diff --git a/docs/remediation/KICKSTART.md b/docs/remediation/KICKSTART.md index 1bfd3c24..6fdf43d8 100644 --- a/docs/remediation/KICKSTART.md +++ b/docs/remediation/KICKSTART.md @@ -34,6 +34,9 @@ mechanically until Build 3 (rotation) makes it automatic. durably with enumerated evidence under its own minted identity → **coordinator head-pinned merge**. The queue guard runs but is **zero-information until RM-03 lands** (D-23) and must not be cited as evidence. **After a `GO`, freeze pushes** — even a doc tweak voids the verdict. The coordinator assigns the gate seat. +- **Query for refutation, never for confirmation.** A subordinate asked to confirm a hypothesis will + agree — the bias is in the question, not the answerer, and agent seats are agreeable by construction. + State the hypothesis as yours, ask for the evidence that KILLS it, and reproduce when it matters. - **Dogfooding:** every fix validated against its live seed case (MISSION.md lists them). - **Tracking → DB** (hard cutover); do NOT re-invest in flat-file tracking. jarvis-brain PDA is off-limits. - **Git identity:** export `MOSAIC_GIT_IDENTITY=` so wrappers author correctly and survive respawn. diff --git a/docs/remediation/MISSION.md b/docs/remediation/MISSION.md index 6a77de6f..f6fcd769 100644 --- a/docs/remediation/MISSION.md +++ b/docs/remediation/MISSION.md @@ -91,6 +91,32 @@ gate/program; the LLM handles only genuine judgment. > seat in the loop. Residual risk bound to **RM-59** (`depends_on: RM-12, RM-21, RM-25`), where the > choke-point executor and spine verify from _outside_ the worktree's authority. +### First-class principle — query for refutation, never for confirmation + +> **A subordinate asked to confirm a hypothesis will agree. Ask it to refute, with evidence.** +> +> The bias is induced by the **query**, not by the answerer's diligence. _"The DB flaked — please +> confirm"_ and _"confirm or refute this, with the log line that proves it"_ are different instruments, +> and they return different answers to the same question. The first harvests agreement; only the second +> can return **"you are wrong, and here is why."** +> +> This matters most with agent subordinates, which are **agreeable by construction**: fluent, eager to +> be useful, and structurally disinclined to tell the dispatcher their premise is false. A confirmation +> query aimed at one is close to a guaranteed yes — so the discipline cannot rest on the answerer being +> rigorous. **It has to be built into how the question is asked.** +> +> Promoted to the charter by Mos (2026-08-01). Origin: the orchestrator hypothesised that a coincident +> `ci-postgres` failure caused a CI test failure and asked the implementing seat to **confirm or refute** +> it. The seat **refuted it** with the log (`ci-postgres:5432 - accepting connections`, migrations +> completed) and identified the real cause. Reproduction on an identical head then settled it. Had the +> query been phrased for confirmation, the agreement would have been returned, **D-21 would have been +> re-classified on a false premise**, and a banked finding would have been silently corrupted. +> +> **Operationally:** state your hypothesis explicitly, mark it as yours, ask for _evidence that kills +> it_, and say what evidence would change your mind. A hypothesis you cannot describe how to falsify is +> not yet a hypothesis. Where the answer is consequential, **reproduce** rather than accept — two +> independent runs beat one confident report. + ## Decision record (authoritative, immutable) - **15/15 proposals decided: 13 accept, 2 modify (P-AUTHORITY-001, P-INBOX-001), 0 reject.**