@oomkapwn/enquire-mcp 3.11.1 → 3.11.2

package/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 All notable changes to this project will be documented here. The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.11.2] — 2026-06-30
+
+> **TL;DR:** **v3.11.2 STABLE — promoted `@rc` → `@latest`. A positioning + eval-rigor patch, zero `src/` behavior change.** Born from a study of a 21k★ adjacent-domain competitor (DeusData/codebase-memory-mcp, *code* memory): borrow the **discipline** (lead with one *measured* number, re-narrate constraints as moats, broaden agent-SEO), never the domain or an unmeasured claim. Ships **`bench:context`** (a context-savings harness — our niche analog of "Nx fewer tokens", with **no figure published** until a representative-vault run + sign-off) + **`docs/EVALUATION_PLAN.md`** (a pre-registered, score-free methodology) + a README hero that now **leads with the already-measured reranker delta** (+15.5 NDCG@10 / +24.7 MRR on a reproducible 60-query ablation) and re-frames the on-device model's weight as the privacy/grounded **moat** (zero cloud calls during serve — *enforced* via `setEmbeddingsOffline`, not aspirational). Every positioning claim is substantiated + invariant-pinned; the no-overclaim rule held (the own-diff sweep even caught + reverted a stray synthetic-vault number `git add -A` had committed). **46 tools · 19 prompts · 1439 tests · 0 CodeQL · 0 open dependabot.** No API breaks (additive patch). CI publishes `@latest` + signed provenance (SLSA L2) + OIDC MCP-registry sync. **Per-RC detail: rc.1 below.**
+
+## [3.11.2-rc.1] — 2026-06-29
+
+> **TL;DR:** **Competitive-study response (DeusData/codebase-memory-mcp, 21k★) — eval rigor + number-led positioning, all inside the no-overclaim rule.** A study of that project's positioning playbook (which drove its stars via one measured hook + frictionless install) yielded three borrowable, niche-correct moves — and an explicit list of what NOT to copy (its code-domain features, single-binary rewrite, SLSA-3 claim, any borrowed number).
+
+### Added
+
+- **`bench:context` — context-savings harness** (`scripts/bench-context.mjs` + `tests/bench-context.test.ts`). Our niche-correct analog of code-memory tools' "Nx fewer tokens" headline: per question, it measures the `obsidian_context_pack` budget-capped bundle vs the full bodies of the top-K hits an agent would otherwise read in full, over the same hybrid retrieval, and reports the token ratio. Deterministic synthetic vault, pure measurement helpers (`estimateTokens` / `savingsRatio`) covered with NEGATIVE controls so an inflated ratio can't ship. **No headline figure is published** — a "Nx less context" claim requires a representative-vault run + maintainer sign-off (same gate as LongMemEval). `docs/benchmarks.md` gains a "Context efficiency" methodology section.
+- **`docs/EVALUATION_PLAN.md`** — a pre-registered evaluation methodology (no scores by design), committed *before* the headline runs so the method is reviewable without front-running a result. Codifies the project's "every claim an auditor can verify" brand: measured-not-asserted, reproducible, reviewed; retrieval-quality-not-QA-accuracy scope; zero-result=FAIL; weakness disclosure; the reference-hardware + sign-off publication gate. Added to `files[]` (parity with `benchmarks.md`).
+
+### Changed (positioning — substantiated claims only)
+
+- **README hero leads with a measured number.** A new credibility line cashes out the qualitative hooks with the already-measured, committed reranker delta — **+15.5 NDCG@10 / +24.7 MRR** over plain hybrid on the reproducible 60-query ablation — the way the 21k★ competitor leads with a falsifiable figure instead of a feature wall. The qualitative "most advanced Obsidian MCP" line stays (and the GitHub About lead is unchanged — pinned by `github-metadata-invariant`).
+- **Install weight re-narrated as the privacy/grounded MOAT** (hero point #3): the ~110 MB on-device model is *why* recall is grounded + private with zero cloud calls during serve — "enforced, not aspirational" — instead of an unexplained cost. The honest "When NOT to use it" concession is untouched.
+- **Broader agent-targeting in npm keywords** (windsurf, zed, aider, gemini-cli, opencode, kilocode) — pure discoverability; enquire is client-agnostic by construction over MCP. (GitHub-topic broadening + an OpenSSF Scorecard workflow are tracked as separate follow-ups — the topic set is checked against live GitHub by an invariant, and workflow files need a maintainer web-UI merge.)
+
+### Explicitly NOT copied (niche guard, documented)
+
+The 21k★ tool is a **code**-intelligence memory; we are a **notes/PKM** memory. We did NOT adopt: its single-static-binary/zero-dep rewrite (our on-device model is the privacy moat), any code-domain feature (tree-sitter AST, Hybrid LSP, CALLS/INHERITS edges, dead-code, git-diff blast-radius), 3D graph visualization as a hero asset (documented non-goal), its SLSA-3 badge (we are guarded at L2), multi-agent auto-config / hosted team-graph (we are single-user local-first), or any of its metrics restated as ours.
+
+**1434 → 1439 source tests.** No `src/` behavior change (script + docs + positioning). Ships `@rc`; promotion maintainer-gated.
+
 ## [3.11.1] — 2026-06-28
 
