dsh-task-time-cost 0.2.8-beta.2 → 0.2.9-beta.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (19) hide show
  1. package/CHANGELOG.md +9 -0
  2. package/COMPATIBILITY.md +2 -2
  3. package/README.en.md +7 -6
  4. package/README.md +7 -6
  5. package/docs/public/guide.en.md +4 -2
  6. package/docs/public/guide.zh.md +4 -2
  7. package/lib/client.js +23 -20
  8. package/lib/index.js +35 -11
  9. package/lib/index.js.map +2 -2
  10. package/package.json +1 -1
  11. package/src/client/details.tsx +1 -1
  12. package/src/client/format.ts +19 -0
  13. package/src/client/index.tsx +21 -37
  14. package/src/client/locale.ts +6 -0
  15. package/src/client/styles.ts +4 -1
  16. package/src/host/engine.ts +17 -4
  17. package/src/host/index.ts +2 -2
  18. package/src/shared/wire.ts +1 -1
  19. package/src/shared/workload.ts +19 -5
package/CHANGELOG.md CHANGED
@@ -1,5 +1,14 @@
1
1
  # Changelog / 发行说明
2
2
 
3
+ ## 0.2.9-beta.1 — narrower evidence-based budgets and simpler details
4
+
5
+ - With enough eligible completed tasks of the same kind and route, workload ranges use the observed central 20–80% distribution instead of preserving the wide template ratio. Future, reset, ineligible and other-route tasks are excluded; sparse data retains reference priors.
6
+ - Broad early ETA references read “Refining estimate” in the compact entry; the dialog retains the full calculated range and explains the uncertainty. No clipping or fabricated precision.
7
+ - The first popup shows observed Tokens and estimated remaining output Tokens separately. Already generated output is subtracted once from the remaining budget; incomplete full totals stay unknown.
8
+ - Move estimate details and settings/calibration under the header’s accessible More button. Closing and reopening returns to the simple summary; management still opens only on demand.
9
+ - Clarify executing-AI plans: count model response rounds rather than individual tool calls, and use parallel tool wall time. No extra estimator model requests.
10
+ - Algorithm p0-0.2.9; original usage, historical price resolutions and initial forecasts remain preserved. Original P0 and whole-task accuracy requirements remain open; this is a preview.
11
+
3
12
  ## 0.2.8-beta.2 — npm preview distribution
4
13
 
5
14
  - Enables public npm publication with the beta tag and adds direct DSH installation instructions.
package/COMPATIBILITY.md CHANGED
@@ -2,11 +2,11 @@
2
2
 
3
3
  | Area | Target or boundary |
4
4
  | --- | --- |
5
- | Candidate | 0.2.8-beta.2 public preview; original P0 incomplete |
5
+ | Candidate | 0.2.9-beta.1 public preview; original P0 incomplete |
6
6
  | Host | Windows DSH Desktop 0.2.2-rc.2 |
7
7
  | Runtime | Node 24+, Cordis 4.1.0, Typert 0.2.2-rc.2, aligned Zod 4.6.6 |
8
8
  | Client | DSH module factory, one external Host React, public Slot contracts and theme tokens |
9
- | Algorithm | p0-0.2.4; current-turn records plus explicit optional session aggregates |
9
+ | Algorithm | p0-0.2.9; comparable workload distributions, current-turn records and optional session aggregates |
10
10
  | Other environments | Other DSH releases, macOS/Linux, ACP and mobile unverified |
11
11
 
12
12
  ## Lifecycle / 生命周期
package/README.en.md CHANGED
@@ -2,7 +2,7 @@
2
2
 
3
3
  **A reference for the wait.** See remaining time, actual duration and cost references beside native Token statistics in DeepSeek Harness, calibrated to local routes without a separate estimator-model call.
4
4
 
5
- [中文](README.md) · [Download preview](https://github.com/leicheng42/dsh-task-time-cost/releases/tag/v0.2.8-beta.2) · [User guide](docs/public/guide.en.md) · [Report an issue](https://github.com/leicheng42/dsh-task-time-cost/issues/new/choose)
5
+ [中文](README.md) · [Download preview](https://github.com/leicheng42/dsh-task-time-cost/releases/tag/v0.2.9-beta.1) · [User guide](docs/public/guide.en.md) · [Report an issue](https://github.com/leicheng42/dsh-task-time-cost/issues/new/choose)
6
6
 
