Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions Documentation/Validation.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,11 @@
# 1.5.0 validation · 2026-10-02

- 62 Swift tests pass on Xcode 27.0 / macOS 27.0.1. New coverage: correction anchoring across curly quotes, full-width punctuation, whitespace and case; invented spans dropped without failing the review; DeepSeek streamed usage (including cache hits and reasoning tokens) and Codex turn-completion usage; required `expressions` in the Codex schema; single-judge quick review routing, aggregation and persistence; legacy reports decoding as full reviews; one exam time-up signal per crossing; clearing an untouched submitted draft; word/character revision diffs; parent-then-earlier revision baselines; built-in bank uniqueness and practice index; review-card sync idempotence and the 1/2/4/8/16-day schedule; backup export/restore round trip with images, cards and drafts, and a second restore that adds nothing.
- Two existing expectations changed by design: an invented correction span no longer rejects the whole review, and a successfully graded, unchanged draft starts fresh.
- An isolated QA copy (separate bundle identifier, store and defaults) was seeded through the new backup restore with six synthetic reviews, then each page was captured: review score, comparison card and word diff, annotated essay, quick-review score card, History deltas and 单评 label, Statistics trends, Mistakes trends and mastery, Practice start and a revealed drill card, question bank sheet, immersive countdown, and Settings. Layout issues found there (sidebar badge truncation, struck words joining neighbours) were fixed. The user's database and defaults were not used.
- No paid live grading was run. DeepSeek usage relies on `stream_options.include_usage`; Codex usage parsing is lenient and absent usage is not an error. Model compliance with the new `expressions` field is untested live.
- macOS version 1.5.0 (8), Universal arm64 + x86_64. SwiftData adds the `ReviewCard` model and defaulted `SavedQuestion.year` / `label`; existing stores migrate lightly. Prompt version 1.3.0.

# 1.4.0 validation · 2026-09-13

- 49 Swift tests pass on Xcode 26.6 / macOS 26.6.2. New coverage includes completion-order progress, concurrent independent inputs, immutable submission and rewrite-parent snapshots, editing a new draft during grading, duplicate-submit rejection, cancellation, judge-specific failure, Codex preflight failure, and no partial score persistence.
Expand Down
7 changes: 4 additions & 3 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -21,8 +21,9 @@ WriteBench 把练习收敛为一条清晰的路径:选题、作答、评阅、
| 专注写作 | 认真评阅 | 留下进步 |
| :--- | :--- | :--- |
| 原生全屏作答,自动保存草稿 | 三位评审独立阅读同一份原文 | 同题目录与逐稿修改路径 |
| 考研、六级使用横线答题区,默认关闭实时词数 | 原题要求、语言、组织与语域分别反馈 | 错误归类、趋势与练习统计 |
| 手写稿识别后先校对,再提交 | 本机取中位数,显示评审分歧 | 所有原稿保留在自己的设备上 |
| 考研、六级使用横线答题区,可开启考试倒计时 | 修改直接高亮在原文上,可点击查看 | 与上一稿对比:分差、维度变化、逐词改动 |
| 内置原创题库,随机抽未练过的题 | 本机取中位数;草稿可用快速单评 | 错题与表达的间隔复习、趋势统计 |
| 手写稿识别后先校对,再提交 | 记录每次评阅的用量与用时 | 一键导出 / 恢复全部本地数据 |

### 支持的考试

Expand Down Expand Up @@ -136,7 +137,7 @@ platforms/android/ Android Studio 项目与手机界面
scripts/ 构建、图标生成与显式联调脚本
```

macOS 自动化测试覆盖 49 个案例,Android 有 5 个领域测试与 2 个实际设备服务测试;Windows 通过评分、持久化、字段校验及实际 OCR 自检。GPT-6 Astra/MAX 已通过实际 Swift 子进程完成样例评卷;DeepSeek 已验证官方模型接口连接。后台、流式输出及版本路径已用隔离测试验证;本次未执行完整付费三评,需用户填入有效 Key 后使用。
macOS 自动化测试覆盖 62 个案例,Android 有 5 个领域测试与 2 个实际设备服务测试;Windows 通过评分、持久化、字段校验及实际 OCR 自检。GPT-6 Astra/MAX 已通过实际 Swift 子进程完成样例评卷;DeepSeek 已验证官方模型接口连接。后台、流式输出及版本路径已用隔离测试验证;本次未执行完整付费三评,需用户填入有效 Key 后使用。

<details>
<summary><strong>更多文档</strong></summary>
Expand Down
15 changes: 15 additions & 0 deletions RELEASE-NOTES.md
Original file line number Diff line number Diff line change
@@ -1,3 +1,18 @@
# 1.5.0 · 练习闭环:限时、对比、复习

- **评分更稳**:评审引用原文时出现弯引号、全角标点、多余空格或大小写差异,会自动对齐到原文;个别找不到的修改建议只丢弃该条,不再让整份评阅失败。
- **考试限时**:Settings 可开启倒计时(考研小作文 15 分钟、大作文 30 分钟、雅思 Task 2 40 分钟等),最后 5 分钟变色,到点提示音;可选到点自动交卷,或继续作答并显示超时。准备页新增“清空重来”。
- **原文标注**:逐句修改直接高亮在你的原文上,主要 / 次要问题分色,点击高亮处查看修改和理由。
- **与上一稿对比**:评阅页新增“与上一稿对比”,显示总分与各维度的分差、主要问题数量变化,以及逐词新增 / 删除对比。History 每一稿旁显示相对上一稿的分差。
- **统计看变化**:新增近期变化(近几篇比之前几篇,百分点)、最近 3 篇平均线、四个维度趋势图、评阅用时与用量汇总;可按三评 / 快速单评筛选。Mistakes 显示每类错误的近期变化,可标记“已掌握”并隐藏。
- **数据备份**:Settings 可导出全部作文、评阅、草稿、题库、目录信息、复习卡和图片为一个 JSON;恢复时只添加本机没有的记录,不覆盖现有内容。不含 API Key 与 Codex 登录信息。
- **用量与快速单评**:每次评阅记录各评审的 token 用量与用时;填入 DeepSeek 价格后显示估算花费。答题页新增“快速单评”(⇧⌘↩),只请一位评审,适合草稿,可在 Settings 选择由哪位评审负责。
- **题库**:内置题目扩充为每个题型 4–6 道原创题,标注已练次数与最近得分;支持随机抽一道未练过的题,我的题库可填写年份与标签并筛选。
- **间隔复习**:新增 Practice 页。逐句修改与评审推荐的表达自动生成复习卡,按 1 / 2 / 4 / 8 / 16 天间隔安排;错题看原句写改法,表达看中文写英文。表达库支持搜索、手动添加和移除。侧栏显示待复习数量。
- 成功评阅后,未再修改的已交草稿自动清空并重置计时,原文保留在 History。评审提示词升级到 1.3.0,新增推荐表达字段;旧记录照常读取。

macOS 15+。数据库新增复习卡与题库年份 / 标签字段,旧数据自动迁移。

# 1.4.0 · 后台评阅与逐稿修改路径

- 交卷后立即回到准备页,三位评审在后台独立工作。顶部显示真实的完成进度(0/3 → 3/3)、等待时间、取消和结果入口,完成时不强制弹出评阅。
Expand Down
Loading
Loading