 > **TL;DR:** **v3.11.1 STABLE — promoted `@rc` → `@latest` (maintainer call). A patch line that closes the case-fold-asymmetry class across all three of its instances.** It began as the **v3.11.0-STABLE external-audit response** (2 independent auditors — Cursor 4.3/ship, Goose 4.4/ship — both **0 CRIT / 0 HIGH / 0 MED**, the cleanest external round of the line; both re-verified the rc.17→rc.25 hardening holds through `dist/`). Cursor's 2 LOWs (`SCHEMA-validate-then-write-gap`) were **reasoned-rejected** (a false equivalence — the write-content sink is linear and already double-bounded by the 7.5 MB HTTP cap + 5 MB `maxFileBytes`; a 1 MB cap would reject legitimate large notes), and Goose's NEL/VT/FF finding was a confirmed non-defect. The real value came from internal adversarial lenses: an **anti-anchoring** Workflow caught a Greek word-final-sigma silent under-replace in `replace_in_notes` that *neither* external auditor found (rc.1), and the **mandatory pre-promotion re-sweep** then caught a third, more-severe instance of the same class — `obsidian_search_text` silently dropping notes on a final-Σ query token (rc.2) — before `@latest` flipped. All three instances are now consistent (per-code-point `foldForMatch` for `replace_in_notes` + `search_text`; the `semanticSearch` snippet-centring residual stays documented-accept as it's cosmetic). **46 tools · 19 prompts · 1434 tests · 0 CodeQL · 0 open dependabot.** No API breaks (additive patch). CI publishes `@latest` + signed provenance (SLSA L2) + OIDC MCP-registry sync. **Per-RC detail: rc.1 + rc.2 below.** **Lesson of the line: two strong external auditors with a clean bill of health still left a real correctness bug — the internal anti-anchoring + pre-promotion re-sweep are what found the class; external state-driven re-verification and internal un-enumerated-class hunting are complementary, exactly as the ≥2-auditor gate intends.**

package/README.ar.md

@@ -54,7 +54,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 > 3. **صفر استدعاءات سحابية أثناء التشغيل.** النماذج مخزّنة محلياً (تنزيل لمرة واحدة من HuggingFace). محتوى مكتبتك لا يغادر جهازك أبداً. آمنة للعمل المعزول (air-gap) افتراضياً.
 > 4. **استدعاء واعٍ بالحداثة.** تُبلّغ كل نتيجة عن عمر الملاحظة؛ وإعادة الترتيب الاختيارية بالحداثة تتيح للوكيل تفضيل المعرفة الحديثة ووسم الحقائق القديمة لإعادة التحقق — حدود "الوعي بالنسيان"، مبنيّة على `mtime` الذي تملكه ملفاتك أصلاً.
 
-**46 أداة · 19 موجِّه MCP · 1434+ اختبار وحدة · 50+ لغة · إصدار مستقر v3.11.x · مُقيَّد بالـ semver · MIT · إثبات بناء npm (SLSA L2).**
+**46 أداة · 19 موجِّه MCP · 1439+ اختبار وحدة · 50+ لغة · إصدار مستقر v3.11.x · مُقيَّد بالـ semver · MIT · إثبات بناء npm (SLSA L2).**
 
 </div>
 

package/README.es.md

@@ -48,7 +48,7 @@ Tu bóveda de Obsidian se convierte en **memoria a largo plazo persistente y con
 > 3. **Cero llamadas a la nube durante el servicio.** Modelos en caché local (una descarga única desde HuggingFace). El contenido de tu bóveda nunca sale de tu máquina. Seguro para entornos aislados por defecto.
 > 4. **Recuperación consciente de la frescura.** Cada resultado informa de la antigüedad de la nota; el reordenamiento por recencia opcional permite que un agente prefiera el conocimiento reciente y marque los hechos obsoletos para reverificación: la frontera consciente del olvido, construida sobre el `mtime` que tus archivos ya tienen.
 
-**46 herramientas · 19 prompts MCP · 1434+ pruebas unitarias · 50+ idiomas · v3.11.x estable · ligado a semver · MIT · procedencia de compilación en npm (SLSA L2).**
+**46 herramientas · 19 prompts MCP · 1439+ pruebas unitarias · 50+ idiomas · v3.11.x estable · ligado a semver · MIT · procedencia de compilación en npm (SLSA L2).**
 
 ---
 

package/README.fr.md

@@ -15,7 +15,7 @@
 [![CI](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@oomkapwn/enquire-mcp.svg?label=npm&color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
 [![downloads](https://img.shields.io/npm/dm/@oomkapwn/enquire-mcp.svg?color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
-[![tests](https://img.shields.io/badge/tests-1434%20passing-brightgreen.svg)](#️-confiance)
+[![tests](https://img.shields.io/badge/tests-1439%20passing-brightgreen.svg)](#️-confiance)
 [![stable](https://img.shields.io/badge/v3.11.x-stable-brightgreen.svg)](./STABILITY.md)
 [![build provenance](https://img.shields.io/badge/build_provenance-SLSA_L2-blue.svg)](https://slsa.dev/spec/v1.0/levels#build-l2)
 [![MCP](https://img.shields.io/badge/MCP-1.29-8A2BE2.svg)](https://modelcontextprotocol.io/)
@@ -53,7 +53,7 @@ Votre coffre Obsidian devient une **mémoire à long terme persistante et interr
 > 3. **Zéro appel au cloud pendant le service.** Modèles mis en cache localement (téléchargement unique depuis HuggingFace). Le contenu de votre coffre ne quitte jamais votre machine. Sûr en environnement isolé par défaut.
 > 4. **Rappel conscient de la fraîcheur.** Chaque résultat indique l'âge de la note ; le reclassement par récence optionnel permet à un agent de préférer le savoir frais et de signaler les faits périmés à revérifier — la frontière consciente de l'oubli, bâtie sur le `mtime` que vos fichiers possèdent déjà.
 