7
7
  [![Preview](https://img.shields.io/github/v/release/leicheng42/dsh-task-time-cost?include_prereleases&label=preview)](https://github.com/leicheng42/dsh-task-time-cost/releases) [![CI](https://github.com/leicheng42/dsh-task-time-cost/actions/workflows/ci.yml/badge.svg)](https://github.com/leicheng42/dsh-task-time-cost/actions/workflows/ci.yml) [![MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
8
8
 
@@ -10,7 +10,8 @@
10
10
 
11
11
  ## What it does
12
12
 
13
- - **Estimate the wait:** compact ETA beside native usage; open it for elapsed time and estimated total.
13
+ - **Estimate the wait:** sufficient comparable local tasks narrow workload ranges using their observed distribution; broad references read “Refining estimate,” with the full range in the popup.
14
+ - **See forecast Tokens:** observed Tokens, estimated remaining output and available total forecasts are separate. Less-used details and settings live under the header’s More (⋯) button.
14
15
  - **Explain cost coverage:** known-model public list-price references with explicit overrides, without treating missing prices as free.
15
16
  - **Learn locally:** route latency/speed, sources, samples and dates, plus JSON/CSV baselines and local statistics export.
16
17
  - **Keep useful references:** retain the last valid time reference with its date when current planning is insufficient; bounded plans from the executing AI help estimate remaining work.
@@ -27,21 +28,21 @@ Alternatively, install the pinned GitHub Release archive:
27
28
 
28
29
  1. In DSH **Plugin Manager → Add plugin**, paste the archive URL below, or download it and select the local archive.
29
30
  2. Enable **Task time and cost estimator** and restart when the Host asks.
30
- 3. Open the **clock beside Token statistics**. **Settings and calibration** shows route samples.
31
+ 3. Open the **clock beside Token statistics** for forecast Tokens. **More (⋯) → Settings and calibration** shows route samples.
31
32
 
32
33
  ```text
33
- https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.8-beta.2/dsh-task-time-cost-0.2.8-beta.2.tgz
34
+ https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.9-beta.1/dsh-task-time-cost-0.2.9-beta.1.tgz
34
35
  ```
35
36
 
36
37
  CLI (replace desktop with your profile):
37
38
 
38
39
  ```sh
39
- dsh plugin --profile desktop add https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.8-beta.2/dsh-task-time-cost-0.2.8-beta.2.tgz
40
+ dsh plugin --profile desktop add https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.9-beta.1/dsh-task-time-cost-0.2.9-beta.1.tgz
40
41
  ```
41
42
 
42
43
  ## Preview scope
43
44
 
44
- **0.2.8-beta.2: Windows / DSH Desktop 0.2.2-rc.2 / Node24+.** Other environments remain unverified. [Compatibility](COMPATIBILITY.md) and [acceptance](docs/public/acceptance.md) retain open gates.
45
+ **0.2.9-beta.1: Windows / DSH Desktop 0.2.2-rc.2 / Node24+.** Other environments remain unverified. [Compatibility](COMPATIBILITY.md) and [acceptance](docs/public/acceptance.md) retain open gates.
45
46
 
46
47
  Time ranges are references, not promises or confidence intervals. Startup values are engineering priors. Full online-task accuracy, controlled provider comparisons and observer overhead remain unmeasured. Exact native running-line placement has no verified public seat; the entry is beside Tokens.
47
48
 
package/README.md CHANGED
@@ -2,7 +2,7 @@
2
2
 
3
3
  **让等待有个参考。** 在 DeepSeek Harness 的 Token 统计旁，查看剩余时间、实际用时和费用参考；预估随本地路由表现持续校准，无需另起估价模型。
4
4
 
