FIX-7-3: calibration findings, defaults decision and docs

[mateuszwalo]

Track: research · Level: core · Effort: ~8h · Phase: 7 · Depends on: FIX-7-2
Source: this plan, section 9

Why this matters

The programme ends with a decision that has to be made in public either way: does BNNR ship
diagnosis-conditioned selection as the default, or does it not.

Steps

1. Write the findings document, in the T20 house style, with the can-claim and cannot-claim
   split.
2. If the FIX-7-1 criterion was met: flip the default search_policy to diagnosis_single,
   ship the threshold profile, and update README, docs and release notes.
3. If it was not: keep exhaustive as the default, publish the negative result, and say plainly
   what would be needed to decide it.
4. Either way: document what the diagnosis measures, what it cannot see, and the
   generalisation boundary of the calibration.

Done when

The default is set deliberately, with the evidence for the choice in the repository, and the
docs no longer promise anything the benchmarks have not shown.

---

How to take this: comment "taking this" and wait to be assigned. Branch fix-7-3-calibration-findings from upstream/main, and put Closes #<this issue number> in your PR. Full workflow: the Cohort Handbook (pinned in Discord).

Plan: plans/FIX-PLAN-T20-2026-08-23.md, section 9.