docs(p022): re-record the #260 sweep from a CI run of its workflow

[PhysShell]

Что и зачем

Запись свипа #260 теперь снята самим воркфлоу, а не локально: байты артефакта
прогона shadow sweep (#260) заменяют прежнюю локальную запись целиком.
Первый коммит правит воркфлоу так, чтобы файл фактов каждой ноги назывался по id
документа и записывался голым именем — прежняя CI-запись несла путь раннера
десять раз, и поле source было константой, а не провенансом. Второй коммит
кладёт запись того прогона, перегенерированный фрагмент и минимальные правки
ноты. Поведение продакшена не меняется нигде.

Тип изменения

• docs — документация
• refactor / chore / test / ci — без изменения поведения

Как проверено

• python tests/shadow_sweep.py — OK, запись именует коммит прогона
• python scripts/render_checkpoint_status.py --check
• python tests/run_tests.py
• ruff check . и mypy
• Артефакт скачан аутентифицированным GET, сверен по sha256 и распакован как
  байты; sha256 файла в дереве равен sha256 скачанного члена архива

Связанные issue

Refs #250, #260

Чеклист

• изменение покрыто тестом/селфтестом — запись проходит через тот же
  интерпретатор и тот же гейт, что и прежняя
• README/docs обновлены при необходимости
• коммиты в conventional-commit стиле

---

Checkpoint packet (#250)

Scope. Re-record #260's sweep from the workflow's own execution, and make
that record portable: each leg's facts file is named by its document id and
recorded by that bare name, so the assembled record names no runner path and
source distinguishes the documents instead of repeating one constant.

Explicit non-goals. No change to the sweep definition, the interpreter
(tests/shadow_sweep.py), the compare driver, the fixtures or BOUNDARY_POLICY.
No production behaviour changes anywhere. The dirty bit, the
definition_sha256 check in the interpreter, .gitattributes and the
derived-status counts stay separate tails.

Python source of truth. Unchanged. Python remains the public engine; the
adapter stays dev-only.

Frozen fixture. None moved. Nothing under tests/fixtures/ is touched.

Fixture regeneration command. The sweep workflow itself —
gh workflow run "shadow sweep (#260)" — then the record is its
shadow-sweep-run-summary artifact, taken as bytes.

Steady-state test command.

python tests/run_tests.py
ruff check . && mypy
python tests/shadow_sweep.py
python scripts/render_checkpoint_status.py --check

Production dependency changes. None.

Behavior changes. None in production. In CI: each sweep leg writes
<runner temp>/<document id>.facts.json instead of <runner temp>/facts.json
and records that bare name; the driver reads the same bytes, the same once.

Acceptance changes. None. The same definition, the same interpreter, the
same gate; only the recorded run is newer.

Local commands. In the note's §7.

GitHub Actions links. The recorded run is named in the note's §3 and in the
record's workflow_run_url.

Known deferred cases.

• The interpreter does not yet check a record's definition_sha256 against the
  definition's bytes, the way the renderer already does for campaigns.
• .gitattributes for docs/evidence, tests/fixtures and docs/generated:
  three measured episodes now show an operator with core.autocrlf producing
  different bytes for one tree.
• The dirty bit and the derived-status counts are untouched here.
• .slnx: MaterialDesignInXamlToolkit still has no classic solution at its pin,
  so it is covered by its directory walk alone.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01WPozrRezSfnH9RvQGfn1Nn

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: f212229d-f380-4a66-829c-813ddd9b6eee

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}