Type Alias QuestionRunnerOptions

Feed the classified date intent into the question planner (advisory) and compute that classification ONCE per question, threaded into the answer stage. AND-gated on enableCalendarAwareness at the runner — no effect unless calendar awareness is also on.

Enrich the document-selection listing with each chart's circle_metadata (level-of-care / program-stay window) so the LLM picks documents with clinical context in view. Pure enrichment — never narrows the candidate set, so it can't starve retrieval of the document holding the answer.

Hydrate the retrieval query with clinical synonyms ('synonym' mode) instead of the raw question ('none'), so chart phrasings that differ from the question (e.g. "denies SI/HI" vs "suicide risk") still match.

Append the template's review criteria to the document-selection prompt so the LLM picks documents relevant to the rubric, not just the question text.

After the answer agent emits a pass/fail status, run a cheap single-shot reconciler (reconcileStatus) over the question + answer prose + citations and OVERWRITE status with the value the answer actually supports. Guards against the answer LLM stamping a label that contradicts its own prose. Off = the emitted status is returned verbatim (today's behavior).

Abstain ("insufficient evidence") instead of invoking the answer agent when retrieval returns no usable context — bounds the damage of a recall miss rather than letting the agent fabricate.

Drop the hard file_name $in Pinecone filter so semantic retrieval isn't blinded by a wrong/narrow title-based document pick.

Reviewer-selected charts; when set they are the entire candidate set.

Shared reference instant for all date reasoning (prod = now, eval = frozen).

The id of the session whose note triggered this run. Used by SessionNoteAuditRunner to attribute sources/citations to the session so the UI can deep-link to it.

When set, the runner skips EMR/patient_charts + Pinecone retrieval and uses ONLY this (already HTML-cleaned) session-note text as the agent context. Set by the session-note trigger path; unset for patient-level.

When set, scope the run to a single treatment episode: the candidate charts the EMR runner considers are filtered to this episode, so charts from other episodes (and unassigned charts) are never reviewed. Unset for unscoped patient-level reviews.