omnilane 0.10.3 → 0.10.4

package/CHANGELOG.md

@@ -6,6 +6,42 @@ semantic version tags.
 
 ## [Unreleased]
 
+## [0.10.4] - 2026-07-26
+
+No lane ordering changes. This release corrects documentation that could send
+work to the wrong model, and records the evidence behind the shipped defaults.
+
+### Changed
+
+- Narrowed the `long-context` lane description in `routing.yaml` and in all five
+  localized README lane tables. It previously called the lane long-document
+  *synthesis* while shipping Gemini first; published multi-needle scores at 1M
+  favour Claude by roughly threefold, while Gemini leads single-needle
+  retrieval. These are different capabilities with different leaders, and the
+  old wording pointed multi-hop work at the wrong candidate. The lane now
+  describes retrieval and volume sweeps, and names the Claude candidate for
+  integration across scattered sources. Ordering is deliberately unchanged: the
+  supporting evidence is secondary and covers prior model generations, which is
+  not a sufficient basis for moving a shipped default.
+
+### Fixed
+
+- `docs/model-capabilities-2026-07.md` quoted the Artificial Analysis Coding
+  Agent Index at v1.1 while the index had re-based twice more. The same model
+  reads 80, 78 or 67 depending on the version and harness a source used, so the
+  figure is not portable across versions. The section now records every observed
+  value with its provenance, documents the v1.3 composition, and states that the
+  index may be cited for ordering but never for a number.
+- Documented the writing evidence behind `taste-final`, which previously rested
+  entirely on general and agentic indexes that do not measure prose. Added
+  EQ-Bench Creative Writing v3, EQ-Bench Longform, and the Lech Mazur
+  story-writing benchmark, each read from the publisher. Recorded that Mazur has
+  not yet evaluated Claude Opus 5 and that Claude Fable 5 leads that board, so
+  the question stays open rather than being presented as settled.
+- Added per-effort cost and throughput to the Intelligence Index table, which is
+  what justifies the shipped effort levels: `xhigh` reaches the same index score
+  as `max` for 30% (Opus 5) and 53% (GPT-5.6 Sol) less per task.
+
 ## [0.10.3] - 2026-07-26
 
 ### Changed
@@ -411,7 +447,8 @@ semantic version tags.
 - Initial shared routing table, cross-vendor dispatcher, runners, installer,
   and baseline lint fixes.
 
-[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.10.3...HEAD
+[Unreleased]: https://github.com/Seraphim0916/omnilane/compare/v0.10.4...HEAD
+[0.10.4]: https://github.com/Seraphim0916/omnilane/compare/v0.10.3...v0.10.4
 [0.10.3]: https://github.com/Seraphim0916/omnilane/compare/v0.10.2...v0.10.3
 [0.10.2]: https://github.com/Seraphim0916/omnilane/compare/v0.10.1...v0.10.2
 [0.10.1]: https://github.com/Seraphim0916/omnilane/compare/v0.9.1...v0.10.1

package/README.ja.md

@@ -119,7 +119,7 @@ flowchart LR
 | ✒️ taste-final | Claude Opus 5 (high) | GPT-5.6 Sol (max) | 対外文章、prompt/ドキュメント推敲、スタイル最終審 |
 | 💬 consult | 明示指定したベンダー/モデル | —(フォールバックなし) | 自然言語で直接相談。`--vendor` を必ず維持 |
 | 🎨 ui-draft | GPT-5.6 Sol (xhigh) | Claude Opus 5 (high) | デザインシステム/参考画像がある場合の UI ドラフト |
-| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 100 万トークン級の長文統合。Pro も agentic 対応、高速反復ループは Flash を優先 |
+| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 100 万トークン級の走査と検索。複数箇所をまたぐ統合には Claude 候補を、高速反復ループは Flash を優先 |
 | ⚡ fast-agentic | Gemini 3.6 Flash (High) | GPT-5.6 Luna (high) | 高速なマルチステップ agentic ループ、マルチモーダル確認 |
 | 📡 live-search | Grok 4.5 | —(off) | リアルタイム X/ウェブ検索とソーシャル文脈 |
 | 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus → OpenCode | Codex クォータ逼迫時の中級コーディング逃し弁 |
@@ -503,6 +503,23 @@ scripts/dispatch.sh --dry-run hardest-coding "…"   # 解決済みプラン、
 