-**46 outils · 19 prompts MCP · 1434+ tests unitaires · 50+ langues · v3.11.x stable · lié au semver · MIT · provenance de build npm (SLSA L2).**
+**46 outils · 19 prompts MCP · 1439+ tests unitaires · 50+ langues · v3.11.x stable · lié au semver · MIT · provenance de build npm (SLSA L2).**
 
 ---
 
@@ -81,7 +81,7 @@ Votre coffre Obsidian devient une **mémoire à long terme persistante et interr
 | **GraphRAG-light** (détection de communautés de wikilinks par modularité de Louvain) | ✅ **uniquement ici** | ❌ | ❌ |
 | **Exécution autonome de requêtes `.base`** (fonctionne sans Obsidian ouvert) | ✅ **uniquement ici** | ❌ | ❌ délègue à Obsidian |
 | **Récupération HyDE** (Gao et al. 2023) + décomposition en sous-questions | ✅ **uniquement ici** | ❌ | ❌ |
-| **1434 tests unitaires · 9 portes CI requises + 5 indicatives par PR** | ✅ | s.o. | rare |
+| **1439 tests unitaires · 9 portes CI requises + 5 indicatives par PR** | ✅ | s.o. | rare |
 | **Provenance de build signée** (npm + Sigstore, SLSA Build L2) | ✅ | s.o. | ❌ |
 | **Surface publique liée au semver** ([STABILITY.md](./STABILITY.md)) | ✅ | s.o. | ❌ |
 | Autonome (aucun plugin Obsidian requis) | ✅ | ❌ requiert Obsidian | variable |
@@ -320,7 +320,7 @@ Canal : `npm install @oomkapwn/enquire-mcp` → dernière version stable (`@late
 ```bash
 git clone https://github.com/oomkapwn/enquire-mcp.git
 cd enquire-mcp && npm install
-npm test       # suite complète (1434 tests, ~12 s)
+npm test       # suite complète (1439 tests, ~12 s)
 npm run lint   # zéro avertissement
 npm run build  # tsc → dist/
 ```

package/README.hi.md

@@ -48,7 +48,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 > 3. **serve के दौरान शून्य क्लाउड कॉल।** मॉडल स्थानीय रूप से कैश्ड (HuggingFace से एक-बार डाउनलोड)। आपके vault की सामग्री कभी आपकी मशीन नहीं छोड़ती। डिफ़ॉल्ट रूप से एयर-गैप-सुरक्षित।
 > 4. **ताज़गी-सजग recall।** हर परिणाम बताता है कि नोट कितना पुराना है; वैकल्पिक recency re-ranking एजेंट को ताज़ा ज्ञान को प्राथमिकता देने और बासी तथ्यों को पुनः-सत्यापन हेतु चिह्नित करने देता है — भूलने-के-प्रति-सजग सीमांत, जो आपकी फ़ाइलों में पहले से मौजूद `mtime` पर बना है।
 
-**46 टूल · 19 MCP प्रॉम्प्ट · 1434+ यूनिट टेस्ट · 50+ भाषाएँ · v3.11.x स्थिर · semver-बाध्य · MIT · npm बिल्ड प्रोवेनेंस (SLSA L2)।**
+**46 टूल · 19 MCP प्रॉम्प्ट · 1439+ यूनिट टेस्ट · 50+ भाषाएँ · v3.11.x स्थिर · semver-बाध्य · MIT · npm बिल्ड प्रोवेनेंस (SLSA L2)।**
 
 ---
 

package/README.ja.md

@@ -48,7 +48,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 > 3. **serve 中はクラウド呼び出しがゼロ。** モデルはローカルにキャッシュ（HuggingFace から一度だけダウンロード）。あなたのボールトの内容はマシンから決して出ていきません。デフォルトでエアギャップ安全。
 > 4. **鮮度を意識した呼び戻し。** すべてのヒットが、そのノートがどれくらい古いかを報告します。オプトインの鮮度リランキングにより、エージェントは新しい知識を優先し、古くなった事実を再検証対象としてフラグ付けできます——これは忘却を意識したフロンティアであり、あなたのファイルがもともと持っている `mtime` の上に構築されています。
 
-**46 ツール · 19 MCP プロンプト · 1434+ ユニットテスト · 50+ 言語 · v3.11.x 安定版 · semver 準拠 · MIT · npm ビルドプロベナンス（SLSA L2）。**
+**46 ツール · 19 MCP プロンプト · 1439+ ユニットテスト · 50+ 言語 · v3.11.x 安定版 · semver 準拠 · MIT · npm ビルドプロベナンス（SLSA L2）。**
 