5
- [English](README.en.md) · [下载预览版](https://github.com/leicheng42/dsh-task-time-cost/releases/tag/v0.2.8-beta.2) · [使用指南](docs/public/guide.zh.md) · [反馈问题](https://github.com/leicheng42/dsh-task-time-cost/issues/new/choose)
5
+ [English](README.en.md) · [下载预览版](https://github.com/leicheng42/dsh-task-time-cost/releases/tag/v0.2.9-beta.1) · [使用指南](docs/public/guide.zh.md) · [反馈问题](https://github.com/leicheng42/dsh-task-time-cost/issues/new/choose)
6
6
 
7
7
  [![Preview](https://img.shields.io/github/v/release/leicheng42/dsh-task-time-cost?include_prereleases&label=preview)](https://github.com/leicheng42/dsh-task-time-cost/releases) [![CI](https://github.com/leicheng42/dsh-task-time-cost/actions/workflows/ci.yml/badge.svg)](https://github.com/leicheng42/dsh-task-time-cost/actions/workflows/ci.yml) [![MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
8
8
 
@@ -10,7 +10,8 @@
10
10
 
11
11
  ## 它能帮你做什么
12
12
 
13
- - **估计还要多久**：紧挨原生 Token 用量的简洁 ETA，点开看实际用时和预计总耗时。
13
+ - **估计还要多久**：本地可比样本足够时，用实际工作量分布收窄预算；跨度过大时显示「预估待收敛」，弹窗保留完整范围。
14
+ - **直接看到预测 Token**：点开就能分别看到已观测 Token、预计剩余输出 Token，以及可确认的预计总 Token；低频详情和设置收进右上角「⋯」。
14
15
  - **看懂费用覆盖**：已知模型自动匹配公共目录参考价，自定义费率优先；缺价和部分已知金额明确区分。
15
16
  - **留下本地经验**：按路由显示延迟、速度、来源、样本数和日期，支持 JSON/CSV 基线与本地统计导出。
16
17
  - **长任务保留参考**：当前计划不足时可保留最后有效时间参考并标注更新时间；执行 AI 的有界计划用于换算剩余工作。
@@ -27,21 +28,21 @@ dsh plugin --profile desktop add dsh-task-time-cost@beta
27
28
 
28
29
  1. 打开 DSH 的**插件管理 → 添加插件**，粘贴下方归档地址；也可先下载归档，再选择本地文件安装。
29
30
  2. 启用**任务耗时与费用预估**，按宿主提示重启。
30
- 3. 在会话底部 **Token 统计旁点击时钟**；从**设置与校准**查看本地路由数据。
31
+ 3. 在会话底部 **Token 统计旁点击时钟**；预测 Token 直接显示，右上角 **⋯ → 设置与校准**查看本地路由数据。
31
32
 
32
33
  ```text
33
- https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.8-beta.2/dsh-task-time-cost-0.2.8-beta.2.tgz
34
+ https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.9-beta.1/dsh-task-time-cost-0.2.9-beta.1.tgz
34
35
  ```
35
36
 
36
37
  CLI（将 desktop 换成你的实际 profile）：
37
38
 
38
39
  ```sh
39
- dsh plugin --profile desktop add https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.8-beta.2/dsh-task-time-cost-0.2.8-beta.2.tgz
40
+ dsh plugin --profile desktop add https://github.com/leicheng42/dsh-task-time-cost/releases/download/v0.2.9-beta.1/dsh-task-time-cost-0.2.9-beta.1.tgz
40
41
  ```
41
42
 
42
43
  ## 预览版适用范围
43
44
 
44
- **0.2.8-beta.2：Windows / DSH Desktop 0.2.2-rc.2 / Node 24+。** 当前支持范围内提供试用；其他平台和宿主版本未验证。[兼容记录](COMPATIBILITY.md)与[验收标准](docs/public/acceptance.md)保留未完成项。
45
+ **0.2.9-beta.1：Windows / DSH Desktop 0.2.2-rc.2 / Node 24+。** 当前支持范围内提供试用；其他平台和宿主版本未验证。[兼容记录](COMPATIBILITY.md)与[验收标准](docs/public/acceptance.md)保留未完成项。
45
46
 
