dsh-community-bench 0.1.1 → 0.2.0

package/CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## 0.2.0 — 2026-10-18
+
+- Brand the consumer as part of `dsh-score`.
+- Show GOE (`E0`–`E5`) and `unrated`; no prior-only `#1` or quality rank.
+- If a surface has no positive measured result, render "No measured evidence yet" instead of prior cards.
+- Relabel confidence as evidence completeness; show the discovery score separately.
+
 ## 0.1.1 — 2026-10-18
 
 - npm-first install docs and live registry links.

package/README-zh.md

@@ -1,6 +1,9 @@
 # dsh-community-bench
 
-**DeepSeek Harness 的应用内社区基准标签页。** 它读取公开的、按任务条件计算的 [DSH Community Bench](https://perrylink.github.io/dsh-attention-index/) registry，展示所有人看到的同一份分数。它**不在本地算分**。
+**DeepSeek Harness 的应用内社区基准标签页，面向 `dsh-score` 公共证据层。** 它读取公开的 [dsh-score](https://perrylink.github.io/dsh-attention-index/) registry，展示所有人看到的同一份证据。它**不在本地算分**。
+
+> **状态：参考消费者 / 维护模式。** 不再新增评分或排名功能。公共证据层是
+> `dsh-score-index`（当前为 `dsh-attention-index`；repo 改名待执行）。
 
 [![npm](https://img.shields.io/npm/v/dsh-community-bench)](https://www.npmjs.com/package/dsh-community-bench)
 [![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE)
@@ -10,12 +13,13 @@
 
 ## 展示什么
 
-- 每个能力面的前 3 名（记忆、视觉、Office、浏览器、编排、安全、MCP、远程、开发、语音、研究、UI、桌面）。
-- 社区分：有实测为 0–100，仅有先验为 0–60；同时展示 tier、prior、bench、覆盖率和置信度。
+- 每个能力面的 GOE（`E0`–`E5`）。v0.2 里所有行都是 `E0 / unrated`。
+- 只有 positive measured 进入质量块；先验行是明确标注的候选，永远不是 `#1`。
+- prior、bench、覆盖率和**证据完整度**（不是置信概率）。
 - 社区注意力席位和贡献者声誉。
-- 公共排行榜链接和刷新按钮。
+- 公共证据页链接和刷新按钮。
 
-这个标签页只是**消费端**：测量和分配都在公共 registry 里，不在本插件里。在线 registry 不可达时回退到内置快照，并明确标注。
+这个标签页只是**消费端**：证据和评级都在公共 registry 里，不在本插件里。在线 registry 不可达时回退到内置快照，并明确标注。
 
 ## 安装
 

package/README.md

@@ -1,6 +1,10 @@
 # dsh-community-bench
 
-**In-app Community Bench tab for DeepSeek Harness.** It reads the public, task-conditioned [DSH Community Bench](https://perrylink.github.io/dsh-attention-index/) registry and shows the same score everyone sees. It does **not** compute a per-user score.
+**In-app Community Bench tab for the `dsh-score` public evidence layer.** It reads the public [dsh-score](https://perrylink.github.io/dsh-attention-index/) registry and shows the same evidence everyone sees. It does **not** compute a per-user score.
+
+> **Status: reference consumer / maintenance mode.** No new scoring or ranking
+> features. The public evidence layer is `dsh-score-index` (currently
+> `dsh-attention-index`; repo rename pending).
 
 [![npm](https://img.shields.io/npm/v/dsh-community-bench)](https://www.npmjs.com/package/dsh-community-bench)
 [![npm downloads](https://img.shields.io/npm/dm/dsh-community-bench)](https://www.npmjs.com/package/dsh-community-bench)
@@ -11,12 +15,13 @@
 
 ## What it shows
 
-- Top 3 per capability surface (memory, vision, Office, browser, orchestration, safety, MCP, remote, dev, voice, research, UI, desktop).
-- Public score / 100 (measured) or / 60 (prior-only), with the tier, prior, bench, coverage and confidence.
+- GOE (`E0`–`E5`) per surface. At v0.2 every row is `E0 / unrated`.
+- Positive measured rows only in the quality block; prior-only rows are labelled candidates, never `#1`.
+- Prior, bench, coverage and **evidence completeness** (not a confidence probability).
 - Community attention blocks and contributor reputation.
-- Link to the public leaderboard and a refresh button.
+- Link to the public evidence site and a refresh button.
 
-The tab is a **consumer**: measurement and allocation live in the public registry, not in this plugin. If the live registry is unreachable, it falls back to the bundled snapshot and labels it.
+The tab is a **consumer**: evidence and grading live in the public registry, not in this plugin. If the live registry is unreachable, it falls back to the bundled snapshot and labels it.
 
 ## Install