 ## 📜 リリース履歴
 
+## v0.10.4 の新機能
+
+- **`long-context` がマルチホップ作業を誤ったモデルに向けなくなりました** — この
+  レーンは長文*統合*を名乗りながら Gemini を第一候補にしていましたが、公開されて
+  いる 1M トークンのマルチニードル評価では Claude が約 3 倍のスコアを示し、Gemini
+  が強いのはシングルニードル検索です。レーンの説明を走査と検索に改め、複数箇所を
+  またぐ統合には Claude 候補を案内します。順序は意図的に据え置き — 根拠が二次情報
+  であり、前世代モデルの測定に基づくためです。
+- **Coding Agent Index を数値として引用しなくなりました** — 同一モデルがバージョン
+  とハーネス次第で 80、78、67 と読み取れます。今後は順序の参照のみに用い、観測値
+  ごとの出所を記録しています。
+- **`taste-final` に文章特化の根拠を追加** — 従来は散文を測らない汎用・エージェント
+  指標のみで順序を決めていました。EQ-Bench Creative Writing v3、EQ-Bench Longform、
+  Lech Mazur の 3 ボードを、いずれも公開元から直接取得して追加しました。
+- **努力度ごとのコストとスループットを追加**。既定が `xhigh` である理由を示します:
+  `max` と同じ指数スコアを、1 タスクあたり 30-53% 安く得られます。
+
 ## v0.10.3 の新機能
 
 - **5 言語すべての README を再構成** — 冒頭で「これは何か、なぜ欲しくなるのか」を

package/README.ko.md

@@ -117,7 +117,7 @@ flowchart LR
 | ✒️ taste-final | Claude Opus 5 (high) | GPT-5.6 Sol (max) | 대외 문장, prompt/문서 다듬기, 스타일 최종심 |
 | 💬 consult | 명시적으로 지정한 벤더/모델 | —(폴백 없음) | 자연어 직접 상담. `--vendor` 를 반드시 유지 |
 | 🎨 ui-draft | GPT-5.6 Sol (xhigh) | Claude Opus 5 (high) | 디자인 시스템/참고 이미지가 있을 때의 UI 초안 |
-| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 100만 토큰급 장문 통합. Pro 도 agentic 지원, 빠른 반복 루프는 Flash 우선 |
+| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 100만 토큰급 훑기와 검색. 여러 곳을 잇는 다중 홉 통합은 Claude 후보를, 빠른 반복 루프는 Flash 우선 |
 | ⚡ fast-agentic | Gemini 3.6 Flash (High) | GPT-5.6 Luna (high) | 빠른 멀티스텝 agentic 루프, 멀티모달 확인 |
 | 📡 live-search | Grok 4.5 | —(off) | 실시간 X/웹 검색과 소셜 맥락 |
 | 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus → OpenCode | Codex 쿼터 소진 시 중급 코딩 안전 밸브 |
@@ -487,6 +487,22 @@ scripts/dispatch.sh --dry-run hardest-coding "…"   # 완전히 해석된 계
 