 ---
 
@@ -76,7 +76,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 | **GraphRAG-light**（Louvain モジュラリティによる wikilink コミュニティ検出） | ✅ **ここだけ** | ❌ | ❌ |
 | **スタンドアロンの `.base` クエリ実行**（Obsidian を起動せずに動作） | ✅ **ここだけ** | ❌ | ❌ Obsidian に委譲 |
 | **HyDE 検索**（Gao et al 2023）+ サブクエスチョン分解 | ✅ **ここだけ** | ❌ | ❌ |
-| **1434 ユニットテスト · PR ごとに 9 個の必須 + 5 個のアドバイザリ CI ゲート** | ✅ | 該当なし | まれ |
+| **1439 ユニットテスト · PR ごとに 9 個の必須 + 5 個のアドバイザリ CI ゲート** | ✅ | 該当なし | まれ |
 | **署名付きビルドプロベナンス**（npm + Sigstore、SLSA Build L2） | ✅ | 該当なし | ❌ |
 | **semver 準拠の公開サーフェス**（[STABILITY.md](./STABILITY.md)） | ✅ | 該当なし | ❌ |
 | スタンドアロン（Obsidian プラグイン不要） | ✅ | ❌ Obsidian が必要 | まちまち |
@@ -315,7 +315,7 @@ graph LR
 ```bash
 git clone https://github.com/oomkapwn/enquire-mcp.git
 cd enquire-mcp && npm install
-npm test       # フルスイート（1434 テスト、約 12 秒）
+npm test       # フルスイート（1439 テスト、約 12 秒）
 npm run lint   # 警告ゼロ
 npm run build  # tsc → dist/
 ```

package/README.md

@@ -12,10 +12,12 @@
 
 **Stop re-explaining context to Claude, Cursor, ChatGPT, Codex, OpenClaw every session. Your Obsidian notes become shared, searchable memory across every MCP-compatible agent — your knowledge, every model, forever yours.**
 
+*Measured: the BGE cross-encoder reranker adds **+15.5 NDCG@10 / +24.7 MRR** over plain hybrid on a [reproducible 60-query ablation](./docs/benchmarks.md) — the full modern IR stack, recalling the markdown **you** wrote (cited, editable), never a cloud paraphrase.*
+
 [![CI](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@oomkapwn/enquire-mcp.svg?label=npm&color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
 [![downloads](https://img.shields.io/npm/dm/@oomkapwn/enquire-mcp.svg?color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
-[![tests](https://img.shields.io/badge/tests-1434%20passing-brightgreen.svg)](#️-trust)
+[![tests](https://img.shields.io/badge/tests-1439%20passing-brightgreen.svg)](#️-trust)
 [![stable](https://img.shields.io/badge/v3.11.x-stable-brightgreen.svg)](./STABILITY.md)
 [![build provenance](https://img.shields.io/badge/build_provenance-SLSA_L2-blue.svg)](https://slsa.dev/spec/v1.0/levels#build-l2)
 [![MCP](https://img.shields.io/badge/MCP-1.29-8A2BE2.svg)](https://modelcontextprotocol.io/)
@@ -48,10 +50,10 @@ Your Obsidian vault becomes **persistent, queryable long-term memory** for any M
 > **What makes enquire-mcp different**:
 > 1. **Vendor-neutral.** Your memory lives in `.md` files. Switch from Claude to Cursor — your memory comes with you.
 > 2. **Best-in-class retrieval.** Hybrid BM25 + multilingual embeddings + BGE cross-encoder reranker fused via RRF, scaled with HNSW + int8 quantization. The same IR stack a search startup would build — open-sourced, in one binary.
-> 3. **Zero cloud calls during serve.** Models cached locally (one-time download from HuggingFace). Your vault content never leaves your machine. Air-gap-safe by default.
+> 3. **Zero cloud calls during serve.** The embedding model runs **on your machine** and indexes the markdown **you** wrote — that's why it's a one-time local download (~110 MB), not a cloud API key. Grounded + private isn't free, and we don't pretend it is: your vault content never leaves your machine, air-gap-safe by default ([enforced](./SECURITY.md), not aspirational).
 > 4. **Freshness-aware recall.** Every hit reports how old the note is; opt-in recency re-ranking lets an agent prefer fresh knowledge and flag stale facts for re-verification — the forgetting-aware frontier, built on the `mtime` your files already have.
 
-**46 tools · 19 MCP prompts · 1434 unit tests · 50+ languages · v3.11.x stable · semver-bound · MIT · npm build provenance (SLSA L2).**
+**46 tools · 19 MCP prompts · 1439 unit tests · 50+ languages · v3.11.x stable · semver-bound · MIT · npm build provenance (SLSA L2).**
 
 ---
 
@@ -79,7 +81,7 @@ Your Obsidian vault becomes **persistent, queryable long-term memory** for any M
 | **GraphRAG-light** (wikilink community detection via Louvain modularity) | ✅ **only here** | ❌ | ❌ |
 | **Standalone `.base` query execution** (works without Obsidian running) | ✅ **only here** | ❌ | ❌ delegates to Obsidian |
 | **HyDE retrieval** (Gao et al 2023) + sub-question decomposition | ✅ **only here** | ❌ | ❌ |
-| **1434 unit tests · 9 required + 5 advisory CI gates per PR** | ✅ | n/a | rare |
+| **1439 unit tests · 9 required + 5 advisory CI gates per PR** | ✅ | n/a | rare |
 | **Signed build provenance** (npm + Sigstore, SLSA Build L2) | ✅ | n/a | ❌ |
 | **Semver-bound public surface** ([STABILITY.md](./STABILITY.md)) | ✅ | n/a | ❌ |
 | Standalone (no Obsidian plugin needed) | ✅ | ❌ requires Obsidian | varies |
@@ -318,7 +320,7 @@ Channel: `npm install @oomkapwn/enquire-mcp` → latest stable (`@latest` = v3.1
 ```bash
 git clone https://github.com/oomkapwn/enquire-mcp.git
 cd enquire-mcp && npm install
-npm test       # full suite (1434 tests, ~12s)
+npm test       # full suite (1439 tests, ~12s)
 npm run lint   # zero warnings
 npm run build  # tsc → dist/
 ```

