fix: keep report evidence consistent across collectors

[oldwinter]

Summary

• identify Python/FastAPI projects without inferring Rails from a generic app/ directory
• project bounded, documented Just recipes and exclude history-only paths from current-file recommendations
• bind Session facts and lead analysis to one privacy-filtered population with redacted consistency checks
• expose dimension tracks as validated progressbars and keep bounded Chinese phrases together in self-contained HTML reports
• render canonical summary-facts Episode coverage in visible Evidence cards while retaining the legacy fallback

Why

• Issue/Story: None. This is focused maintenance from a validated Better Harness review of a real FastAPI/Just repository, followed by spec-first regressions and browser QA.
• Outcome: project and Session evidence no longer contradicts itself, report actions point to current executable surfaces, and HTML reports are more accessible and readable in Chinese.

Traceability and Scope

• Specs: docs/specs/2026-07-30-core-change-watch-profile-and-current-paths.md; docs/specs/2026-07-30-session-population-binding.md; docs/specs/2026-07-30-html-dimension-progressbar-semantics.md; docs/specs/2026-07-30-html-cjk-line-breaking.md; docs/specs/2026-07-30-html-summaryfacts-episode-coverage.md
• Acceptance criteria: framework-specific classification; bounded unverified Just entrypoints; current-path filtering; shared privacy-scoped Session population and fail-closed bindings; complete score-bound progressbar semantics; bounded zh-CN phrase wrapping with escaped fallback; canonical Episode coverage precedence with legacy compatibility.
• Owners changed: Core Change Watch projection, Session evidence-bundle orchestration/contracts, self-contained HTML renderer/validator, tests, docs, and changelog.
• Non-goals: no score manipulation, synthetic Episodes, raw Session persistence, installed-plugin edits, dependency changes, release, or deployment.

Change Type

• Feature
• Bug fix
• Tests only
• Documentation/community
• Refactor with no intended behavior change
• Dependency, packaging, or infrastructure

Test and Review Evidence

Check	Result
Focused Core Change Watch, Session population, Evidence Bundle, task-loop, report-run, CLI, and WorkBuddy integration suites	161/161 pass on tree 7b981dc2b4a98214c3c3cdbf81acb682c050c290
npm run check	1006/1006 tests pass
npm package verification	Pass, 343 entries
runtime ZIP verification	Pass, 366 entries
git diff --check, conflict-marker scan, unresolved-index scan, and exact-tree audit	Pass
Independent code review and completion gate	PASS on the same tree SHA, no findings

Manual and visual evidence:

• Chrome 151 full-page QA at 375, 768, and 1280 pixels.
• Target phrases 口径, 入口, 结论, and 命令 had zero split occurrences at every viewport and no document horizontal overflow.
• The final report visibly matched 5/5 Sessions and 14/12 total/edited Episodes, retained zero findings, and scored 81/100 (exact mean 80.6).
• Five progressbars exposed exact values 93/84/94/82/50; axe reported zero violations; console and page errors were empty; dialog focus restored after close.

Risk and Recovery

• Compatibility: filesystem projections use portable Node path handling; segmentation is locale-gated and falls back to escaped readable text when Intl.Segmenter is unavailable.
• Package/schema impact: shipped source, docs, and tests only; no dependencies, installed plugin cache, generated report artifact, or score data changed.
• Rollback: the repair commits are independently revertible; upstream integration used normal merge commits without rewriting published history.
• Residual risk: the current merge tree was validated on Linux with cross-platform fixtures. An earlier head passed the repository's Ubuntu Node 22/24, macOS, and Windows CI matrix. CI for the current head is waiting for the repository-required fork workflow approval.

AI Involvement

• Level: Assisted
• Review: diffs and rendered reports were inspected, focused/full/package checks ran, real target projections were exercised, browser interactions were verified, and independent code/completion reviews passed on the exact staged tree.

Checklist

• Followed AGENTS.md, CONTRIBUTING.md, and canonical-owner guidance.
• Kept the change focused and excluded unrelated local/generated state.
• Kept tests and documentation aligned with delivered behavior.
• Checked Markdown links and cross-platform behavior.
• Ran package/runtime verification and updated CHANGELOG.md.
• Confirmed the right to contribute under the MIT License.

