cite-review

Goal

Every \cite{prop_N} in the tex must faithfully represent proposition N on two axes:

• Content — the sentence asserts what the proposition's content says. Stronger, weaker, or categorically different is a citation error.
• Strength — the sentence's confidence matches the proposition's qualifier. Asserting flatly what the graph holds at possibly is a citation error even when the content matches word for word.

The second axis exists because the qualifier is a single credibility scale: the graph now states how much it believes each proposition, so prose can contradict it without misquoting it.

Done means: every citation reviewed on both axes; every mismatch resolved by (a) a corrected proposition plus a repointed citation, (b) rewritten or removed LaTeX, or (c) a qualifier change you actually judged and can defend. Consequences this skill cannot close on its own — a new proposition left unestablished, a conclusion whose evidence you removed, an unjudged rebuttal — are named as open gaps, not silently left.

Phase 1 — Extract In Resident Context

Identify the .tex file(s). If the user did not specify a path, find .tex files in the project directory.

Scan the tex for every \cite{...} containing prop_N keys. A single \cite{} may carry multiple keys, e.g. \cite{prop_1, prop_3}. For each key in each cite, emit one tuple (proposition_id, file_path, line_number) — a flat list, one entry per proposition per cite occurrence. Do not read surrounding text, call get_argument, or open any PDF here. Hand the list to Phase 2.

Legacy statement_N keys are stale, not renamed. They point into the retired Statement ontology, whose graph is archived read-only. Never assume statement_7 means prop_7 — the number would silently land on an unrelated proposition. Re-identify what each legacy key was citing, find or create the proposition that says it, and repoint the key. Report any key you cannot re-identify instead of guessing.