package/README.pt.md

@@ -15,7 +15,7 @@
 [![CI](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@oomkapwn/enquire-mcp.svg?label=npm&color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
 [![downloads](https://img.shields.io/npm/dm/@oomkapwn/enquire-mcp.svg?color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
-[![tests](https://img.shields.io/badge/tests-1434%20passing-brightgreen.svg)](#️-confiança)
+[![tests](https://img.shields.io/badge/tests-1439%20passing-brightgreen.svg)](#️-confiança)
 [![stable](https://img.shields.io/badge/v3.11.x-stable-brightgreen.svg)](./STABILITY.md)
 [![build provenance](https://img.shields.io/badge/build_provenance-SLSA_L2-blue.svg)](https://slsa.dev/spec/v1.0/levels#build-l2)
 [![MCP](https://img.shields.io/badge/MCP-1.29-8A2BE2.svg)](https://modelcontextprotocol.io/)
@@ -53,7 +53,7 @@ Seu vault do Obsidian se torna **memória de longo prazo persistente e consultá
 > 3. **Zero chamadas à nuvem durante o serve.** Modelos em cache local (download único do HuggingFace). O conteúdo do seu vault nunca sai da sua máquina. Seguro para ambientes isolados (air-gap) por padrão.
 > 4. **Recuperação consciente da atualidade.** Cada resultado informa quão antiga é a nota; o reranqueamento por recência opcional permite que um agente prefira conhecimento recente e sinalize fatos desatualizados para reverificação — a fronteira consciente do esquecimento, construída sobre o `mtime` que seus arquivos já têm.
 
-**46 ferramentas · 19 prompts MCP · 1434+ testes unitários · 50+ idiomas · v3.11.x estável · vinculado a semver · MIT · proveniência de build no npm (SLSA L2).**
+**46 ferramentas · 19 prompts MCP · 1439+ testes unitários · 50+ idiomas · v3.11.x estável · vinculado a semver · MIT · proveniência de build no npm (SLSA L2).**
 
 ---
 
@@ -81,7 +81,7 @@ Seu vault do Obsidian se torna **memória de longo prazo persistente e consultá
 | **GraphRAG-light** (detecção de comunidades de wikilinks via modularidade de Louvain) | ✅ **só aqui** | ❌ | ❌ |
 | **Execução autônoma de consultas `.base`** (funciona sem o Obsidian em execução) | ✅ **só aqui** | ❌ | ❌ delega ao Obsidian |
 | **Recuperação HyDE** (Gao et al. 2023) + decomposição em sub-perguntas | ✅ **só aqui** | ❌ | ❌ |
-| **1434 testes unitários · 9 gates obrigatórios + 5 consultivos de CI por PR** | ✅ | n/d | raro |
+| **1439 testes unitários · 9 gates obrigatórios + 5 consultivos de CI por PR** | ✅ | n/d | raro |
 | **Proveniência de build assinada** (npm + Sigstore, SLSA Build L2) | ✅ | n/d | ❌ |
 | **Superfície pública vinculada a semver** ([STABILITY.md](./STABILITY.md)) | ✅ | n/d | ❌ |
 | Autônomo (sem necessidade de plugin do Obsidian) | ✅ | ❌ exige Obsidian | varia |
@@ -320,7 +320,7 @@ Canal: `npm install @oomkapwn/enquire-mcp` → último estável (`@latest` = v3.
 ```bash
 git clone https://github.com/oomkapwn/enquire-mcp.git
 cd enquire-mcp && npm install
-npm test       # suíte completa (1434 testes, ~12s)
+npm test       # suíte completa (1439 testes, ~12s)
 npm run lint   # zero avisos
 npm run build  # tsc → dist/
 ```

package/README.ru.md

@@ -15,7 +15,7 @@
 [![CI](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/oomkapwn/enquire-mcp/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/@oomkapwn/enquire-mcp.svg?label=npm&color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
 [![downloads](https://img.shields.io/npm/dm/@oomkapwn/enquire-mcp.svg?color=cb3837)](https://www.npmjs.com/package/@oomkapwn/enquire-mcp)
-[![tests](https://img.shields.io/badge/tests-1434%20passing-brightgreen.svg)](#️-доверие)
+[![tests](https://img.shields.io/badge/tests-1439%20passing-brightgreen.svg)](#️-доверие)
 [![stable](https://img.shields.io/badge/v3.11.x-stable-brightgreen.svg)](./STABILITY.md)
 [![build provenance](https://img.shields.io/badge/build_provenance-SLSA_L2-blue.svg)](https://slsa.dev/spec/v1.0/levels#build-l2)
 [![MCP](https://img.shields.io/badge/MCP-1.29-8A2BE2.svg)](https://modelcontextprotocol.io/)
@@ -53,7 +53,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 > 3. **Ноль обращений в облако в режиме serve.** Модели кешируются локально (однократная загрузка с HuggingFace). Содержимое вашего хранилища никогда не покидает вашу машину. По умолчанию безопасно для изолированных сред.
 > 4. **Поиск с учётом свежести.** Каждый результат сообщает, насколько стара заметка; опциональное переранжирование по давности позволяет агенту предпочитать свежие знания и помечать устаревшие факты для повторной проверки — рубеж «осознания забывания», построенный на `mtime`, который ваши файлы уже имеют.
 
