[ARSH871-bot]

The Phase 3 backlog document now lists the seven specialist picks agreed on 13 October (US-58 to US-64, #391 to #397), and records #220 as Shubham's. Until this merges, docs/design/phase-3-backlog.md misstates the milestone: seven of its stories are missing and one has the wrong owner.

What changed

All of it goes through the generator. The document is regenerated, not edited, because tests/test_phase3_backlog_doc.py requires it to match tools/make_phase3_backlog.py byte for byte.

File	Change
tools/make_phase3_backlog.py	The seven picks join EXISTING, the list the script folds in rather than creates. create skips them, so no run can open them a second time. US-53's owner becomes Shubham. A Specialist epic, a row in "Where these came from", and a line in the order-of-work block, which the script checks against the stories.
docs/design/phase-3-backlog.issues.json	US-58 to US-64 mapped to #391 to #397
docs/design/phase-3-backlog.md	Regenerated with python -m tools.make_phase3_backlog doc, which makes no GitHub calls

The distribution line is computed from the lists, not typed:

before   38 stories. Arsh has 28. Ankeet has 4, Shubham has 3, Samika has 3.
after    45 stories. Arsh has 29. Ankeet has 7, Shubham has 4, Samika has 5.

Sizes show as TBC. The issues leave sizing to each owner at sprint planning on 19 October, and the document should not guess one.

Evidence

pytest tests/test_phase3_backlog_doc.py tests/test_tools_invocation.py    20 passed, 10 skipped
pytest tests/                                                            1869 passed, 13 skipped   (same as main)

The guards were checked against a mutation, not just run green. With US-64 removed from the numbers file and one line of the document hand-edited:

FAILED test_document_matches_the_generator
FAILED test_every_story_has_an_issue_number
2 failed, 3 passed

Both restored: 5 passed. The committed document has 0 CR bytes and is byte-identical to document(), so the check holds on CI's Linux runners as well as on Windows.

On #249

This is my second open PR beside #386, which #249 says I should not do. It is opened under that rule's exception for fixing something already wrong: the document misstates the milestone until it merges. #387 to #390 are next in my queue.

Reviewer: @shubhamkataria2005, because the one change here that is not mechanical is US-53 (#220) being recorded as yours.

[shubhamkataria2005]

Approving. Verified by regeneration, not by reading the diff:

• Merged onto current main, ran python -m tools.make_phase3_backlog doc (no GitHub calls), and the output is byte-identical to the committed document. tests/test_phase3_backlog_doc.py: 5 passed.
• The ownership it records matches GitHub. #220 is assigned to me, and #391–#397 are assigned exactly as the specialist table says: Arsh #391–392, Ankeet #393–395, Samika #396–397.
• The distribution line is computed, not typed, which is the right call for the one sentence people quote.

Taking #220 (US-53) — I will comment there once I have sized it, ahead of planning on 19 October.