docs(locomo): record v2-r13 evidence-driven no-op

docs/superpowers/plans/2026-03-19-v2-r13-locomo-evidence-single-knob-implementation-plan.md

@@ -0,0 +1,45 @@
+# V2-R13 LoCoMo Evidence-Only Single-Knob Implementation Plan
+
+> **For agentic workers:** REQUIRED: Use superpowers:subagent-driven-development (if subagents available) or superpowers:executing-plans to implement this plan. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Run a constrained evidence check for three stopword candidates and avoid code change unless acceptance rule is met.
+
+**Architecture:** A/B evaluation only; repository code remains unchanged when criteria are not met.
+
+**Tech Stack:** Python offline evaluator + existing LoCoMo gate scripts.
+
+---
+
+### Task 1: Candidate Evaluation
+
+**Files:**
+- Create: `/tmp/v2_r13_candidate_eval.py` (temporary)
+
+- [ ] Evaluate `games`, `kind`, `names` against current baseline.
+- [ ] Compare `overall` + `multi-hop` jointly.
+
+### Task 2: Execute Decision
+
+**Files:**
+- No retrieval source edits if acceptance rule fails
+
+- [ ] Apply no-op decision if no candidate satisfies acceptance rule.
+
+### Task 3: Verification
+
+**Files:**
+- No source edits
+
+- [ ] `cargo test --workspace`
+- [ ] `./.venv/bin/pytest python/tests -q`
+- [ ] `corepack pnpm --dir packages/node test`
+- [ ] `corepack pnpm --dir packages/mcp test`
+- [ ] `REBUILD_PYTHON=0 tools/locomo/run-full-gate.sh 5`
+
+### Task 4: Finish Round
+
+**Files:**
+- Create: `docs/superpowers/releases/2026-03-19-v2-r13-locomo-evidence-single-knob-results.md`
+
+- [ ] Record A/B evidence and no-op rationale.
+- [ ] Commit docs-only round.

docs/superpowers/releases/2026-03-19-v2-r13-locomo-evidence-single-knob-results.md

@@ -0,0 +1,35 @@
+# V2-R13 LoCoMo Evidence-Only Single-Knob Results
+
+## Summary
+
+This round intentionally avoids retrieval code changes unless one candidate gives clear dual gain.
+
+Candidates evaluated: `games`, `kind`, `names`.
+
+## Offline A/B Evidence
+
+Baseline:
+
+- overall: `0.5822`
+- multi-hop: `0.4255`
+- open-domain: `0.3370`
+- single-hop: `0.6314`
+
+Candidates:
+
+- `games`: overall `0.5822`, multi-hop `0.4291`, single-hop `0.6302`
+- `kind`: unchanged vs baseline
+- `names`: overall `0.5817`, multi-hop `0.4220`
+
+Decision rule:
+
+- require both overall + multi-hop improvement with no obvious side effect
+
+Decision:
+
+- no candidate passed; no retrieval code change made
+
+## Outcome
+
+- V2-R13 is a docs-only, evidence-driven no-op round.
+- over-design avoided by enforcing the single-knob acceptance gate.

docs/superpowers/specs/2026-03-19-v2-r13-locomo-evidence-single-knob-design.md

@@ -0,0 +1,43 @@
+# V2-R13 LoCoMo Evidence-Only Single-Knob Design
+
+## Goal
+
+Avoid over-design by allowing at most one stopword tweak only when evidence shows clear gain.
+
+Round policy:
+
+- run offline A/B for exactly three candidates (`games`, `kind`, `names`)
+- only change code if `overall` and `multi-hop` both improve
+- if not, ship no retrieval code change
+
+## Baseline
+
+- overall: `0.5822`
+- multi-hop: `0.4255`
+- open-domain: `0.3370`
+
+## A/B Results
+
+- `games`: overall `0.5822`, multi-hop `0.4291`, single-hop `0.6302`
+- `kind`: unchanged from baseline
+- `names`: overall `0.5817`, multi-hop `0.4220`
+
+## Decision
+
+No candidate satisfies the acceptance rule (`overall` + `multi-hop` both up without side effects).
+
+Chosen action: no retrieval logic change in this round.
+
+## Test Strategy
+
+1. `cargo test --workspace`
+2. `./.venv/bin/pytest python/tests -q`
+3. `corepack pnpm --dir packages/node test`
+4. `corepack pnpm --dir packages/mcp test`
+5. `REBUILD_PYTHON=0 tools/locomo/run-full-gate.sh 5`
+
+## Success Criteria
+
+1. no code changes in retrieval path
+2. required 4-command matrix passes
+3. full-gate remains PASS