feat(governance): Rounds 36-50 panel — plan-only and plain-review boundaries

Extend Socratic panel through Round 50 with unanimous agreements on
plan-only-no-code and plain-review-vs-risk holdouts, production-negation
router boundaries, ROUTE-002/003 expansion, ROUTING_CONTRACT R9, and
governance doc sync.

Author: johnteee

CONTRIBUTING.md

@@ -28,6 +28,7 @@ Before tuning `route_paraphrase_eval.py`:
 
 1. Add fresh holdout phrases to `route-002-holdout-eval.yaml` and/or adversarial cases to `route-003-adversarial-eval.yaml`.
 2. Run `make all` — `validate_route_fixture.py` enforces minimum fixture coverage.
+3. When fixing review-vs-risk or plan-only boundaries, add holdout phrases **before** editing `route_paraphrase_eval.py` (see `ROUTING_CONTRACT.md` R8–R9).
 3. Record governance-surface decisions in `PROJECT_KNOWLEDGE.md` Decision Index when routing contract changes.
 
 See [ROUTING_CONTRACT R8](reflective-prompt-library/plans/ROUTING_CONTRACT.md) and the [Governance Maintenance Playbook](reflective-prompt-library/GLOSSARY.md).

reflective-prompt-library/GLOSSARY.md

@@ -280,11 +280,32 @@ fairness without translating canonical `SKILL.md` contracts.
 its English equivalent intent group.
 
 
+---
+
+
+## Plan-Only Without Code / 僅規劃不寫程式
+
+A request for tickets, acceptance criteria, rollout plans, or specs with explicit
+no-code context. Routes to `reflective-spec-plan`, not `reflective-implement`.
+
+**Operational test:** Phrase includes planning artifact signals **and** no-code
+context (e.g. "plan only", "without touching the repo", 不要改程式).
+
+---
+
+## Plain Review (Non-Production) / 非正式環境審查
+
+Code or diff review for correctness, readability, or regressions **without**
+production-risk assessment. Routes to `reflective-review`, not `reflective-risk`.
+
+**Operational test:** Review/diff/PR context plus production negation (e.g.
+"not production deploy", 不是正式環境風險) or readability-only scope.
+
 ---
 
 ## Governance Maintenance Playbook / 治理維護手冊
 
-Ongoing upkeep after panel close (Rounds 1–35). Not agent instructions — operator checklist.
+Ongoing upkeep after panel close (Rounds 1–50). Not agent instructions — operator checklist.
 
 **Operational test:** Before router tuning, add fresh ROUTE-002/003 holdout phrases; run `make validate`; record decisions in `PROJECT_KNOWLEDGE.md` Decision Index when governance surface changes.
 

reflective-prompt-library/PROJECT_KNOWLEDGE.md

@@ -41,7 +41,7 @@ operating its own agent runtime. See [README.md](README.md#north-star).
 > does not silently expand. Completed milestones are retired from this section;
 > their outcomes live in the Decision Index.
 
-**Ongoing (post Round 35 panel):** maintain ROUTE-001/002/003 evals, governance validators,
+**Ongoing (post Round 50 panel):** maintain ROUTE-001/002/003 evals, governance validators,
 and holdout expansion before router tuning — see [GLOSSARY.md](GLOSSARY.md) Governance
 Maintenance Playbook. No open implementation blockers from Rounds 8–20; deferred promotions
 are recurrence-gated — see [panel backlog](plans/multi-agent-panel-consensus-2026-06-25.md#recurrence-gated-backlog-not-panel-blockers).
@@ -81,6 +81,7 @@ are recurrence-gated — see [panel backlog](plans/multi-agent-panel-consensus-2
 - 2026-06-20 Synthesize external-adoption case studies, then correct the evidence and promotion-gate scope after the Test Plan routing counterexample → [record](plans/external-adoption-case-studies-2026-06-20.md)
 - 2026-06-21 Evaluate Hyperplan / multi-agent adversarial planning — no change (runtime = non-goal; methodology overlaps; possible gaps not promoted) → [record](plans/external-adoption-case-studies-2026-06-20.md)
 - 2026-06-25 Round 21 panel audit — no implementation blockers; recurrence-gated backlog formalized; QUALITY_GATES drift fixed → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
+- 2026-06-25 Rounds 36–50 panel — plan-only/plain-review holdouts, production-negation boundaries, ROUTE-002/003 expansion → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Rounds 8–20 panel — milestones closed, ROUTE-003 adversarial eval, skill examples gate, maintenance playbook → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Round 7 panel — AGENTS harness-policy alignment, zh-TW ROUTE-002 holdout, metrics sync → [record](plans/multi-agent-panel-consensus-2026-06-25.md)
 - 2026-06-25 Round 6 panel — undocumented-decisions warning, ROUTE-002 holdout expansion, benchmark fixture CI, reject full SKILL i18n → [record](plans/multi-agent-panel-consensus-2026-06-25.md)

