fix(stage1): apply corrections both reviewers agreed on, downgrade unsourced status

Implements only what reviewer 1 and reviewer 2 independently and concordantly
required, at the build sources rather than in generated files:

- NSDUH 2022/2023/2024 YUSUITHK: strip the codebook editorial note about the
  2022 section move from question_text. It was never shown to respondents.
  The removed text is kept verbatim in notes.
- NSDUH 2021 YUSUITHK: drop the stray space before the final question mark.
- Solomon Islands items: review_status provisional -> needs_source. No country
  questionnaire, codebook or fact sheet ships under Dataset/; the only citation
  is the generic GSHS data dictionary. Keyed off the missing instrument, not
  hardcoded per country.

Also downgrades two design claims that A-20260920-067 verified as unsupported:
Jamaica and Wallis and Futuna report a census with a 100% school response rate
but carried the same Taylor-linearization plan as the two-stage cluster
samples. design_status becomes census_variance_treatment_unresolved and two
conflict records are added. variance_method itself is untouched - that is a
Stage 4 statistical decision needing its own evidence.

Deliberately not done: the instruction sentence missing from the 2021 stem is
left absent rather than written in, since that would be composing official
question text; and the 84 adjudications remain human work. Only 21 of 84 rows
are concordant, so 63 need adjudication.