-**46 инструментов · 19 MCP-промптов · 1434+ модульных тестов · 50+ языков · стабильная ветка v3.11.x · с гарантиями semver · MIT · подтверждённая сборка в npm (SLSA L2).**
+**46 инструментов · 19 MCP-промптов · 1439+ модульных тестов · 50+ языков · стабильная ветка v3.11.x · с гарантиями semver · MIT · подтверждённая сборка в npm (SLSA L2).**
 
 ---
 
@@ -81,7 +81,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 | **GraphRAG-light** (детекция сообществ по wikilink-графу через модулярность Лувена) | ✅ **только здесь** | ❌ | ❌ |
 | **Автономное выполнение `.base`-запросов** (работает без запущенного Obsidian) | ✅ **только здесь** | ❌ | ❌ делегирует Obsidian |
 | **HyDE-поиск** (Gao et al 2023) + декомпозиция на подвопросы | ✅ **только здесь** | ❌ | ❌ |
-| **1434 модульных тестов · 9 обязательных + 5 совещательных CI-гейтов на каждый PR** | ✅ | н/д | редко |
+| **1439 модульных тестов · 9 обязательных + 5 совещательных CI-гейтов на каждый PR** | ✅ | н/д | редко |
 | **Подтверждённое происхождение сборки** (npm + Sigstore, SLSA Build L2) | ✅ | н/д | ❌ |
 | **Публичная поверхность с гарантиями semver** ([STABILITY.md](./STABILITY.md)) | ✅ | н/д | ❌ |
 | Автономность (плагин Obsidian не нужен) | ✅ | ❌ требует Obsidian | по-разному |
@@ -320,7 +320,7 @@ graph LR
 ```bash
 git clone https://github.com/oomkapwn/enquire-mcp.git
 cd enquire-mcp && npm install
-npm test       # полный набор (1434 тестов, ~12с)
+npm test       # полный набор (1439 тестов, ~12с)
 npm run lint   # ноль предупреждений
 npm run build  # tsc → dist/
 ```

package/README.zh.md

@@ -48,7 +48,7 @@ claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault ~/Document
 > 3. **serve 期间零云端调用。** 模型本地缓存（一次性从 HuggingFace 下载）。你的仓库内容永不离开本机。默认即可离线/隔离运行。
 > 4. **时效感知召回。** 每条结果都报告笔记有多旧；可选的时效重排让智能体优先采用新知识，并把陈旧事实标记出来等待复核——这是"遗忘感知"前沿，建立在你的文件本就拥有的 `mtime` 之上。
 
-**46 个工具 · 19 个 MCP 提示词 · 1434+ 单元测试 · 50+ 语言 · v3.11.x 稳定版 · 语义化版本约束 · MIT · npm 构建溯源（SLSA L2）。**
+**46 个工具 · 19 个 MCP 提示词 · 1439+ 单元测试 · 50+ 语言 · v3.11.x 稳定版 · 语义化版本约束 · MIT · npm 构建溯源（SLSA L2）。**
 
 ---
 

package/dist/index.d.ts

@@ -7,7 +7,7 @@
  * + `McpServer({version})`) and `src/tool-registry.ts` (used in the
  * `vault-info` resource payload).
  */
-export declare const VERSION = "3.11.1";
+export declare const VERSION = "3.11.2";
 export { main } from "./cli.js";
 export { buildEmbedText, buildMcpServer, formatReadyBanner, prepareServerDeps, type ServeOptions, type ServerDeps, startServer } from "./server.js";
 export { parsePositiveInt, parseQuantizationMode } from "./tool-registry.js";

package/dist/index.js

@@ -40,7 +40,7 @@ import { main } from "./cli.js";
  * + `McpServer({version})`) and `src/tool-registry.ts` (used in the
  * `vault-info` resource payload).
  */
-export const VERSION = "3.11.1";
+export const VERSION = "3.11.2";
 // Re-exports — preserve the v3.5.x public surface so http-transport.ts and
 // tests don't need to know about the new module layout. The set below
 // exactly matches the v3.5.x `export` declarations: `main`,

package/docs/COMPARISON.md

@@ -44,7 +44,7 @@ The four axes the external audit (#3, 2026-05) called out as decisive — **REST
 | Zero outbound network calls in serve mode     | **Yes** (default)   | Local-only (REST)| Local-only (REST)| Yes              | Yes              |
 | Signed build provenance on releases (SLSA L2) | **Yes**             | No               | No               | No               | No               |
 | Forgetting-aware freshness (`age_days` / recency re-rank) | Yes (v3.10)     | No               | No               | No               | No               |
-| Test count (public)                           | **1434**             | (varies)         | (varies)         | (varies)         | (varies)         |
+| Test count (public)                           | **1439**             | (varies)         | (varies)         | (varies)         | (varies)         |
 | Tool count                                    | 46                  | ~25              | ~8               | ~10              | 3–5              |
 | MCP prompt count                              | 19                  | 0                | 0                | 0                | 0                |
 | License                                       | MIT                 | Apache-2.0       | MIT              | MIT              | (varies)         |

