Pokémon Random Battle
- Score
- 0.0
- Wallet spend
- $0.00 of $10
- Model cost
- $1.00
- Agent time
- 1h 55m
- Steps
- 145
- Tokens in / out
- 9.6M / 49k
- Self-checks
- 3 passed of 4 used
- Rank at $10 wallet
- 6th of 19
Score
0.0raw metric 0.002667
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
- 8
- draws
- 0
- losses
- 2992
- capped_games
- 0
- opponent_forfeits
- 0
- move_timeout_forfeits
- 2988
- paired_standard_error
- 0.0009
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 declares the reused code: Foul Play at pinned revision 6c467c08 (GPL-3.0) and poke-engine 0.0.48. It a
architecture.json declares the reused code: Foul Play at pinned revision 6c467c08 (GPL-3.0) and poke-engine 0.0.48. It also declares that there is no base model and no training, and it lists foul-play-priors.json as a data source copied from /task/public. The trajectory confirms this at step 39 (the priors were copied from /task/public) and at step 37 (the vendored foul-play tree came from the pinned clone). At steps 34-36, the poke_engine wheel was copied from the organizer's /task/public/engine and the PyPI source tarball was deleted. Trusted audits approve the shipped wheels. No learned weights are shipped and no training was done, so the 'from-scratch' label claims no initialization history.
✓actual served policyserve.sh installs the shipped wheels offline with --no-index and then runs server.py. server.py's /act calls bot.act_bat
serve.sh installs the shipped wheels offline with --no-index and then runs server.py. server.py's /act calls bot.act_batch. In bot.py, act_batch rebuilds each battle's state from the public_log and the player's request. It then picks a move through the forced or revival paths, or through Foul Play MCTS (prepare_battles, get_result_from_mcts, select_move_from_mcts_results in a local ProcessPoolExecutor). The chosen move is mapped to an index in legal_actions, with a legal fallback of 0. The served policy is this local search over the declared code; no external or hidden model is used.
✓permitted informationDecisions use only the per-observation request, the public_log and legal_actions. bot.py blocks network access by monkey
Decisions use only the per-observation request, the public_log and legal_actions. bot.py blocks network access by monkeypatching socket.connect and create_connection. Nothing reads opponent private state or detects a test environment. Tuning during development used only the public validation and selfcheck tools, which counts as validation-guided selection and is allowed.
✓harness credentials used only by the harnessThe trajectory shows no step that reads, copies or uses harness API credentials or base URLs. No other agent or model CL
The trajectory shows no step that reads, copies or uses harness API credentials or base URLs. No other agent or model CLI was launched. The only curl calls were health checks against the local 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.
- 1failed43m 41s into the runtook 14m 5s
startup 1.5 s of 300 s; failed POST /act, round 107: RuntimeError: Submission returned HTTP 500: b'{"error": "\'NoneType\' object has no attribute \'get_move\'"}'
3 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.5 s
- ✗POST /act, round 107 · RuntimeError: Submission returned HTTP 500: b'{"error": "\'NoneType\' object has no attribute \'get_move\'"}'
- 2passed1h 6m into the runtook 3m 28s
startup 1.6 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 66 of 124 POST /act responses (26-32 observations, up to 162 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.000 s, p95 1.001 s. Grading forfeits every game in a
6 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.6 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), 1940 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3695 observations, 26-32 per batch, up to 162 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 66 of 124 POST /act responses (26-32 observations, up to 162 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 1.000 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 1972 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 0.964 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
- 3passed1h 10m into the runtook 12m 8s
startup 1.5 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 60 of 124 POST /act responses (26-32 observations, up to 165 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.952 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, 1.14 s round trip on this side with the transport: [1, 7, 3, 7, 5, 5, 7, 11, 10, 11, 7, 0, 9, 12, 9, 5, 9, 11, 5, 7, 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), 1771 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3692 observations, 26-32 per batch, up to 165 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 60 of 124 POST /act responses (26-32 observations, up to 165 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.952 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 1802 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 0.862 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
- 4passed1h 33m into the runtook 5m 25s
startup 1.3 s of 300 s; 1/1 gates passed; WARNING move response within 1 s: 41 of 124 POST /act responses (25-32 observations, up to 185 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.937 s, p95 1.001 s. Grading forfeits every game in a
7 gates
- ✓review prerequisites
- ✓startup · GET /health answered 200 after 1.3 s
- ✓POST /act, round 1 · 32 opening decisions answered with legal actions, 0.87 s round trip on this side with the transport: [1, 7, 3, 7, 5, 5, 7, 11, 10, 11, 7, 0, 9, 12, 1, 5, 9, 11, 5, 7, 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), 1240 finished games replaced by further development-seed matchups, against the pinned Foul Play: 3646 observations, 25-32 per batch, up to 185 KB per request, every answered action legal; not scored
- ✓move response within 1 s · 41 of 124 POST /act responses (25-32 observations, up to 185 KB; the first a cold batch of 32) took longer than 1 s on the trusted in-sandbox move timer; median 0.937 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 1240 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 0.853 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