reflective-prompt-library/plans/QUALITY_GATES_SUMMARY.md

@@ -260,8 +260,8 @@ The implementation aligns with research findings:
 | Skills with governance | 9/9 | ✅ Complete |
 | Benchmark tasks | 23 | ✅ Ready |
 | Routing consistency | 100.0% | ✅ Passes ROUTE-001 expanded boundary eval |
-| Holdout routing consistency | 100.0% | ✅ ROUTE-002 (28 groups, 82 paraphrases) |
-| Adversarial routing consistency | 100.0% | ✅ ROUTE-003 (9 groups, 20 paraphrases) |
+| Holdout routing consistency | 100.0% | ✅ ROUTE-002 (32 groups, 91 paraphrases) |
+| Adversarial routing consistency | 100.0% | ✅ ROUTE-003 (11 groups, 24 paraphrases) |
 | Skill example coverage | 9/9 | ✅ validate_skill_examples.py |
 | Linting errors | 0 | ✅ Clean |
 
@@ -292,7 +292,7 @@ ROUTE-002 measures unseen phrasing separately from ROUTE-001. Round 7 (2026-06-2
 ### Done ✅
 
 1. **CI/CD** — `.github/workflows/python-tools.yml` runs `make all` on push/PR
-2. **ROUTE-001/002/003 in CI** — 128 + 82 + 20 paraphrases at 100% consistency (seeded fixtures); `validate_route_fixture.py` gates minimum coverage
+2. **ROUTE-001/002/003 in CI** — 128 + 91 + 24 paraphrases at 100% consistency (seeded fixtures); `validate_route_fixture.py` gates minimum coverage
 3. **Governance validators** — links, lint, governance metadata, PROJECT_KNOWLEDGE, benchmark fixture, skill examples
 4. **Harness policy docs** — CONTRIBUTING, AGENTS, SKILL_INSTALLATION, maintenance playbook
 

reflective-prompt-library/plans/ROUTING_CONTRACT.md

@@ -79,6 +79,15 @@ This contract applies to:
 - `validate_route_fixture.py` enforces minimum group and phrase counts in CI.
 - Holdout expansion is maintenance, not proof of broad semantic routing.
 
+### R9: Production-negation and plan-only boundaries
+
+- Phrases that **negate** production risk (e.g. "not production deploy", 不是正式環境風險)
+  must not auto-route to `reflective-risk` when the user requests plain code review.
+- Planning-only requests with explicit no-code context route to `reflective-spec-plan`,
+  even when incidental words like "change" appear in "code changes".
+- Add holdout cases for these boundaries before tuning router keyword lists.
+
+
 ## Router Output Contract
 
 Use this minimal output shape for routing responses:

reflective-prompt-library/plans/multi-agent-panel-consensus-2026-06-25.md

@@ -811,8 +811,209 @@ User directive: continue Socratic panel for as many rounds as feasible.
 
 - `make all`: **62 pytest**, ROUTE-001 **100%** (16 groups / 128), ROUTE-002 **100%** (28 / 82), ROUTE-003 **100%** (9 / 20), governance **9/9**, skill examples **9/9**, route fixture gate **pass**
 