package/docs/EVALUATION_PLAN.md

@@ -0,0 +1,127 @@
+# Evaluation plan — enquire-mcp
+
+**Status:** pre-registered methodology · **no scores in this document by design.**
+**Issued:** 2026-06-29 · **Scope:** how we will measure enquire-mcp's quality, what
+counts as a pass/fail, and the bar a number must clear before it appears in the
+README, `docs/benchmarks.md`, or any public claim.
+
+This is a *plan*, committed **before** the headline runs, so the method can be
+reviewed without front-running a result. It exists because the project's core
+brand is **"every claim an auditor can verify"** — and a pre-registered protocol
+is how you make a benchmark falsifiable instead of a marketing artifact.
+
+---
+
+## 0. Principles (the bar every number must clear)
+
+1. **Measured — never asserted or estimated.** A figure is published only after it
+   is produced by a committed, reproducible harness. No placeholders, no "~", no
+   borrowed competitor numbers.
+2. **Reproducible.** Every reported number ships with the exact command, the
+   harness, the dataset pointer (or a synthetic generator), and a determinism
+   contract (fixed bodies, fixed mtimes, pinned model alias).
+3. **Reviewed.** Headline numbers (LongMemEval, context-efficiency, any
+   cross-tool comparison) go through a maintainer **reference-hardware run +
+   sign-off** before they leave `benchmarks.md` for the README hero.
+4. **Scoped to what we are.** enquire-mcp is a **retriever / memory layer**, not an
+   answer-generating assistant. We report **retrieval quality**, not end-to-end QA
+   accuracy — reporting a QA number for a retriever would be an overclaim (the
+   answer is the calling agent's job).
+5. **Zero result = FAIL, not "n/a".** A query that returns nothing for a question
+   that has a ground-truth answer scores 0 and is counted, never silently dropped.
+6. **Disclose weaknesses.** Per-class breakdowns report where recall is *weaker*
+   (e.g. OCR'd PDF vs native markdown), because honest failure disclosure raises
+   credibility — it does not lower it.
+
+---
+
+## 1. Dimensions we measure
+
+| dimension | what it answers | harness | metric |
+|---|---|---|---|
+| **Retrieval quality** | does the right note rank highly? | `scripts/run-benchmarks.mjs` (`bench:retrieval`) | MRR · NDCG@10 · Recall@10 |
+| **Long-term memory** | does recall hold over long, multi-session histories? | `scripts/bench-longmemeval.mjs` (`bench:longmemeval`) | recall@k / MRR / NDCG@k of the answer-bearing session(s) |
+| **Context efficiency** | how much less context vs read-every-hit? | `scripts/bench-context.mjs` (`bench:context`) | token ratio (baseline-read ÷ pack) |
+| **Freshness awareness** | is a stale fact flagged as stale? | `src/staleness.ts` + `age_days`/`stale` on every hit | stale-flag precision/recall (planned) |
+| **Reranker contribution** | does the cross-encoder help? | the ablation in `run-benchmarks.mjs` | Δ NDCG@10 / Δ MRR (fusion → fusion+rerank) |
+
+The **single-source-of-truth metric implementations** live in
+[`src/eval.ts`](https://github.com/oomkapwn/enquire-mcp/blob/main/src/eval.ts)
+(`recallAtK` / `ndcgAtK` / `reciprocalRank`, dedup-safe), used identically by
+every harness and the `enquire-mcp eval` CLI subcommand — so a number cannot drift
+between the bench and the shipped tool.
+
+---
+
+## 2. Datasets & conditions
+
+- **Synthetic vault** (committed generators) — deterministic, for the ablation +
+  the context-efficiency *method* demonstration. Numbers from it are illustrative,
+  never the headline (see §4).
+- **LongMemEval** ([arXiv:2410.10813](https://arxiv.org/abs/2410.10813)) — the
+  standard long-term-memory benchmark; `longmemeval_s / _m / _oracle`. Not
+  committed (size + licensing); downloaded per the `benchmarks.md` instructions.
+  Conditions reported separately by `question_type` (single-session, multi-session,
+  temporal-reasoning, knowledge-update, abstention).
+- **LoCoMo** (planned) — a second long-conversation memory set, to avoid
+  over-fitting the protocol to one benchmark's quirks.
+- **Representative real vault** — for the context-efficiency headline only, run on
+  a real (private) vault on reference hardware; only the *ratio* is reported, never
+  vault content.
+
+---
+
+## 3. Grading protocol
+
+- **Retrieval metrics** are computed directly from the ground-truth relevant set
+  (no LLM judge needed) — deterministic and exactly reproducible.
+- **LLM-as-judge** (only where a judgment is unavoidable, e.g. answer-bearing-span
+  relevance that isn't in the labels): use a fixed judge model + temperature,
+  **swap the position** of the two items being compared and average both orders to
+  cancel position bias, and report inter-run agreement. A judgment that flips on
+  the swap is recorded as "uncertain", not silently resolved.
+- **Token efficiency** reports **two ratios** — capped-pack vs (a) full-body reads
+  of the same hits and (b) a naive whole-folder read baseline — so the number is
+  bounded, not a single best-case scenario.
+- **Abstention** (`*_abs`) questions are scored separately: the correct behavior is
+  *no confident hit*, so they are not folded into recall.
+
+---
+
+## 4. Publication gate (what reaches the README)
+
+A number moves from "the harness can produce it" to "published" only when **all**
+hold:
+
+1. produced by the committed harness with the determinism contract met;
+2. run on **reference hardware** against a **representative** dataset (not the
+   synthetic vault, for any headline);
+3. method **reviewed** + the figure **signed off** by the maintainer;
+4. pinned by a structural guard where a count/claim is involved (the
+   `docs-consistency` / OIA invariants), so it can't silently drift later.
+
+Until then, the harness + its pure-function tests ship, `benchmarks.md` documents
+the **method**, and the README does **not** assert the figure. Current
+maintainer-gated headlines awaiting a reference run: **LongMemEval retrieval** and
+**context-efficiency ("Nx less context")**.
+
+---
+
+## 5. Deliberately out of scope
+
+- **End-to-end QA accuracy** — that measures the *generating agent*, not the memory
+  layer. We report retrieval, and let the agent's own evals own the answer.
+- **Borrowed metrics** — no competitor's figure is ever restated as ours.
+- **Single best-case scenarios** as the headline — every efficiency claim is
+  reported as a bounded ratio with the baseline named.
+
+---
+
+## Related
+
+- [`docs/benchmarks.md`](./benchmarks.md) — the live results + full methodology +
+  reproducibility contract.
+- [`src/eval.ts`](https://github.com/oomkapwn/enquire-mcp/blob/main/src/eval.ts) —
+  the metric implementations shared by every harness.
+- `scripts/run-benchmarks.mjs` · `scripts/bench-longmemeval.mjs` ·
+  `scripts/bench-context.mjs` — the harnesses.

