snooker-testbed-canary-7049627-stage0-v1

Phase-3 Option-B canary: stage-0 'sure-shot' curriculum bootstrap. Curriculum stages [0,1,2,3,4] where stage 0 places 1 red midway between cue and top_right pocket. Also fixes per-worker episode tracking (was only worker 0). Job 7049627, torch, 17m45s wall, ran 2026-04-24 16:18-16:35 local. VERDICT: FAIL on go/no-go BUT stages 0 and 1 both cleared BEFORE step 50k (much faster than canary 7011838 which took 300k+ steps to advance 1->2). Stage-2 performance peaks at 1.94 (vs random 2.10) but mean is 0.82 — still below random. Policy std still drifting UP (1.006 -> 1.059) — the fundamental gradient-starvation issue persists. Curriculum bootstrap works for getting through stages but doesn't fix the underlying entropy-bonus-wins-over-signal problem.

Dataset Info

• Rows: 10
• Columns: 9

Columns

Column	Type	Description
step	Value('int64')	Global PPO timestep
curriculum_stage	Value('int64')	Stage at this step. Stage 0/1 completed pre-step-50k; all evals are stage 2.
mean_score	Value('float64')	Snooker points scored, mean over 16 eps
max_score	Value('float64')	Best episode score in the eval
mean_shots	Value('float64')	Mean shots per episode (200 = truncated)
mean_efficiency	Value('float64')	mean_score / mean_shots
mean_foul_rate	Value('float64')	Fraction of shots that were fouls (0-1). Real fouls, parser-fix already applied.
episodes	Value('int64')	16 eval episodes per eval
source_file	Value('string')	Original metrics_step_*.json

Generation Parameters

{
  "script_name": "sbatch/train_sim_canary.sbatch (stage-0 curriculum mode)",
  "model": "stable-baselines3 PPO, MlpPolicy net_arch=[256,256]",
  "description": "Phase-3 Option-B canary: stage-0 'sure-shot' curriculum bootstrap. Curriculum stages [0,1,2,3,4] where stage 0 places 1 red midway between cue and top_right pocket. Also fixes per-worker episode tracking (was only worker 0). Job 7049627, torch, 17m45s wall, ran 2026-04-24 16:18-16:35 local. VERDICT: FAIL on go/no-go BUT stages 0 and 1 both cleared BEFORE step 50k (much faster than canary 7011838 which took 300k+ steps to advance 1->2). Stage-2 performance peaks at 1.94 (vs random 2.10) but mean is 0.82 \u2014 still below random. Policy std still drifting UP (1.006 -> 1.059) \u2014 the fundamental gradient-starvation issue persists. Curriculum bootstrap works for getting through stages but doesn't fix the underlying entropy-bonus-wins-over-signal problem.",
  "hyperparameters": {
    "algorithm": "PPO",
    "total_timesteps": 500000,
    "n_envs": 8,
    "n_steps": 512,
    "batch_size": 2048,
    "learning_rate": 0.0003,
    "ent_coef": 0.01,
    "gamma": 0.99,
    "curriculum_stages": [
      0,
      1,
      2,
      3,
      4
    ],
    "advancement_threshold": -400.0,
    "curriculum_min_episodes_per_stage": 100,
    "curriculum_window_size": 50,
    "reward_shot_cost": 0.01,
    "reward_pot_bonus": 0.5,
    "reward_completion_bonus": 10.0,
    "reward_position_shaping": 0.0,
    "eval_episodes": 16,
    "code_version": "2026-04-23-phase2",
    "commit": "a5021a0"
  },
  "input_datasets": [],
  "experiment_name": "snooker-testbed",
  "job_id": "torch:7049627",
  "cluster": "torch",
  "artifact_status": "final",
  "canary": true,
  "final_sidecar": {
    "step": 500000,
    "std": 1.058772087097168,
    "code_version": "2026-04-23-phase2",
    "curriculum_stage": 2,
    "curriculum_stage_idx": 2,
    "episodes_in_stage": 2309,
    "recent_episode_rewards": [
      -772.9999999999985,
      -789.4999999999986,
      -746.4999999999985,
      -707.4999999999986,
      -765.9999999999986,
      -731.4999999999986,
      -738.9999999999985,
      -755.9999999999986,
      -748.4999999999985,
      -805.4999999999985,
      -736.9999999999986,
      -759.9999999999985,
      -757.9999999999986,
      -757.4999999999986,
      -796.9999999999986,
      -717.9999999999986,
      -742.4999999999986,
      -806.9999999999986,
      -750.4999999999986,
      -732.4999999999986,
      -785.9999999999986,
      -767.4999999999985,
      -751.9999999999986,
      -700.4999999999986,
      -756.9999999999986,
      -808.9999999999986,
      -752.9999999999985,
      -707.4999999999986,
      -782.4999999999985,
      -742.4999999999985,
      -748.9999999999985,
      -796.9999999999985,
      -736.9999999999985,
      -765.9999999999986,
      -722.9999999999985,
      -751.4999999999985,
      -729.9999999999985,
      -721.4999999999986,
      -694.9999999999986,
      -735.9999999999986,
      -731.9999999999985,
      -772.4999999999985,
      -708.9999999999986,
      -735.4999999999986,
      -767.4999999999986,
      -743.9999999999986,
      -698.9999999999986,
      -725.4999999999986,
      -713.4999999999986,
      -729.4999999999986
    ]
  }
}

Experiment Documentation

For complete experiment details, see https://github.com/aditijc/snooker-testbed

Usage

from datasets import load_dataset

dataset = load_dataset("aditijc/snooker-testbed-canary-7049627-stage0-v1", split="train")
print(f"Loaded {len(dataset)} rows")

---