ARC-AGI × MDLxDCC Arena

What this page is — and is not

This is a private research prototype for failure mining, primitive discovery, routing, and MDL/DCC diagnostics. It is not an ARC Prize submission, not an official benchmark claim, and not an exposure of proprietary MDLxDCC core internals.

The useful measurements here are train-fit exacts, HIGH confidence, LOO survival, new primitive-family wins, near-miss structure, exact per runtime, and whether the arena learns where to spend DCC budget.

Visual evidence layer

The local arc_report.html pages are valuable because they show train input, target, winner prediction, exact/near-miss status, categories, candidate counts, and failure buckets. Instead of embedding every run report, the v0.4 package should carry a compact atlas plus links back to per-run reports.

Small preview from selected exacts

Package rule: place the generated key bundle beside this page as arc_v04_key_bundle/. Keep the huge original out_arc_v04_overnight_w10 outside the MDLxDCC package unless needed locally.

Completed v0.4 overnight findings

Top runs by exact / HIGH / LOO

Best ROI routes

Chain2: weak globally, useful when routed

The 50-task global baseline with chain2 and the no-chain2 ablation both produced 4 exact / 3 HIGH. The no-chain2 run used roughly half the runtime. That keeps chain2 out of the global default.

But routed chain2 produced 5 exact rows across 5 unique tasks, with 3 robust LOO=1.0 rows. The useful cases are mostly color/morph/crop compositions such as line_connect → component_recolor or fill_holes → global_color_replace.

Primitive-family signal

Row counts are not unique-task counts because the same task can appear in several category slices. Still, the repeated winners are informative: tile_repeat, fill_holes, substitution_expand, line_connect, recolor_by_map, wire_connect, scale_up_integer, crop_largest_object, and extract_object are the current strongest families.

First-discovery progression

Across the 40-run bundle, v0.4 finds 39 unique exact train-fit task ids; 33 have best LOO=1.0. Baseline global runs found 4; the rest came from slicing and routing.

Failure buckets and runtime bombs

needs_depth2_composition is too broad. It appears almost everywhere and must be split into object matching, macro-grid, line completion, marker instruction, crop/resize, color-role transfer, local cellular rule, and canvas transformation buckets. identity and global_color_replace near-misses are also too noisy and should be treated as diagnostic hints, not strong routes.

v0.5 update plan

• Memory: add arc_az.json and optional task cache storing task signature hash, categories, winning families, failed families, near-misses, runtime cost, and best candidate summaries.
• Routed budget: add category evidence scores, --cat-report, and --chain2-policy routed; skip expensive lanes when shape evidence says impossible.
• Better categories: keep wide tags, but add confidence/reasons and support combinations such as --cat line,color --cat-mode all and not:slow_risk.
• Failure mining: split the broad buckets and add exact_but_loo_bad, identity_fallback_suspect, and near_miss_small_rule_gap subtypes.
• Primitive pack: prioritize macro-grid cell rules, marker rays, object matching, symmetry completion, crop-between-markers, color-role fill, counting, line-gap bridge, local cellular rules, and object motion/contact.
• Visual reporting: make the compact atlas a first-class artifact; do not ship every massive arc_report.html inside the MDLxDCC package.

Reproducibility and artifact layout

Canonical source files remain the arena code, batch file, summarizer, and JSON outputs. The compact key bundle is the MDLxDCC-facing artifact.

python arc_mdlxdcc_arena_v0_4.py --data-dir D:\8Z\8z\ARC\ARC-AGI-2 --split dev --max-tasks 50 --outdir out_arc_v04_quick_w10 --mode both --fresh --budget wide --workers 10 --progress-every 5
python summarize_arc_overnight.py out_arc_v04_overnight_w10
python arc_collect_keydata_v04.py out_arc_v04_overnight_w10 --out out_arc_v04_keybundle

Expected local package link target: arc_v04_key_bundle/arc_visual_atlas.html, arc_v04_key_bundle/arc_report_hub.html, arc_v04_key_bundle/run_index.csv, and arc_v04_key_bundle/task_index.csv.