46
47
  时间范围是参考，不是完成承诺或统计置信区间。默认性能基线是工程起点；完整在线整任务准确率、供应商对比和观察开销尚未测量。精确“深度求索中，用时…”旁的位置缺公开扩展槽，目前入口位于 Token 旁。
47
48
 
package/docs/public/guide.en.md CHANGED
@@ -4,9 +4,11 @@
4
4
 
5
5
  ## Entry and estimates
6
6
 
7
- Open the clock/estimate beside native Token statistics. Details expand on demand. Completed actions show elapsed time without duplicate Tokens. Active time, manual waits, remaining reference and estimated total are distinct; retained historical references carry a timestamp.
7
+ Open the clock/estimate beside native Token statistics. The simple popup shows elapsed time, estimated total time, observed Tokens and estimated remaining output Tokens separately. A complete total Token forecast appears when known. The header’s More (⋯) button contains details and settings/calibration; reopening starts with the summary. Completed actions show elapsed time without duplicate Tokens. Retained time references carry a timestamp.
8
8
 
9
- Complete normal tasks to collect local calibration. Profiles show route, source, date, samples and available metrics. Defaults of10s first text,50 visible tok/s,30s total call and0.5/2.5 reference factors are engineering priors, not provider measurements.
9
+ Very broad early ranges show “Refining estimate” in the entry while the dialog keeps the full computed range. At least 5 eligible completed tasks of the same kind and route can supply their central 20–80% workload distribution. Sparse samples retain engineering references. Executing-AI plans count model response rounds rather than tools or files; parallel tool time follows the critical path.
10
+
11
+ Complete normal tasks to collect local calibration. Profiles show route, source, date, samples and available metrics. Defaults of10s first text,80 visible tok/s,30s total call and0.5/2.5 reference factors are engineering priors, not provider measurements.
10
12
 
11
13
  Provider, endpoint identity, model, mode, reasoning settings and tier distinguish routes. Unsupported endpoint identity remains unknown. Only configure a stable identity free of secrets, never an API key or credential-bearing URL.
12
14
 
package/docs/public/guide.zh.md CHANGED
@@ -4,11 +4,13 @@
4
4
 
5
5
  ## 入口与详情
6
6
 
7
- 点击 Token 统计旁的时钟/预估打开浮窗。预估详情按需展开；已结束回合的操作行显示「用时」，不重复 Token。工作时间、等待用户、剩余参考和预计总时间分别理解。历史保留参考附带更新时间。
7
+ 点击 Token 统计旁的时钟/预估打开简洁浮窗，直接查看已用时间、预计总时间、已观测 Token 与预计剩余输出 Token。完整总 Token 可确认时单独显示；预测不会被已用量覆盖。右上角「⋯」打开低频的预估详情和设置与校准，关闭重开恢复摘要。已结束回合的操作行只显示「用时」。历史保留参考附带更新时间。
8
+
9
+ 当时间区间仍过宽时，入口显示「预估待收敛」，点开仍可查看完整范围。至少 5 个同类、同路由的合格已完成任务可提供工作量的实际 20%–80% 分布；样本不足时保留工程参考。执行 AI 的计划按模型响应轮数估计，不能把每个工具或文件都计为一次模型请求；并行工具按关键路径耗时计算。
8
10
 
9
11
  ## 首次校准
10
12
 
11
- 先正常完成若干回合。设置中的路由画像按最近真实观测排序，显示来源、日期、样本数和可测指标。默认首正文10秒、可见正文50 tok/s、总调用30秒及0.5/2.5参考倍率均为工程起点；无合格样本时仍属参考，不代表供应商实测。
13
+ 先正常完成若干回合。设置中的路由画像按最近真实观测排序，显示来源、日期、样本数和可测指标。默认首正文10秒、可见正文80 tok/s、总调用30秒及0.5/2.5参考倍率均为工程起点；无合格样本时仍属参考，不代表供应商实测。
12
14
 
13
15
  同名模型在不同 provider、端点、模式、推理设置和服务档位下不视为同一路由。SDK不能统一提供的端点身份保持 unknown；需要覆盖时只填不含秘密的稳定身份，不填 API Key 或带凭证的 URL。
14
16