Delete ARC-GEN-100K.zip

Upload ARC-GEN-100K.zip

Move all solver files to own-solver/ — keep .gitattributes and notebooks at root

Add total_cost to summary stats

Wave 2+flood fill scan: 0 matches. Update LEARNING with findings.

Wave 2 scan: 0 matches for composition chains, mode extensions, bg fill. Tasks are harder than expected.

Update registry with Wave 1 solvers

Add Wave 1 solvers: downsample_stride, symmetry_complete, extract_inner, add_border, sparse_fill, channel_filter

Rewrite Phase 3: merged expert + original solvers, organized by architecture type, honest estimates

Fix: --strict_score default False (warnings only), --strict_size default True (halt on oversized)

Fix: delete leftover .onnx on unsolved tasks, --strict_score default False

Fix 4: main.py — add --strict_score, stop if any model unscorable

Fix 3: solver_registry — score_network gate after validate, reject if (None,None,None)

Fix 2: validators.py — add onnx.checker, banned op scan (UPPERCASE), static shape check

Fix 1: profiler.py — remove silent fallback, loud warning when onnx_tool missing

LEARNING.md: log profiler.py silent fallback mistake + Kaggle rejection + fix plan

main.py: add --strict_size (default True) — stops entire run if any .onnx > 1.44MB

solver_registry: check file size after onnx.save, before validate — skip oversized immediately

Add file size check: reject oversized .onnx before validation, --strict_size stops run on violation

Fix: unsolved_count × 0 in est_lb — only count solved score

Fix SKILL.md: correct constraints (1.44MB per onnx file, Script not If, all static shapes required)

Fix submission.py: per-onnx-file size check, not zip; remove excluded task skip"

Fix BANNED_OPS (Script not If) and clarify MAX_FILESIZE is per .onnx file

Update SKILL.md: v5.2 structure, new solvers, no excluded tasks, current scores

set unsolved_count score =0

Remove new_solvers.py (split into gravity.py, edge.py, mode.py)

Update registry imports: gravity.py, edge.py, mode.py (replaces new_solvers.py)

Split: mode fill solver into mode.py

Split: edge detection solver into edge.py

Split new_solvers.py into gravity.py, edge.py, mode.py

Wire new solvers into registry (gravity, edge, mode_fill)

Add gravity, edge detect, mode fill solvers (v5.2)

Rewrite Phase 3 with research-backed blueprints + honest projections

Clean up main.py: remove excluded tasks messaging (all 400 count)

Remove fake task exclusions — all 400 tasks count, unsolved=0

Update LEARNING.md with Exp 3 PCA/SVD full results + v5.1 entry

Update TODO.md with Exp 3 full PCA/SVD results

Exp 3: PCA/Truncated SVD before lstsq — implemented, tested, 0 new solves

Upload TODO.md with huggingface_hub

Upload TODO.md

Update TODO.md: evidence-based Phase 2 overhaul with confidence ratings, literature citations, coordinated experiment sequence

Delete neurogolf_solver.py

Update LEARNING.md for v5 refactor + new entries

Update SKILL.md for v5 refactored package

Update README.md for v5 refactored package

v5 refactor: add main.py entry point (with W&B init)

v5 refactor: add submission.py (with W&B logging)

v5 refactor: add solvers/conv.py (with lstsq crash fix)

v5 refactor: add solvers/solver_registry.py

v5 refactor: add solvers/tiling.py