opponent sets from the local priors, then poke-engine runs MCTS or expectiminimax. The choice is mapped back to legal_actions and its index is validated. On timeout or error the bot falls back to a linear heuristic over action_features. There is no remote decision service and no undeclared code path. Only /act and /health are exposed.
✓permitted informationDecisions use only the observation's request, public_log, legal_actions and action_features, plus the local priors file.
Decisions use only the observation's request, public_log, legal_actions and action_features, plus the local priors file. Opponent hidden information is sampled from public priors, not read from simulator state. bot/fp_env.py loads the sets by assigning datasets.raw_pkmn_sets from the local JSON and calling _initialize_pkmn_sets directly. That bypasses RandomBattleTeamDatasets.initialize/get_sets_file, so the upstream remote URL in vendor/foul-play/fp/data/sets/randbats.py is never fetched. Nothing in bot/ calls .initialize(. I found no network access, private evaluation access, test detection or cached actions.
✓harness credentials used only by the harnessI searched the trajectory and found no step that uses, copies or exports harness credentials, or that launches extra har
I searched the trajectory and found no step that uses, copies or exports harness credentials, or that launches extra harness CLIs with them. One harness file the participant read mentions MLE_JUDGE_API_KEY, but nothing used it. Credential-related activity was Spin CLI budget and compute calls, which policy allows.
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.
- 1passed5h 11m into the runtook 3m 4s
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, 1.01 s round trip on this side with the transport: [1, 7, 1, 7, 5, 5, 7, 11, 9, 9, 7, 0, 9, 11, 1, 5, 9, 11, 5, 7, 11, 11, 11, 9, 0, 11, 5, 0, 0, 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), 93 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3575 observations, 24-32 per batch, up to 306 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 306 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.801 s, p95 0.836 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 0.812 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
- 2passed7h 12m into the runtook 2m 42s
startup 1.7 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 1.7 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 1.00 s round trip on this side with the transport: [1, 7, 1, 7, 5, 5, 7, 11, 9, 9, 7, 0, 7, 11, 1, 5, 9, 11, 5, 7, 11, 11, 11, 9, 0, 11, 5, 0, 0, 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), 87 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3570 observations, 24-32 per batch, up to 334 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 334 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.863 s, p95 0.870 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 0.860 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
- 3passed8h 10m into the runtook 4m 25s
startup 2.1 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 2.1 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 1.38 s round trip on this side with the transport: [1, 7, 1, 7, 5, 5, 7, 11, 9, 9, 7, 0, 7, 11, 1, 5, 9, 11, 5, 7, 11, 11, 11, 9, 0, 11, 5, 0, 0, 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), 85 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3579 observations, 24-32 per batch, up to 323 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 323 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.867 s, p95 0.873 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 0.865 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
- 4passed10h 40m into the runtook 3m 5s
startup 2.4 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 2.4 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 1.07 s round trip on this side with the transport: [1, 7, 1, 7, 5, 5, 7, 11, 9, 9, 7, 0, 9, 11, 1, 5, 9, 11, 5, 7, 11, 11, 11, 9, 0, 11, 5, 0, 0, 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), 89 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3543 observations, 24-32 per batch, up to 326 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 326 KB; the first a cold batch of 32) on the trusted in-sandbox move timer; median 0.852 s, p95 0.865 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 0.842 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.
The arm came back within noise (+0.0078 ± 0.0197 paired), so the shipped configuration stays. Let me record that measurement in the docs and logs.