# rdfexport arrow-classifier parity report (2025-10-22T19:08:58Z)

## Summary
- Ported the original DrawIO arrow handling into the override so non-strict edges, label fallbacks, and literal detection mirror the historical Python implementation.
- Regenerated `legacy/draw_io_parser.py` via meta builder to pick up the refreshed override bundle.
- Updated debug fixtures (`debug/map.json`, AA37 preserve-html scenario) and CLI tests to reflect the corrected arrow classifications.
- Added pytest coverage under `legacy/tests/test_cell_classifier.py` to enforce parity with the legacy parser and to assert strict-mode failures are raised.

## Testing
- `bun run check`
- `python -m debug --scenario knut-olborgs-forskningsnotater`
- `python -m debug --scenario rg-18-210-walkerton-inquiry-in-ric-o-original`
- `bun run test:log:linux`

Known failures remain limited to the longstanding baseline issues (DrawIO metadata strip-html overrides, invalid namespace fixtures, metadata fixture regeneration) and the Bun integration test `runDrawioPipeline preserves literal HTML when stripHtml disabled`. No new regressions were introduced.