84 item_version_id unchanged, response_dictionary byte-identical, both signed
reviewer CSVs unchanged. Gate 1 audit passes all seven machine deliverables and
stays blocked solely on two_actual_human_reviewers.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Jinotech
2026-09-20 22:03:41 +12:00
co-authored by Claude Opus 5
parent bf7bfb7a35
commit 21c7e211d0
10 changed files with 153 additions and 29 deletions
+26 -1
View File
@@ -1,6 +1,6 @@
# 项目进度档案
版本:1.21
版本:1.22
建立日期:2026-09-20
当前总状态:`in_progress`
当前阶段:阶段 1 `in_progress`;题库、回答字典、919,863 行回答长表及三调查设计字段/方法均已核验;Gate 1 仅剩主要结局的两名实际人工审核者签署
@@ -988,3 +988,28 @@ Gate 0 当前判定:`passed`。四项交付物齐备;目标总体政策、
1. 由签署人确认并将 `review_date` 修正为 ISO `YYYY-MM-DD`,随后重跑 `validate_human_reviews.py`
2. 完成 84 项裁决并据此更新题库审核状态;
3. 重跑 Gate 1 逐项验收,全部满足后方可将阶段 1 标为 `passed`
### A-20260920-071:实施两名审核者一致要求的更正并降级未获来源支持的状态
- 时间:2026-09-20T22:02:53+12:00;阶段 1Gate 1 保持 `in_progress`
- 动作:行动前完整读取工作流与本档案并核对干净 Git 状态。读取两份已签署审核表,仅实施**两名审核者独立且一致**要求的更正,并对经复核确认没有来源支持的状态做降级。全部修改在生成脚本层,未改动任何已签署审核行,未替任何人裁决。
- 输入:`primary_outcome_review_reviewer_1.csv``primary_outcome_review_reviewer_2.csv``item_bank.csv``response_dictionary.csv``survey_design.csv``nsduh_complete_variable_dictionary.csv`、A-20260920-067 已核实的 WHO 实况表证据。
- **审核一致性统计(本次新发现,须记录):** 84 行中仅 **21 行两人决定一致**14 `approve_with_correction` + 7 `needs_source`),**63 行不一致**。不一致分布:`needs_source`/`approve` 27 行、`needs_source`/`approve_with_correction` 18 行、`approve_with_correction`/`approve` 18 行;按调查为 YRBS 45 行、GSHS 18 行。reviewer 1 无一行给出无条件 `approve`reviewer 2 给出 45 行。75% 的不一致率本身即为需要裁决时重点关注的证据。
- 实施的更正(均为两人一致要求,且经本次独立复核):
1. **NSDUH 2022/2023/2024 `YUSUITHK` 删除代码本编辑说明。**`question_text` 以 "Note: Beginning in 2022, questions ... were moved from the youth mental health service utilization section to the youth experiences section." 开头,该句是 DAS 字典的编辑注记,非受访者所见。reviewer 1 要求“删除开头的模块迁移 Note”,reviewer 2 标为 `CORRECTION (wording)`。删除后被删文本**逐字保留**在 `notes` 中,未丢失。
2. **NSDUH 2021 `YUSUITHK` 修正题干末尾多余空格**`kill yourself ?``kill yourself?`),reviewer 2 明确指出该空格属 DAS 渲染artifact。
3. **Solomon Islands 三题 `review_status` 由 `provisional` 降为 `needs_source`。** 两名审核者均独立返回 `needs_source`,理由为 `Dataset/` 下不存在该国问卷、代码本或实况表(递归检索 "solomon"/"SLB" 无命中),其唯一引用为通用 `GSHS_数据字典.xlsx`。降级逻辑写为“无国别问卷则 needs_source”,不针对国别硬编码。
- 实施的状态降级(依据 A-20260920-067 已核实证据,非本次新主张):
4. **牙买加与瓦利斯和富图纳 `design_status` 由 `source_verified_fields` 降为 `census_variance_treatment_unresolved`。** 两地实况表载明使用普查("A census was used…" / "Un recensement a été utilisé…",学校应答率 100%),但其权重/PSU/层与 Taylor 线性化计划此前与两阶段整群样本组件逐字节相同。权重/PSU/层字段予以保留,仅将方差处理标为未决,并新增 2 条 `design_treatment_not_established_for_census` 冲突记录。`variance_method` 本身未改动——那是需要独立证据的阶段 4 统计决定,不在本次范围。
- 未实施及理由(**明确不做**,留待人工):
- **2021 缺失的作答提示句未补写。** 20222024 含 "You can answer 'I'm not sure' or 'I don't want to answer' to any question."2021 无此句,而 2021 同样提供编码 3I'm not sure)与 4I don't want to answer)。两名审核者均指出应补回。但向 `question_text` 写入题目文本属改写官方原文,违反工作流 §5.2“不补写成官方原文”,已作为待人工处理项登记,未由 AI 执行。
- **84 项裁决未执行。** 按 A-20260920-063 防伪边界,裁决属人工决定。
- `LANGVER` 仍未用于语言状态;NSDUH 2021 与 20222024 的题干结构差异仍存。
- 检查:两个脚本 `py_compile` 通过后运行 registry 与 package。计数 84 题 / 396 编码 / 28 设计行 / 9 连接 / **34 冲突(新增 2** / 84 队列。逐项 diff 验证:84 个 `item_version_id` 不变、列结构不变、回答字典 396 行**完全未变**;题库仅 `question_text` 4 行、`text_hash` 4 行、`notes` 3 行、`review_status` 3 行变化。`text_hash` 由 Gate 审计独立重算校验通过,确认哈希与新文本一致(2022/2023/2024 修正后题干相同,哈希合并为同一值,符合预期)。两份已签署审核表 SHA-256 保持 `cbd79847…``5fe95245…` 不变。
- Gate 1 接受审计:七项机器交付物全部 `passed`(冲突 34 行、未关闭 13 项),唯一 `unmet_requirements` 仍为 `two_actual_human_reviewers`。Gate 1 保持 `in_progress`
- 未解决问题:63 行审核不一致待裁决;`review_date` 仍为 `2026/9/20`(非 ISO);2021 提示句缺口;13 项未关闭冲突;`LANGVER` 未使用。
- 下一步:
1. 由两名审核者裁决 63 行不一致决定,并确认 `review_date` 的 ISO 格式修正;
2. 裁决完成后重跑 `validate_human_reviews.py` 与 Gate 1 验收;
3. Gate 1 通过后方可进入阶段 2(可行性、可识别性和连接审计)。