You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
No ingest path extracts equations, so find_equation's TED route never fires #550
Neither run_bulk_ingest, run_delta nor tools/notebook_ingest.py calls step 1. On a corpus built the normal way there is no equations table, so find_equation always degrades to dense_only_fallback, and the TED route the eval gate (tests/eval/test_equations_parity.py) measures is unreachable in production.
Found while scoping #310's second half (#549), which builds the definitions index during notebook ingest but deliberately not equation trees, since there is nothing to build them from.
Suggested remedy
Run equation extraction per paper inside the ingest loop: it needs the same LaTeXML HTML the chunker reads. Then build the trees once at the end of a notebook ingest, as #549 does for definitions.
For the staging bulk path, the equations table lives inside the LanceDB dataset, so it is promoted by cutover with the chunks. Unlike theorem names, it has no cutover hazard.
Done when
A freshly ingested notebook has an equations table with trees.
find_equation on MathML input reports ted_fused on it.
A test drives an ingest end to end and checks both.
Activity