/tor-procurement — build, review and grade procurement ToRs

One job: get a procurement ToR to open-market standard. The spine is references/open-market-checklist.md — 22 items that took the AI-for-Research ToR four revision rounds (v0.4→v0.7, July 2026) to reach by hand. A new ToR starts at that standard; an existing ToR is measured against it.

Mode routing

• build — no ToR exists yet, or Ane wants a fresh start.
• review — a ToR exists; Ane wants strong/weak points and a grade. Read-only.
• finalise — a review happened or Ane brings agreed revisions; apply them, nothing else.

If the mode is ambiguous, ask in one line before working.

Shared rules — all modes

• Read references/open-market-checklist.md before touching the ToR.
• Factual reliability. Never invent a budget figure, name, date, address, or funding source. Anything unconfirmed goes in square brackets AND into the "Internal preparation note — REMOVE BEFORE PUBLICATION" strip list (checklist item 21). A plausible placeholder in a published ToR can be acted on as if true.
• Intake before building. Confirm the critical facts with Ane before constructing anything; working files go stale (session-efficiency protocol).
• A published ToR is external-facing: Tier 1 plain English, translatability test (bidders are often non-native readers), acronyms spelled on first use.
• When the ToR carries MEL content (indicators, evaluation design, learning questions), citation standards apply via /mel-framework-citation.