 ## 📜 릴리스 기록
 
+## v0.10.4 새 기능
+
+- **`long-context` 가 더 이상 다중 홉 작업을 잘못된 모델로 보내지 않습니다** — 이
+  레인은 장문 *통합*을 표방하면서 Gemini 를 1순위로 두었지만, 공개된 1M 토큰 멀티
+  니들 점수는 Claude 가 약 3배 앞서고 Gemini 가 강한 쪽은 싱글 니들 검색입니다.
+  레인 설명을 훑기와 검색으로 좁히고, 여러 곳을 잇는 통합에는 Claude 후보를
+  안내합니다. 순서는 의도적으로 유지 — 근거가 2차 자료이고 이전 세대 모델을 측정한
+  것이기 때문입니다.
+- **Coding Agent Index 를 수치로 인용하지 않습니다** — 같은 모델이 버전과 하네스에
+  따라 80, 78, 67 로 읽힙니다. 이제 순서 참조로만 쓰고, 관측값마다 출처를 기록합니다.
+- **`taste-final` 에 글쓰기 전용 근거 추가** — 이전에는 산문을 측정하지 않는 범용·
+  에이전트 지표만으로 순서를 정했습니다. EQ-Bench Creative Writing v3, EQ-Bench
+  Longform, Lech Mazur 세 보드를 모두 발행처에서 직접 가져와 추가했습니다.
+- **노력 수준별 비용과 처리량 추가**. 기본값이 `xhigh` 인 이유를 보여줍니다:
+  `max` 와 같은 지수 점수를 작업당 30-53% 저렴하게 얻습니다.
+
 ## v0.10.3 새 기능
 
 - **5개 언어 README 전면 재구성** — 문서 첫머리에서 "이게 무엇이고 왜 필요한가"를

package/README.md

@@ -122,7 +122,7 @@ actually resolves.
 | ✒️ taste-final | Claude Opus 5 (high) | GPT-5.6 Sol (max) | User-facing prose, prompt/doc polish, style arbitration |
 | 💬 consult | Explicit named vendor/model | — (no fallback) | Direct natural-language consultation; always keep `--vendor` |
 | 🎨 ui-draft | GPT-5.6 Sol (xhigh) | Claude Opus 5 (high) | UI drafts only WITH a design system / reference images |
-| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 1M-token synthesis; Pro is agentic-capable, while fast repeated loops prefer Flash |
+| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 1M-token sweeps and retrieval; for multi-hop synthesis prefer the Claude candidate, and Flash for fast repeated loops |
 | ⚡ fast-agentic | Gemini 3.6 Flash (High) | GPT-5.6 Luna (high) | Fast multi-step agentic loops, multimodal checks |
 | 📡 live-search | Grok 4.5 | — (off) | Realtime X/web search and social context |
 | 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus → OpenCode | Codex-quota relief valve for mid-tier coding |
@@ -533,6 +533,25 @@ working notes, including per-benchmark caveats, live in
 
 ## 📜 Release history
 
+## What's new in v0.10.4
+
+- **`long-context` no longer points multi-hop work at the wrong model** — the
+  lane called itself long-document *synthesis* while shipping Gemini first, but
+  published multi-needle scores at 1M favour Claude by roughly threefold while
+  Gemini leads single-needle retrieval. The lane now describes retrieval and
+  volume sweeps and names the Claude candidate for integration work. Ordering is
+  unchanged; the evidence is secondary and covers prior model generations.
+- **The Coding Agent Index is no longer quoted as a number** — the same model
+  reads 80, 78 or 67 depending on index version and harness. It is now cited for
+  ordering only, with every observed value and its provenance recorded.
+- **`taste-final` has writing evidence behind it** — previously ordered from
+  general and agentic indexes that do not measure prose. Added EQ-Bench Creative
+  Writing v3, EQ-Bench Longform, and the Lech Mazur benchmark, read from the
+  publishers.
+- **Per-effort cost and throughput** added to the model notes, showing why the
+  defaults use `xhigh`: it reaches the same index score as `max` for 30-53% less
+  per task.
+
 ## What's new in v0.10.3
 
