dm-bench 0.1.0

Torn-document reassembly benchmark. Synthetic, seeded, licence-clean pages are torn into non-overlapping fragments; a solver must place every fragment back on the page with a rigid pose. Full benchmark card, metrics and baseline: docs/BENCHMARK.md. Code: Arittra-Bag/Dataset-Maker.

Contents

tier	val pages	test-dev pages	test pages
easy	39	13	48
medium	37	16	50
hard	33	19	52

• puzzles/<tier>/<split>/<page_id>/: puzzle.json plus fragment PNGs (RGBA).
• answers/<tier>/{val,test-dev}/: ground-truth poses and adjacency.
• test answers are held out. They are derived from a maintainer secret; the commitment is in benchmark.json (test_commitment).
• SHA256SUMS (bytes) and CONTENT.sha256 (decoded content) cover every file. release_sha256 for this release: 6fe00ba8fac1e39b62cfa78a266095ba7e24c594d0fb2b5442b1148266b185ed.

The train split is not shipped; regenerate it with python -m src.bench build --splits train from the code repository.

Use

python -m src.bench verify <this folder>
python -m src.bench solve --release <this folder> --out solutions
python -m src.bench eval  --release <this folder> --solutions solutions --splits val test-dev

Scoring test needs the maintainer; send solutions for the test split.

Baseline (edge-greedy@0.1, eval_version 1.0, held-out test)

tier	direct_acc [95% CI]	neighbor_acc	perfect pages
easy	0.993 [0.981, 1.000]	0.994	0.958
medium	0.865 [0.820, 0.907]	0.851	0.300
hard	0.246 [0.224, 0.267]	0.640	0.000

Source: docs/results/dm-bench-0.1.0_edge-greedy-0.1_eval-1.0.json in the code repository.