Φ
ACTIVE MISSION

Playing ARC-AGI-3

Pure symbolic prediction against the hardest interactive reasoning benchmark in AI. No LLM. No GPU. No fine-tuning. Observe → predict → act. Against the live engine it completed 24 of 25 games and 176 levels — one game is still standing. The gym numbers below are the offline harness, kept separate on purpose.

i7-3630QMNo GPUNo LLM in Solve Loop~5 WattsSolo Researcher

Gym Benchmark

1,000 procedurally generated games · 5 core-knowledge families · 3 difficulty levels each · 0.6 seconds total runtime

0%
Solve Rate
0.000
Mean RHAE
0
Games Tested
0
Failures
0.0s
Total Runtime

Measured on internal gym (arc3_gym.py) — procedural puzzles matching ARC-3 protocol. Not the official ARC-AGI-3 leaderboard. Official eval pending benchmark release.

Family Breakdown

navigate
100%RHAE 1.000BFS Pathfinding

Agent→goal shortest path. Walls avoided. Computed in microseconds.

push
100%RHAE 0.896Composite BFS

BFS on (agent, block) state space. Finds minimum-move push sequence.

paint
100%RHAE 1.000Target Copy

Read target half → copy to work region. 16 clicks. Zero guessing.

lights
100%RHAE 0.666GF(2) Elimination

Lights Out → linear algebra. Gaussian elimination over GF(2). 2–3 clicks.

symmetry
100%RHAE 1.000Mirror Copy

Detect filled half → mirror column-by-column. 8 clicks.

The Paradigm

GRADIENT CHASING (before)

Click a colour. Did the score go up? No? Try another. Blind exploration exhausts the action budget. Navigate scraped to 52.5%. Everything else: 0%.

10.5% overall solve rate

THALAMIC PREDICTION (after)

Read the grid. Detect the puzzle type. Compute the exact solution. Execute in minimum actions. Zero guessing. Every move is predetermined.

100.0% overall solve rate

The shift: stop treating the environment as a black box. The answer is in the grid. Read it, compute it, paint it. Prediction over exploration. Measurement over iteration.

Solve Pipeline

1Observe

Read the raw grid. No preprocessing, no embeddings. Just the matrix of cell values.

2Detect

Universal family detection from grid structure alone. No labels, no hints from the environment.

3Predict

Compute the exact solution using the matched algorithm: BFS, GF(2), pattern-copy, or state-space search.

4Execute

Emit the minimum action sequence. Every click and move is predetermined before the first action fires.

Universal Family Detection

No labels from the environment. No hints. The classifier reads grid structure alone and routes to the correct solver.

Separator column (all-5)paint
All cells ∈ {1, 3}lights
100% dense half + zero halfsymmetry
1 agent + 1 block + 1 goalpush
1 agent + 1 goal + wallsnavigate

Architecture Components

arc3_predict_policy.py

Thalamic prediction policy. Observe → detect → solve → execute. Zero gradient chasing. Plans the full solution before the first action.

arc3_gym.py

Procedural game generator. Speaks exact ARC-3 FrameData/GameAction protocol. 5 families, graded 0–254 scoring, solvability-verified.

Family Detection

Universal classifier. Reads grid structure—separator columns, cell value distributions, entity counts, half-density. No family labels used.

Solver Suite

BFS pathfinding, composite (agent,block) BFS, target-copy, GF(2) Gaussian elimination, mirror-copy. Each solver is exact—zero trial-and-error.

The Jump

10.5%
Gradient Policy
Blind probing. Budget-limited.
100.0%
Predictive Policy
Exact solution. Zero failures.
ACTIVE DEVELOPMENT — OFFICIAL EVAL PENDING