MED01-P2 — Kickoff prompts (Stage 1 and Stage 2)

Status: v1 · 2026-10-03 · written at P2 freeze (core brief v1.0 + Probe Appendix v1.0, FROZEN 2026-10-03). Built on the approved P1 kickoff (kickoff-prompts.md, v3). Git only; not mirrored to Drive.

How to launch (Danny)

1. Same prompts, word for word, to both researchers (§M1). Any edit applies to both or to neither.
2. Clean session for each. Not the editorial-review chats, and not a chat that inherits project context or memory:
   • ChatGPT: a chat outside the project, as a Temporary Chat or with memory / “reference chat history” off.
   • Gemini: a new chat with saved info / personal context off; not a project Gem. Never reuse Gemini’s P2 editorial chat — it holds illustrative cues from its review.
3. Stage 1 attachments (from git, not Drive exports): medical/MED01/brief-P2.md (core v1.0) and medical/PROTOCOL-MEDICAL.md (v0.5). Do not attach the Probe Appendix yet.
4. Web search / browsing on.
5. Stage 1 filing. Paste the complete Stage 1 output here. Claude files it verbatim in raw-inputs/<model>/ (git only). If a reply is cut off, ask for a continuation in the same chat; it is part of the same submission.
6. Stage 2. Only after Claude confirms that researcher’s Stage 1 is committed: in the same chat, send the Stage 2 prompt below with medical/MED01/brief-P2-probes.md (appendix v1.0) attached. Paste the addendum here; Claude files it separately.
7. Firewall: nothing from either researcher reaches the other, or Drive, until both researchers’ Stage 1 and Stage 2 are filed. Then both passes are mirrored together, and the P2 archive audit and editorial dispositions may follow.

Stage 1 prompt (identical for ChatGPT and Gemini)

You are one of two independent researchers on MED01-P2, a historical evidence map of medical cannabis and cannabinoid research, 2000–2017. Another AI researcher is working from the same frozen brief. You will not see each other's work, and you should not try to anticipate it.

YOUR TASK SPECIFICATION is the attached file "brief-P2.md" — MED01-P2 core brief v1.0, FROZEN 2026-10-03. The second attachment, "PROTOCOL-MEDICAL.md" (v0.5), is the rulebook the brief cites by section. Read the whole brief before you begin. It is authoritative: where this message and the brief differ, the brief wins.

This is STAGE 1 of a two-stage pass (brief §0a). Answer brief §1–§5 only. A second, separate set of questions will be given to you in this chat after your Stage 1 output has been filed; do not try to anticipate it.

CONTEXT ISOLATION. Treat the attached frozen brief as the sole substantive specification. Do not rely on, reconstruct, or import conclusions from any prior conversation about MED01, its first part (1960–1999), the author and researcher's previous writing, archive material, editorial reviews of this brief, expected milestones, or another model's work. If any such prior context is visible to you, disregard it when selecting findings and sources, and say at the top of your pass that it was visible. Do not open or request any other project file or shared drive. Build the evidence map independently, from the literature and the frozen brief.

Key points, all from the brief:
- EVIDENCE MAP of a literature, not an efficacy review. No present-day efficacy or safety verdicts, and no advice.
- Window: primary sources published 2000-01-01 to 2017-12-31, assigned by earliest public publication date; boundary cases flagged; processes allocated milestone by milestone (§0c).
- Sources marked period / prior / retrospective (rule 6a), or with the evolving-record fields (rule 6b). Retrospective sources are searched through the MED01-wide cutoff of 2026-09-20; what was KNOWN by end-2017 must trace to period evidence.
- Full extraction only for studies directly supporting a clinical or harms candidate claim, or that you designate keystone; survey lines for the rest; gap codes, never blanks (rule 3).
- Exact product, formulation and route every time (rule 5a). Harms with the same rigor as benefits (rule 7).
- Open every identifier live and copy the resolved title; never complete a DOI, PMID or registry number from memory (rules 10, 10a). Quote only by copy from a source opened in this session (rule 11).
- The four canonical PubMed counts use the exact term and URL pattern in §4a, changing only the dates.
- Candidate claims normally 20–30, hard maximum 35 — a ceiling, not a target.
- "None identified in my searches" is never "none exists."
- Finish with self-flagged weak points and the full search log (§4a).

If a platform output limit prevents delivery in one message, continue immediately in consecutively labeled parts without waiting for user input. Together those messages constitute one frozen Stage 1 submission. Do not revise earlier parts while continuing.

Deliver ONE complete Stage 1 submission, as text in this chat, in the output format of brief §5. State at the top which brief version you worked from, which model you are, the date(s) run, and whether live browsing was available.

Stage 2 prompt (identical for ChatGPT and Gemini — same chat, after Stage 1 is filed)

Your Stage 1 submission has been filed and is now frozen as received.

STAGE 2. The attached file "brief-P2-probes.md" is the MED01-P2 Probe Appendix v1.0, FROZEN 2026-10-03, frozen together with the core brief. All rules of the core brief continue to apply (boundaries §0c, evidence standards §4, search logging §4a, context isolation).

Write a separate PROBE ADDENDUM in the output format at the end of the appendix:
- Do NOT revise, restate or correct your Stage 1 submission. If the probes show your Stage 1 map missed something, report it under "Omissions in my Stage 1 map".
- The probes are audit probes, not a statement of what matters. Answer them; do not re-weight your Stage 1 map around them.
- "Which / each / every" means every instance identified in your documented searches — state your search boundary.
- Public-claim labels are plain-language probe summaries, not protocol dispositions; report period evidence first and any later reinterpretation separately and dated.
- Optional extra candidates as PCAND-01 … (at most 8).

Same continuation rule as Stage 1 if you hit an output limit. Deliver ONE complete addendum as text in this chat, stating at the top which appendix version you worked from.