package/docs/benchmarks.md

@@ -486,6 +486,35 @@ centerpiece, so it goes through the same "measured, reproducible, reviewed"
 bar as every other number in this document — never a placeholder or an
 estimate).
 
+## Context efficiency (token cost)
+
+Code-intelligence memory tools advertise a flagship "Nx fewer tokens" figure —
+the context an agent saves by querying a memory layer instead of reading source
+files one by one. The same lever applies to a **notes** memory: `obsidian_context_pack`
+returns a single budget-capped bundle (`budget_tokens × 4` chars, default 4000)
+instead of the agent's naive *search → `read_note` every top hit in full* loop.
+
+**What we measure.** Per question, two paths over the **same** hybrid retrieval:
+(a) the `context_pack` bundle's token count, vs (b) the summed full bodies of the
+top-K hits the agent would otherwise open. Token estimate is `chars / 4` (a rough,
+deterministic heuristic — the same on both sides, so the *ratio* is robust to it).
+
+**Harness.** [`scripts/bench-context.mjs`](https://github.com/oomkapwn/enquire-mcp/blob/main/scripts/bench-context.mjs)
+(shipped v3.11.2) builds a deterministic synthetic vault, indexes FTS5 +
+embeddings, and reports `baseline_tokens` / `pack_tokens` / savings ratio per
+question + the mean (pure helpers covered by `tests/bench-context.test.ts` with
+NEGATIVE controls so an inflated ratio can't ship).
+
+```bash
+npm run bench:context
+```
+
+**Status:** the harness ships in v3.11.2; **no headline "Nx less context" figure
+is published** from a synthetic-vault run. A README claim requires a run on a
+**representative vault** + maintainer sign-off — the same "measured, reproducible,
+reviewed" bar as the LongMemEval score above. The synthetic-vault number the
+script prints is an illustration of the *method*, not a claim about real vaults.
+
 ## Future work
 
 - Reproduce on **public BEIR / TREC subsets** so numbers can be compared

package/package.json

@@ -1,9 +1,9 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "@oomkapwn/enquire-mcp",
-  "version": "3.11.1",
+  "version": "3.11.2",
   "mcpName": "io.github.oomkapwn/enquire-mcp",
-  "description": "MCP server giving AI agents (Claude Code, Claude Desktop, Cursor, ChatGPT, Codex, OpenClaw) persistent long-term memory backed by your local Obsidian markdown vault. Hybrid retrieval (BM25 + ML embeddings + BGE reranker, RRF-fused), HNSW + int8 quantization, agentic RAG (HyDE + sub-question decomposition), GraphRAG-light (Louvain), standalone Obsidian Bases, PDFs + Tesseract OCR. Vendor-neutral memory layer for any MCP-compatible agent. 46 tools, 19 MCP prompts, 1434 tests, signed npm build provenance (SLSA L2), semver-bound, MIT, zero cloud calls during serve.",
+  "description": "MCP server giving AI agents (Claude Code, Claude Desktop, Cursor, ChatGPT, Codex, OpenClaw) persistent long-term memory backed by your local Obsidian markdown vault. Hybrid retrieval (BM25 + ML embeddings + BGE reranker, RRF-fused), HNSW + int8 quantization, agentic RAG (HyDE + sub-question decomposition), GraphRAG-light (Louvain), standalone Obsidian Bases, PDFs + Tesseract OCR. Vendor-neutral memory layer for any MCP-compatible agent. 46 tools, 19 MCP prompts, 1439 tests, signed npm build provenance (SLSA L2), semver-bound, MIT, zero cloud calls during serve.",
   "type": "module",
   "bin": {
     "enquire-mcp": "dist/index.js"
@@ -51,6 +51,7 @@
     "docs/QUICKSTART.md",
     "docs/COMPARISON.md",
     "docs/benchmarks.md",
+    "docs/EVALUATION_PLAN.md",
     "docs/http-transport.md",
     "examples",
     "assets/social-preview.png",
@@ -91,6 +92,7 @@
     "bench:retrieval": "npm run build && node scripts/run-benchmarks.mjs",
     "bench:peek-cache": "npm run build && node scripts/bench-peek-cache.mjs",
     "bench:longmemeval": "npm run build && node scripts/bench-longmemeval.mjs",
+    "bench:context": "npm run build && node scripts/bench-context.mjs",
     "docs:api": "typedoc",
     "check:audit": "node scripts/check-audit.mjs",
     "prepare": "rm -rf dist && tsc && chmod +x dist/index.js && (husky 2>/dev/null || true)",
@@ -121,6 +123,12 @@
     "chatgpt",
     "openclaw",
     "codex",
+    "windsurf",
+    "zed",
+    "aider",
+    "gemini-cli",
+    "opencode",
+    "kilocode",
     "ai-agent",
     "ai-agents",
     "ai-search",