Context
ordgraph-proto/scripts/probe_smoke.py and parts of probe_query_doc.py duplicate ordinal kernel logic in Python:
- rank bucketization
- contingency table construction
- RankQuant symmetric score
- band agreement
- top-group overlap
- bucket L1 distance
- coarsened/quantized contingency projections
That logic belongs with ordvec so probes and downstream harnesses do not fork kernel semantics.
Acceptance criteria
- Add Python and/or CLI utilities that emit the same pair-evidence projections currently used by OrdGraph probes.
- Include golden/parity fixtures against current
ordgraph-proto probe output.
- Support clean downstream use by OrdGraph and OrdinalDB eval harnesses.
- Keep corpus-specific logic out of
ordvec; this should be kernel/projection generation only.
Non-goals
- No graph/event/falsifier semantics in
ordvec.
- No arXiv/enterprise fixture ownership in
ordvec.
Draft source: /home/ndspence/Downloads/ordgraph_extraction_issue_queue.md
Context
ordgraph-proto/scripts/probe_smoke.pyand parts ofprobe_query_doc.pyduplicate ordinal kernel logic in Python:That logic belongs with
ordvecso probes and downstream harnesses do not fork kernel semantics.Acceptance criteria
ordgraph-protoprobe output.ordvec; this should be kernel/projection generation only.Non-goals
ordvec.ordvec.Draft source:
/home/ndspence/Downloads/ordgraph_extraction_issue_queue.md