Skip to content

Latest commit

 

History

History
47 lines (39 loc) · 2.58 KB

File metadata and controls

47 lines (39 loc) · 2.58 KB

English data readiness

Measure Current development pack
Data version 2026.07.3
Compiler-accepted answers 1,000
Editorially approved answers 1,000
Senses 1,000
English draft clues 2,000
Clues per sense exactly 2
Answers with English clue coverage 1,000 (100%)
Cell-length range 3–10
Catalog size 3,616,768 bytes
Solver index size 483,321 bytes

The pack exists to verify deterministic compilation and the real runtime path at the first scale milestone. Its thematic vocabulary, answer-specific definitions, and semantic clues passed Crosseno editorial review on 2026-07-26. The source also passed schema, eligibility, identity, and normalization-collision checks with zero compiler rejections.

Automated checks report zero direct normalized-leakage failures, zero exact clue duplicates, and 2,000 unique clue texts. Each clue pair also passes a token-similarity threshold and a metadata-template rejection check. These checks establish draft quality, not factual or editorial approval. See resources/draft-quality-report.json.

The 1,000 answers are distributed by cell count as follows: 60 of length 3, 245 of length 4, 255 of length 5, 235 of length 6, 128 of length 7, 51 of length 8, 17 of length 9, and 9 of length 10. A supported generation envelope still requires representative benchmark results; individual dimensions and seeds can be unsatisfiable.

The current deterministic matrix covers 5×5, 7×7, and 9×9 dimensions; fast and balanced strategies; all three difficulty levels; and seeds 1, 42, and 12345. All 54 requests succeeded and passed the example’s repeated-build reproduction check. Three seeds per group demonstrate integration compatibility but are not enough to claim a general production success rate. Detailed entry and explored-node totals are in resources/benchmark-report.json.

Targets:

  1. Development release: completed in 0.2.1, including contractual, legal, packaging, artifact, and clean-installation checks for the reviewed 1,000-answer pack.
  2. Preview: 25,000 answers, two clues for at least 90% of senses, representative dialects, and published benchmark results.
  3. Production: 100,000 reviewed eligible answers, two clues for at least 95% of senses, stable supported-size/strategy success gates, full provenance and redistribution review, artifact-size acceptance, and an installation/support matrix.

Counts alone never promote readiness. Review quality, clue leakage, difficulty calibration, dialect behavior, licensing, reproducibility, and generation benchmarks are mandatory acceptance inputs.