deepresearch-searchlisted

# Deep Research Search (Stage 2b) Takes the Stage-0 brief and uses the **Gemini Deep Research Agent** to autonomously plan, search, read, and synthesize the prior literature, then parses the resulting cited report into `<stem>.json` + `<stem>.bib` for dedup and screening. It complements the other deep searches: Undermind and Scholar Labs are browser-driven; this one is a pure API call. The stage is two halves: - **`scripts/deepresearch_search.py`** — calls the Interactions API (`deep-research-max-preview-04-2026` by default), runs the task in the background, polls to completion, and saves the report + raw response. - **`scripts/deepresearch_ingest.py`** — UI-independent parsing. It reads the agent's `KEY PAPERS` section (and, as a fallback, citations in the raw response), best-effort-enriches via Crossref, and writes the pipeline JSON. Importable, and runnable standalone on a saved report. ## The flow `brief` → wrapped into a literature-review prompt that ends with a parseable `KEY PAPERS` section (`Title | Authors | Year | Venue | DOI or URL`, one per line) → `POST /v1beta/interactions` with `background=true`, `store=true`, agent `deep-research-max-preview-04-2026` → poll `GET /v1beta/interactions/{id}` until `completed` → save `deepresearch_report.md` + `deepresearch_raw.json` → parse the `KEY PAPERS` lines → enrich → `stage2b_deepresearch.json` (+ `.bib`). Deep Research is asynchronous and takes minutes (the API caps a task at 60 minutes; most finish under