+
+## Round 36 — Plan-only holdout (2026-06-25)
+
+User directive: continue Socratic panel for as many rounds as feasible.
+
+### BC: Add `plan_only_no_code_holdout` to ROUTE-002?
+
+| Lens | Position |
+| --- | --- |
+| Codex | **Agree** — "code changes" in plan-only phrasing misroutes without holdout |
+| Opus | **Agree** — falsifiable boundary before keyword tuning |
+| Gemini | **Agree** — plan-only saves implement context cost |
+| Composer | **Agree** — IDE users mix plan + code words casually |
+| Sakana | **Agree** — boundary clarity over new skill |
+| GLM | **Agree** — TW plan-only phrases need parity |
+
+**Consensus:** **Agree** — ROUTE-002 group + plan-only boundary rule.
+
+## Round 37 — Plain review vs risk (2026-06-25)
+
+### BD: Add `plain_review_not_risk_holdout`?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — production keyword false positives on plain PR review |
+
+**Socratic Q:** What fails if every "production" mention routes to risk?
+**Answer:** Readability/regression reviews get over-rigor and wrong workflow.
+
+**Consensus:** **Agree** — holdout group + production-negation boundary.
+
+## Round 38 — Router boundary repair (2026-06-25)
+
+### BE: Plan-only and production-negation rules in `route_paraphrase_eval.py`?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — holdout added first (R8); then boundary repair |
+
+**Consensus:** **Agree** — `no_code_context` hoist, `production_negated`, `plan_only_signals`, `plain_review_signals`.
+
+## Round 39 — Adversarial implement/research traps (2026-06-25)
+
+### BF: ROUTE-003 `implement_not_plan_trap` and `research_not_brief_trap`?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — separate adversarial traps from holdout tuning set |
+
+**Consensus:** **Agree** — +2 ROUTE-003 groups.
+
+## Round 40 — ROUTING_CONTRACT R9 (2026-06-25)
+
+### BG: Codify production-negation and plan-only as normative?
+
+| Lens | Position |
+| --- | --- |
+| Opus/Codex | **Agree** — contract documents fairness edge cases |
+| All | **Agree** — R9 in `ROUTING_CONTRACT.md` |
+
+**Consensus:** **Agree** — R9 added.
+
+## Round 41 — Fixture minimum bump (2026-06-25)
+
+### BH: Raise `validate_route_fixture.py` minimums?
+
+| Lens | Position |
+| --- | --- |
+| Codex | **Agree** — prevents accidental holdout shrinkage |
+| All | **Agree** — ROUTE-002 32/91, ROUTE-003 11/24 |
+
+**Consensus:** **Agree** — minimum constants updated; existing pytest covers.
+
+## Round 42 — zh-TW plan/review holdouts (2026-06-25)
+
+### BI: TW fairness groups for plan-only and plain review?
+
+| Lens | Position |
+| --- | --- |
+| GLM | **Agree** — 不是正式環境風險 must not force risk workflow |
+| All | **Agree** — `zh_tw_plan_only_holdout`, `zh_tw_plain_review_holdout` |
+
+**Consensus:** **Agree**.
+
+## Round 43 — GLOSSARY boundary terms (2026-06-25)
+
+### BJ: Operational definitions for plan-only and plain review?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — EN definitions; TW headings in glossary |
+
+**Consensus:** **Agree** — GLOSSARY entries + playbook rounds 1–50.
+
+## Round 44 — Metrics sync (2026-06-25)
+
+### BK: QUALITY_GATES_SUMMARY metrics?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — 32/91 and 11/24 |
+
+**Consensus:** **Agree**.
+
+## Round 45 — Decision Index (2026-06-25)
+
+### BL: PROJECT_KNOWLEDGE entry for Rounds 36–50?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — single Decision Index pointer |
+
+**Consensus:** **Agree**.
+
+## Round 46 — Cheatsheet cues (2026-06-25)
+
+### BM: SKILL_TRIGGER_CHEATSHEET EN/TW boundary lines?
+
+| Lens | Position |
+| --- | --- |
+| Composer/GLM | **Agree** — progressive disclosure for adopters |
+| All | **Agree** |
+
+**Consensus:** **Agree**.
+
+## Round 47 — Tenth skill reaffirm (2026-06-25)
+
+### BN: Re-open tenth core skill?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Reject** — promotion gate unchanged |
+
+**Consensus:** **Reject**.
+
+## Round 48 — Benchmark CI reaffirm (2026-06-25)
+
+### BO: LLM benchmark in CI?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Reject** — fixture gate only |
+
+**Consensus:** **Reject**.
+
+## Round 49 — CONTRIBUTING R8–R9 pointer (2026-06-25)
+
+### BP: Routing maintenance checklist extension?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — plan-only/review-vs-risk holdout-before-tune step |
+
+**Consensus:** **Agree**.
+
+## Round 50 — Final extended seal (2026-06-25)
+
+### BQ: Close panel at Round 50?
+
+| Lens | Position |
+| --- | --- |
+| All six | **Agree** — options A–BQ resolved; maintenance only remains |
+
+**Consensus:** **Agree close** — panel sealed at **Round 50**; Decision Index entry.
+
+### Rounds 36–50 verdict table
+
+| ID | Option | Verdict | Action |
+| --- | --- | --- | --- |
+| BC | Plan-only holdout | **Agree** | ROUTE-002 group |
+| BD | Plain review holdout | **Agree** | ROUTE-002 group |
+| BE | Router boundaries | **Agree** | plan-only + production-negation |
+| BF | Adversarial traps | **Agree** | ROUTE-003 +2 groups |
+| BG | R9 contract | **Agree** | ROUTING_CONTRACT.md |
+| BH | Fixture minimums | **Agree** | validate_route_fixture.py |
+| BI | zh-TW holdouts | **Agree** | ROUTE-002 TW groups |
+| BJ | GLOSSARY terms | **Agree** | plan-only + plain review |
+| BK | Metrics sync | **Agree** | QUALITY_GATES_SUMMARY |
+| BL | Decision Index | **Agree** | PROJECT_KNOWLEDGE |
+| BM | Cheatsheet cues | **Agree** | EN + zh-TW |
+| BN | Tenth skill | **Reject** | — |
+| BO | Benchmark CI | **Reject** | — |
+| BP | CONTRIBUTING | **Agree** | R8–R9 step |
+| BQ | Final seal | **Agree** | Panel sealed Round 50 |
+
+**All roles agree.**
+
+## Implemented Changes (Rounds 36–50)
+
+- `route-002-holdout-eval.yaml`: plan-only, plain-review, zh-TW plan/review groups (91 paraphrases / 32 groups)
+- `route-003-adversarial-eval.yaml`: implement-not-plan, research-not-brief traps (24 paraphrases / 11 groups)
+- `route_paraphrase_eval.py`: production-negation, plan-only, plain-review boundaries
+- `validate_route_fixture.py`: minimum coverage 32/91 and 11/24
+- `ROUTING_CONTRACT.md`: R9
+- `GLOSSARY.md`, `QUALITY_GATES_SUMMARY.md`, `PROJECT_KNOWLEDGE.md`, `CONTRIBUTING.md`
+- `SKILL_TRIGGER_CHEATSHEET.md` + `.zh-TW.md`
+
+## Verification (Rounds 36–50)
+
+- `make all`: **62 pytest**, ROUTE-001 **100%** (16/128), ROUTE-002 **100%** (32/91), ROUTE-003 **100%** (11/24)
+
 ## Panel status (final)
 
