Nhoryun-Omega-III / DATA_DICTIONARY.md
PureOne's picture
Upload Nhoryun Omega III research dataset release 3.0.0
868c741 verified
|
Raw History Blame Contribute Delete
3.69 kB
# Structured dataset dictionary
All subsets are UTF-8 JSON Lines: one record per line. Every subset uses the organizational split name `train`. There is no hold-out partition. All source paths are relative to the dataset repository root.
| Configuration | Primary fields | Meaning |
| --- | --- | --- |
| `claims` | `id`, `title`, `status`, `statement` | Claim identity and exact source-ledger wording. `proved_in_manuscript` describes the release's evidence status; it is not an external certification. |
| `claims` | `assumptions`, `proof_locations`, `verification_evidence`, `limits`, `relation_to_prior_work`, `source_path` | Lists of assumptions, chapter path plus theorem label, audit paths, explicit limits, attribution, and source ledger. Empty values mean the field was absent in the ledger. |
| `manuscript_sections` | `id`, `source_path`, `format`, `sha256`, `text` | Complete chapter text in LaTeX, copied verbatim. Supporting preamble, bibliography, and figures remain in `research/`. |
| `witness_cases` | `m`, `a_exact`, `strong_convexity_exact`, `step_exact`, `condition_number_exact` | Declared finite-run construction parameters. Fraction-valued exact quantities are strings, and counts are integers. |
| `witness_cases` | `rational_curvature_margin_exact`, `analytic_curvature_margin_exact`, `gaussian_sigma_formula`, `gradient_tail_bound_exact`, `value_tail_bound_exact` | Audited rational margins and analytic-transfer bounds. `gaussian_sigma_formula` is a symbolic expression, not a measured integral. |
| `witness_cases` | `number_negative_second_differences`, `largest_second_difference_display`, `first_decrement_display`, `cumulative_gain_display`, `trajectory_sha256`, `all_exact_checks_passed`, `source_path`, `evidence_kind` | Original verification results. Fields ending in `_display` are decimal approximations of the unsmoothed witness. |
| `audit_runs` | `script`, `script_sha256`, `exit_code`, `passed`, `elapsed_seconds`, `report_path`, `log_path`, `source_path` | Original suite records. Timing belongs to the source run and is not a hardware benchmark. |
| `trajectory_m32` | `m`, `k`, `objective_variant`, `x0_exact`, `x1_exact`, `loss_exact` | Exact rational states and losses, regenerated by the included witness implementation for `m=32`. The objective variant is `unsmoothed_rational_witness`. |
| `trajectory_m32` | `decrement_exact`, `second_difference_exact`, `loss_decimal_approx`, `decrement_decimal_approx`, `second_difference_decimal_approx`, `source_path` | Consecutive loss differences and display values. The final state has no next decrement, and the final two states have no next second difference; unavailable entries are JSON `null`. |
| `files` | `path`, `size_bytes`, `sha256`, `category`, `media_type` | Index of distributed payload files. The manifest itself and outer `SHA256SUMS` are excluded from the row index to avoid self-reference. Outer `SHA256SUMS` covers the manifest too. |
The 34 trajectory rows contain 33 decrements and 32 strictly negative second differences. Exact strings are parsed with `fractions.Fraction(value)` in Python. A decimal approximation should not be substituted for its exact rational counterpart in a proof check.
The `research/data/witness_m32_plot_data.csv` source is preserved separately. It contains rounded display values for the same unsmoothed witness. No JSONL table evaluates the Gaussian-convolved analytic objective numerically.
`dataset_metadata.json` records subset counts, release identifiers, PDF checksum, rights status, and the index-generation entrypoint. `research/metadata/research_status.json` remains authoritative for mathematical assumptions and evidence boundaries.