- **kg_llm_extractor rewrites AGE dollar-quote tag in triples so drawers indexing palace source code don't fail at add_triple (#313)** ([`HEAD`](https://github.com/techempower-org/mempalace/commit/HEAD))

Drawers indexing palace-daemon / mempalace source code contain

``mp_age_q`` verbatim (the AGE dollar-quote tag the cypher

wrapper uses to delimit its outer SQL literal). When the LLM

extracted triples about those drawers — e.g. subject="LINE 1",

predicate="select", object="* FROM cypher($1, $mp_age_q$" — the

triple's text carried the tag substring straight to

``_cypher_literal``, which raised ``Cypher literal contains the

AGE dollar-quote tag 'mp_age_q'; reject upstream in the

sanitizer.`` The drawer's KG presence was incomplete and the

worker logged a warning per failed write.

The error message itself flagged the fix location ("reject

upstream in the sanitizer"). This change adds a substring

rewrite at ``_validate`` — the boundary where the LLM hands

triples back — replacing ``mp_age_q`` with ``MP_AGE_Q_LIT``.

The case-sensitive substring check in ``_cypher_literal`` no

longer fires on the rewritten value (Python's ``in`` operator

is case-sensitive). Predicate-normalization happens before the

rewrite, so the lowercased predicate's tag substring is the

one that gets caught and the final predicate ends up with the

upper-case placeholder — readable, KG-queryable, safe.

The roundtrip test (``test_validate_output_survives_cypher_literal``)

proves the end-to-end fix: a triple whose object is the verbatim

``_AGE_DQ_TAG = "mp_age_q"`` source line now passes through

``_cypher_literal`` for all three fields without raising.

*Tests:* 6 — tests/test_kg_extractor.py (rewrites subject + object + predicate, leaves clean triples unchanged, handles multiple occurrences, end-to-end roundtrip survives _cypher_literal)

*Files:* `mempalace/kg_llm_extractor.py`, `tests/test_kg_extractor.py`