 - **Restructured READMEs in all five languages** — the reader now meets a plain

package/README.zh-CN.md

@@ -110,7 +110,7 @@ flowchart LR
 | ✒️ taste-final | Claude Opus 5 (high) | GPT-5.6 Sol (max) | 对外文字、prompt 与文档打磨、风格终审 |
 | 💬 consult | 明确指定的厂商/模型 | —(不降级) | 自然语言直接咨询;必须保留 `--vendor` |
 | 🎨 ui-draft | GPT-5.6 Sol (xhigh) | Claude Opus 5 (high) | 有设计规范/参考图时的 UI 出稿;开放式视觉品味交给 taste-final |
-| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 百万 token 长文整合;Pro 可跑 agentic 工作,高速重复循环仍优先 Flash |
+| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 百万 token 扫读与检索;要跨段落多跳整合请改用 Claude 候选,高速重复循环仍优先 Flash |
 | ⚡ fast-agentic | Gemini 3.6 Flash (High) | GPT-5.6 Luna (high) | 快速多步骤 agentic 循环、多模态检查 |
 | 📡 live-search | Grok 4.5 | —(off) | 实时 X/网络搜索与社群脉络 |
 | 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus → OpenCode | Codex 额度吃紧时的中量级编码溢流道;事实性声明须另行查证 |
@@ -462,6 +462,20 @@ plan 模式,或只给只读工具集)。要改文件必须同时给 `--mode work
 
 ## 📜 版本历程
 
+## v0.10.4 新功能
+
+- **`long-context` 不再把多跳任务指向错的模型**——这条通道原本自称长文*整合*,
+  首选却是 Gemini;但公开的百万 token 多针分数显示 Claude 领先约三倍,Gemini
+  强的是单针检索。通道说明已改为扫读与检索,跨来源整合请改用 Claude 候选。
+  排序刻意不动:那批证据是二手且测的是上一代模型,不足以移动已发布的默认值。
+- **Coding Agent Index 不再被当数字引用**——同一个模型在不同版本与 harness 下
+  读出 80、78、67 三种值。现在只用来看排序,并记录每个观测值的出处。
+- **`taste-final` 补上写作专项证据**——此前完全靠不测文笔的通用与 agentic 指标
+  排序。已加入 EQ-Bench Creative Writing v3、EQ-Bench Longform 与 Lech Mazur
+  三个榜,全部取自发布者第一手。
+- **补上逐档位成本与吞吐**,说明默认为何用 `xhigh`:拿到与 `max` 相同的指数
+  分数,每任务却便宜 30-53%。
+
 ## v0.10.3 新功能
 
 - **五种语言的 README 全面重整**——文档开头改成先讲清楚「这是什么、我为什么会

package/README.zh-TW.md

@@ -110,7 +110,7 @@ flowchart LR
 | ✒️ taste-final | Claude Opus 5 (high) | GPT-5.6 Sol (max) | 對外文字、prompt 與文件打磨、風格終審 |
 | 💬 consult | 明確點名的廠商/模型 | —(不降級) | 自然語言直接諮詢;必須保留 `--vendor` |
 | 🎨 ui-draft | GPT-5.6 Sol (xhigh) | Claude Opus 5 (high) | 有設計規範/參考圖時的 UI 出稿;開放式視覺品味交給 taste-final |
-| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 百萬 token 長文整合;Pro 可跑 agentic 工作,高速重複迴圈仍優先 Flash |
+| 📚 long-context | Gemini 3.1 Pro (High) | Claude Opus 5 (high) | 百萬 token 掃讀與檢索;要跨段落多跳整合請改用 Claude 候選,高速重複迴圈仍優先 Flash |
 | ⚡ fast-agentic | Gemini 3.6 Flash (High) | GPT-5.6 Luna (high) | 快速多步驟 agentic 迴圈、多模態檢查 |
 | 📡 live-search | Grok 4.5 | —(off) | 即時 X/網路搜尋與社群脈絡 |
 | 🚰 coding-overflow | Grok 4.5 | Kimi K3 → Qwen3 Coder Plus → OpenCode | Codex 額度吃緊時的中量級編碼溢流道;事實性宣稱須另行查證 |
@@ -472,6 +472,20 @@ plan 模式,或只給唯讀工具集)。要改檔必須同時給 `--mode work`
 
