[clubanderson]

Closes #12373. Docs only — nothing is switched on; consent is recorded on the issue before any start date.

Part of RFC #12062 (proposal C, Study). cc @mrbobbytables

What

src/docs/research/ablation-plan-2026-10.md (named to match the pre-registration's dated convention; the issue text says 2026-Q4):

• Auto-merge unchanged everywhere stated up front, with per-arm reasoning: shadow gates are record-only (review.shadow / intent.shadow never hold or weaken require_approval / enforce); the cross-family reviewer changes which model reviews, adds no approval requirement.
• Consenting hives table (Bluefin / Console / Hive) with consent-link checkboxes and hive-id placeholders.
• Arms quoted verbatim from pre-reg §Experimental arms; each hive runs both arms sequentially, 30 days each, 7-day washout; hash rule (sha256(hive id) first hex digit parity) decides arm order.
• Schedule table with TBD start/stop placeholders and "who flips it".
• Exact config diffs citing real hive.yaml leaves (review.shadow, intent.shadow, agents.<reviewer>.review_models.{pool, exclude_author_model, exclude_author_family, fallback}), and a one-revert rollback.
• Metrics (H3, H7 rows and source definitions) copied verbatim from the pre-registration; where results are read: /api/evidence/summary shadow_gate + reviewer_accuracy_by_model, hub /api/hub/evidence family_false_approve_rate, hivectl evidence export (✨ hivectl: evidence export --since --format csv|jsonl (same schema as the Evidence charts) #12367), backfill baseline (✨ research: evidence backfill tooling + runs — Bluefin first, Console (L0→L6), then Hive/tunaos/Frostyard/Open Horizon #12371).
• "Predictions we expect to lose" restated (H2, H5, H7 as a result against default gating, H3 possibly underpowered); results reported win or lose on the issue and in the paper.
• Linked from preregistration-2026-10.md (one navigation line under Tracking, explicitly not an amendment) and review-evidence.md. Changelog fragment added-12373-ablation-plan.md.

Guardrail (RFC #12062 Non-goals)

Record/read-only. Nothing here reads into or alters merge eligibility / auto-merge; no code changes.

Local: check-docs-links.py and check-docs-citations.py clean.

[kubestellar-prow]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign danathar for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[github-actions]

Open PR overlap check

Base branch: v5

Conflicts are checked with every PR merged onto the current tip of v5, so a PR that only needs a rebase is not reported as conflicting with its siblings.

PR	Title	Relation	Files
#12386	✨ evidence: panel 5 — claim vs verdict per week	same files	src/docs/review-evidence.md
#12395	✨ hivectl: evidence backfill tooling — schema v1 bundles from GitHub history (#12371)	same files	src/docs/review-evidence.md
#12388	✨ dashboard: Evidence panel 4 — boundary enforcement (attempted vs blocked writes per tier; sentinel hits/week)	same files	src/docs/review-evidence.md
#12409	✨ evidence: schedule the monthly complexity/static-analysis snapshot per repo and expose it in the summary API	same files	src/docs/review-evidence.md
#12389	✨ evidence: panel 6 — work balance (issues opened by kind vs closed)	same files	src/docs/review-evidence.md
#12390	✨ intent: per-tier evaluated / would-block counters in the audit log + durable counter file	same files	src/docs/review-evidence.md

[hivecommons-hive]

merged by Hive trusted-author auto-merge: author @clubanderson holds role owner and GitHub permission admin on repo; CI green at 30ff020.

[github-actions]

Thank you for your contribution! Your PR has been merged.

We'd love to hear how your experience was: share feedback