[Copilot]

Pull request overview

该 PR 旨在修复 Better Harness 报告在不同采集器/渲染器之间的“证据口径不一致”问题：通过冻结并绑定同一份隐私过滤后的 Session population，确保 Session facts 与 lead analysis 使用同一人口快照；同时改进 Core Change Watch 的项目画像与“当前可操作路径”投影；并强化自包含 HTML 报告的可访问性语义与中文可读性。

Changes:

• 新增 Session population 冻结/选择/准入绑定与一致性校验，并在 evidence-bundle 中将 population 作为共享输入传递给 facts 与 lead，冲突时 fail-closed（稳定红码）。
• Core Change Watch：框架识别改为“必须有框架特异证据”，新增根 Justfile 的静态 recipes 投影（unverified argv），并将 recommendedReads / followUpActions 过滤为当前真实存在的路径（保留历史证据但不推荐已删除文件）。
• HTML renderer：维度分数轨道补齐 progressbar 语义并在 validator 中绑定数量与分数；zh-CN 可见文本使用 Intl.Segmenter 进行有界短词 keep-together 渲染与回退。

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
test/task-loop-source.test.mjs	覆盖传入冻结 population 时的复用与 lead 绑定输出断言。
test/session-population.test.mjs	新增 population/selection/admission 绑定与一致性校验的单测。
test/harness-report-render-cli.test.mjs	新增 progressbar 语义与 zh-CN 分词 keep-together 的渲染/验证回归。
test/core-change-watch.test.mjs	新增 FastAPI vs Rails 判别、Just recipes、历史路径过滤的回归覆盖。
test/better-harness-evidence-bundle.test.mjs	覆盖 evidence-bundle 共享 population、冲突 fail-closed、隐私红码与计数一致性。
scripts/session-analysis/session-population.mjs	新增冻结 population 与绑定/校验工具（population/selection/admission）。
scripts/harness-analysis/task-loop-source.mjs	支持从冻结 population 构建 inventory，并输出 lead 侧 sessionBinding。
scripts/harness-analysis/report-run.mjs	将 sessionPopulation 透传给 source 生成，并在 evidence 输出中携带 sessionBinding。
scripts/harness-analysis/renderers/html.mjs	progressbar 语义与 validator 绑定；zh-CN 可见文本分词渲染与 CSS 约束。
scripts/harness-analysis/evidence-bundle/session-evidence.mjs	新增 collectSessionPopulation，并将 population 注入 facts collector 以对齐人口。
scripts/harness-analysis/evidence-bundle/index.mjs	evidence-bundle 先采集共享 population，再并行采集/分析并输出一致性诊断。
scripts/core-change-watch/project-profile.mjs	框架识别收紧；新增根 Justfile recipes 静态投影为 entry candidates。
scripts/core-change-watch/evidence-pack.mjs	recommendedReads / followUpActions 过滤为当前存在路径（保留历史证据但不推荐已删除文件）。
scripts/core-change-watch/core-candidates.mjs	将 app/api 的通用路径理由语言中立化，并为 FastAPI 添加路径信号。
references/project-harness/core-change-watch.md	更新 Core Change Watch 的公开指导：框架证据、Just recipes、当前路径投影边界。
docs/specs/2026-07-30-core-change-watch-profile-and-current-paths.md	记录 Core Change Watch 证据与当前路径投影的 spec 与验收证据。
docs/specs/2026-07-30-session-population-binding.md	记录 Session population 绑定/隐私/失败闭合策略的 spec 与验收证据。
docs/specs/2026-07-30-html-dimension-progressbar-semantics.md	记录维度 progressbar 语义与验证绑定的 spec 与验收证据。
docs/specs/2026-07-30-html-cjk-line-breaking.md	记录 zh-CN 分词 keep-together 渲染与回退策略的 spec 与验收证据。
CHANGELOG.md	记录用户可感知的行为变化（Core Change Watch、Session 绑定、HTML 语义与中文换行）。

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.