 ## 📜 版本歷程
 
+## v0.10.4 新功能
+
+- **`long-context` 不再把多跳任務指向錯的模型**——這條通道原本自稱長文*整合*,
+  首選卻是 Gemini;但公開的百萬 token 多針分數顯示 Claude 領先約三倍,Gemini
+  強的是單針檢索。通道說明已改為掃讀與檢索,跨來源整合請改用 Claude 候選。
+  排序刻意不動:那批證據是二手且測的是上一代模型,不足以移動已出貨的預設。
+- **Coding Agent Index 不再被當數字引用**——同一個模型在不同版本與 harness 下
+  讀出 80、78、67 三種值。現在只用來看排序,並記錄每個觀測值的出處。
+- **`taste-final` 補上寫作專項證據**——先前完全靠不測文筆的通用與 agentic 指標
+  排序。已加入 EQ-Bench Creative Writing v3、EQ-Bench Longform 與 Lech Mazur
+  三個榜,全部取自發布者第一手。
+- **補上逐檔位成本與吞吐**,說明預設為何用 `xhigh`:拿到與 `max` 相同的指數
+  分數,每任務卻便宜 30-53%。
+
 ## v0.10.3 新功能
 
 - **五種語言的 README 全面重整**——文件開頭改成先講清楚「這是什麼、我為什麼會

package/VERSION

@@ -1 +1 @@
-0.10.3
+0.10.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnilane",
-  "version": "0.10.3",
+  "version": "0.10.4",
   "description": "One routing table, every harness — classify subtasks into lanes and dispatch each lane to the best vendor's agentic CLI (Codex, Claude, Gemini, Grok) using your existing subscription logins.",
   "bin": {
     "omnilane": "bin/omnilane"

package/routing.yaml

@@ -24,7 +24,7 @@ hard-judgment:   claude claude-opus-5 xhigh | codex gpt-5.6-sol max   # AA Intel
 taste-final:     claude claude-opus-5 high | codex gpt-5.6-sol max    # user-facing prose, prompt/doc polish, Chinese phrasing, style arbitration
 consult:         codex gpt-5.6-sol max | claude claude-opus-5 high | grok grok-4.5 - | gemini "Gemini 3.1 Pro (High)" -  # direct named-model consultation; use --vendor to prevent fallback
 ui-draft:        codex gpt-5.6-sol xhigh | claude claude-opus-5 high  # only with a design system / reference images; open-ended visual taste -> taste-final
-long-context:    gemini "Gemini 3.1 Pro (High)" - | claude claude-opus-5 high | codex gpt-5.6-sol high  # all have 1M context; prefer Gemini for cost + long-doc synthesis, Flash for fast loops
+long-context:    gemini "Gemini 3.1 Pro (High)" - | claude claude-opus-5 high | codex gpt-5.6-sol high  # all have 1M context; Gemini leads single-needle retrieval at full length and is the cheapest way to sweep volume, Flash for fast loops. For multi-hop synthesis across a large corpus prefer the claude candidate: published multi-needle scores at 1M favour Claude by a wide margin (see docs/model-capabilities-2026-07.md)
 fast-agentic:    gemini "Gemini 3.6 Flash (High)" - | codex gpt-5.6-luna high  # speed + agentic tool loops; 3.6 Flash (released 2026-07-21): AA Intelligence 50, #1 output speed 303.6 tok/s, -17% output tokens vs 3.5 Flash per Google
 live-search:     grok grok-4.5 - | off          # native X/web search lane; no real substitute
 coding-overflow: grok grok-4.5 - | kimi kimi-k3 - | qwen qwen3-coder-plus - | opencode - - | off  # codex-quota relief valve: mid-tier coding; Grok 4.5 (GA 2026-07-16) Terminal-Bench 83.3 / SWE-Bench Pro 64.7, but AA hallucination 54% — verify factual claims. qwen3-coder-plus = 2025-09-23 snapshot alias (Qwen 3.6 Plus exists; re-evaluate before swapping). kimi/qwen model fields are CLI aliases — adjust to your login. opencode "-" model = its own configured default.