-**Sealed 2026-06-25** after **Round 35** (options A–BB). Extended deliberation complete. Future work is recurrence-gated maintenance per playbook — expand holdout before router tuning; no open implementation blockers.
+**Sealed 2026-06-25** after **Round 50** (options A–BQ). Extended deliberation complete. Future work is recurrence-gated maintenance per playbook — expand holdout before router tuning; no open implementation blockers.
 
-_Ongoing maintenance: [GLOSSARY.md](../GLOSSARY.md) Governance Maintenance Playbook — `validate_route_fixture.py` + ROUTE-002/003 holdout expansion before `route_paraphrase_eval.py` tuning._
+_Ongoing maintenance: [GLOSSARY.md](../GLOSSARY.md) Governance Maintenance Playbook — `validate_route_fixture.py` + ROUTE-002/003 holdout expansion before `route_paraphrase_eval.py` tuning (R8–R9)._

reflective-prompt-library/plans/route-002-holdout-eval.yaml

@@ -296,6 +296,44 @@ holdout_sets:
       - clarify the goal and assumptions before we pick a workflow
       - what should we optimize for before choosing skills
 
+  - name: plan_only_no_code_holdout
+    expected_workflow: reflective-spec-plan
+    expected_rigor: medium
+    expected_enhancements:
+      - acceptance_criteria
+    phrases:
+      - draft rollout plan only no code changes
+      - write tickets and acceptance criteria without touching the repo
+      - spec the delivery plan before any implementation starts
+
+  - name: plain_review_not_risk_holdout
+    expected_workflow: reflective-review
+    expected_rigor: high
+    expected_enhancements:
+      - quality_review
+    phrases:
+      - review this PR for style and logic before merge
+      - check the diff for readability not production deploy
+      - inspect the patch for regressions without production risk assessment
+
+  - name: zh_tw_plan_only_holdout
+    expected_workflow: reflective-spec-plan
+    expected_rigor: medium
+    expected_enhancements:
+      - acceptance_criteria
+    phrases:
+      - 把規格寫出來但不要改程式
+      - 只寫工單和驗收標準不要動 code
+
+  - name: zh_tw_plain_review_holdout
+    expected_workflow: reflective-review
+    expected_rigor: high
+    expected_enhancements:
+      - quality_review
+    phrases:
+      - 審查 PR 的可讀性不是正式環境風險
+
+
 evaluation_rules:
   - name: route_equivalence
     assert: same_intent_group_same_workflow

reflective-prompt-library/plans/route-003-adversarial-eval.yaml

@@ -100,6 +100,25 @@ adversarial_sets:
       - lessons learned retro after this sprint
       - sprint retrospective with lessons learned
 
+  - name: implement_not_plan_trap
+    expected_workflow: reflective-implement
+    expected_rigor: low
+    expected_enhancements:
+      - tests
+    phrases:
+      - ship the code fix for the off-by-one bug
+      - implement the approved spec in the repository
+
+  - name: research_not_brief_trap
+    expected_workflow: reflective-research
+    expected_rigor: medium
+    expected_enhancements:
+      - source_check
+    phrases:
+      - compare official docs for both libraries before deciding
+      - source-backed comparison of migration paths
+
+
 
 evaluation_rules:
   - name: route_equivalence