# Codex Report — 2025-10-21 22:35 UTC

## Summary
- Investigated literal vs individual classification for AA37 "lolabout" node and confirmed the DrawIO cell classifier already marked it as a literal.
- Updated the RDF serialisation override to preserve literal/object tuples and to coerce relative/invalid prefix IRIs back to the configured base namespace.
- Regenerated the legacy parser via `bun run build:py` so generated code matches the override changes.
- Extended pytest coverage: one test ensures the literal survives end-to-end and another confirms invalid prefixes fall back to the base namespace.
- Reformatted debug metadata (`debug/map.json`) with Prettier to capture the updated classifications without introducing `ns1:` URIs.
- Added an Unreleased changelog entry documenting the fix.

## Tests
- `uv run pytest legacy/tests/test_debug_cli_regression.py` (fails in known areas: missing rdf:type `picoL:j`, properties absent; upstream regression)
- `bun run check`
- `bun run test:log:linux` (expected failures logged; 6 failing fixtures, 23 pass, 8 skip)

## Follow-ups / Notes
- Numerous `uv run pytest` and `bun run test:log:linux` failures correspond to pre-existing baseline mismatches; no additional remediation attempted in this pass.
- Debug TTL fixtures regenerated by the regression harness were reverted to avoid unrelated churn.
- Ensure future work considers adding targeted regression coverage for mixed object/datatype properties to avoid regressions of the tuple storage logic.