Datasets:
Structured release indexes
All three configurations use UTF-8 JSON Lines and the split name index. They are inventories and summaries of the supplied release, not new observations, human annotations, training pairs, or independent benchmark tasks. No training/validation/test partition is implied.
The original manuscript, claim ledger, raw records, and protocols remain authoritative. Repeated metrics are not independent samples. Use per-seed and per-lineage raw records for statistical analysis.
claims: data/claims.jsonl
| Field | Type | Meaning |
|---|---|---|
claim_id |
string | Stable packaging index identifier |
topic |
string | Short research topic |
statement |
string | Scoped summary of a result or nonclaim |
evidence_kind |
string | Proof, recorded check, measurement, inherited evidence, or unestablished status |
scope |
string | Assumptions and limits needed to interpret the statement |
source_path |
string | Primary source file relative to repository root |
Claim summaries were prepared from the supplied ledger and result files. A proved label refers to a written proof under model assumptions, not proof-assistant verification or a historical novelty determination. Counts of index rows are not counts of discoveries.
metrics: data/metrics.jsonl
| Field | Type | Meaning |
|---|---|---|
metric_id |
string | Stable metric identifier |
study |
string | Research component |
metric |
string | Human-readable scalar name |
value |
float64-compatible number | Selected numeric value copied from source JSON |
unit |
string | Count, seconds, normalized utility, ratio, or proportion |
evidence_kind |
string | Recorded exact count, measurement, or inherited descriptive statistic |
scope |
string | Context and qualifications |
source_path |
string | Original JSON file relative to repository root |
json_pointer |
string | RFC 6901 pointer to the original scalar |
Values have not been pooled across unrelated studies. Original integer values are stored as JSON numbers; all listed counts are small enough to be represented exactly as binary64 values. Bootstrap intervals are descriptive, inherited, and not adjusted for multiple comparisons. Metric summaries do not convert utility differences into intelligence percentages.
artifacts: data/artifacts.jsonl
| Field | Type | Meaning |
|---|---|---|
path |
string | Original research file path, relative to repository root |
bytes |
integer | Exact file byte size |
sha256 |
string | SHA-256 of those file bytes |
role |
string | Navigational role inferred from its top-level research directory |
The index covers the preserved research/ files, including their own original manifest. It does not recursively unpack nested ZIPs or count their internal records as separate rows. Newly added Hub packaging files are covered by the root manifest instead of this index.
Provenance and license
Source identity and source archive hash appear in packaging/SOURCE_PROVENANCE.json. The source data are generated research records, and the index summaries were created for this Hub edition. No new personal dataset or external corpus is added. Authorship and preparer credit reproduce the supplied release.
New packaging descriptions and derived index material are offered under CC BY 4.0 to the extent applicable rights exist; packaging scripts are MIT. Original material retains the notices described in LICENSE.md.