V5.3-r3 human-calibrated rubric / 人工校准评分细则 - #60
Open
yyuan308 wants to merge 1 commit into
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary / 概要
Adds a separately versioned V5.3-r3 Linear Algebra development candidate after course-owner calibration. It does not rewrite the frozen V5.3-r2 draft.
Encodes explicit error-depth anchors, independent final-result points, equivalent symbolic-expression handling, independent Q3 bonus handling, and an explicit Q3 fundamental-method scoring gate.
Updates mirrored grading skills so an incorrect independently allocated final result does not retain final-result points merely because a prior local error caused it. No-double-counting still protects dependent process criteria.
Adds strict validation that a scoring-gate trace uses the declared gate, cap, deduction type, and no extra deductions.
新增独立版本化、经课程负责人校准的 Linear Algebra V5.3-r3 开发候选;不会改写已冻结的 V5.3-r2 草案。
将按错误深度的分值锚点、最终答案独立分、符号表达式等价、独立 Q3 bonus 和 Q3 基础方法错误门槛写入可执行 rubric。
同步 Codex/Claude 通用评分 skill:错误的独立最终答案不因上游局部错误而保留答案分;不重复扣分仍保护依赖性的过程分。
新增严格校验:scoring-gate trace 必须使用已声明的 gate、分数上限、扣分类型,且不得混入其他扣分。
Tests / 测试
Targeted public tests: 31 passed.
Full public core suite completed without a reported failure.
Static rubric validation passed for the new execution-contract rubric.
定向公共测试:31 项通过。
完整公共 core 测试已完成,未报告失败。
新 execution-contract rubric 的静态校验通过。
Privacy and limits / 隐私与限制
No Data files, student work, identities, transcripts, per-student gold, raw outputs, or private locations are staged.
This PR contains no private gold revision, packet, model call, metric, heldout access, or production/teacher-replacement claim.
The r3 plan requires separate explicit authorization before private development-gold revision, packet construction, or any model run.
暂存内容不含 Data 文件、学生作答、身份、转录、逐人 gold、原始输出或私有位置。
本 PR 不包含私有 gold 修订、packet、模型调用、指标、留出集访问,也不作生产可用或教师替代声明。
r3 计划要求:私有开发集 gold 修订、packet 构建和任何模型运行都必须另行获得明确授权。
Next step / 下一步
After review and merge, decide separately whether to authorize a new private development-only gold revision. Only after that can a matched G1 packet and the authorized Codex plus DeepSeek V4 Flash development comparison be prepared; heldout remains sealed.
审查并合并后,请单独决定是否授权新的私有开发集 gold 修订。仅在此之后才可准备同一份 G1 packet 并开展获授权的 Codex 与 DeepSeek V4 Flash 开发集比较;留出集继续封存。