Skip to content

emrg: say what each store field is for, and that a shape change moves all of them - #2917

Merged
argszero merged 1 commit into
masterfrom
feature/store-field-duties
Nov 16, 2026
Merged

argszero merged 1 commit into
masterfrom
feature/store-field-duties

Conversation

@argszero

Copy link
Copy Markdown
Owner

Closes #2916

emrg/server/ios_app_prompt.md §5.2 stated that store copy has to be shown by the build it describes, and §5.3 stated which fields the platform freezes. Neither said what each field a submission writes is for, so a round that found sentences false went first to the review notes — the one surface a reviewer reads, while the store page is the one everybody reads.

What the template now says

Two paragraphs in §5.2, beside the invariant they complete:

  • a job per field — the description says what the thing is for and how it answers the reader's problem, in that order, and parameters are not a selling point; the promotional text answers only "why install it now"; the subtitle compresses that into the one line a search result shows; the review notes are facts a reviewer would otherwise have to guess, not copy; the privacy and support pages make claims and are checked like any other page.
  • the shape-change rule — a list where there was one item, a page count, a number of items, a duration, who is allowed to do what: each makes some sentence false in the same moment, so all the fields describing it are read back against the build in that round, and the notes are corrected last rather than only.

Stated as the consequence of the invariant already there, not as a second copy of it. The platform mechanics (409 STATE_ERROR, the localization PATCH-only rule, reviewSubmissions) stay in §5.3, and a guard leg fails if they reappear in §5.2.

Guard

tests/test_ios_app_store_copy_fields.py: the section carries a job per field and the shape-change rule, verbatim; the rule is stated once; §5.2 keeps the invariant that gives the shape-change paragraph its premise; §5.3's mechanics are not restated; and a control that a section carrying none of the sentences reports every one missing.

Mutation arm — removing both paragraphs reddens it (2 failed, 3 passed), and the blob was restored to its committed sha.

Verified

  • uv run --no-sync pytest tests/ -q — 6481 passed, 36 skipped.

CI is pending. No installation's own copy, app or account is named in this change.

@argszero argszero left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ LGTM — cycle cyc20261116-155958

Read head f02080b1 against master 5c3d128c (merge base is master's tip) and CI run 43826901204, both legs green (test 3m1s, test-windows 6m9s).

The gap this closes is a real one in the template as it stood: the copy section said every claim had to be backed by the build, but never said which of the five fields carries which part, so the shortest field to correct — the review notes — was also the easiest to correct alone while the description and the subtitle kept a sentence the build no longer showed. The two added paragraphs answer that directly: one gives each field a single job and says what each is not, the other makes a change of shape move every field that describes it and puts the notes last rather than only.

Both are written in the template's own voice and structure — declarative, no dates, no note about what used to be true, and they sit in §5.2, the section that already owns store copy, rather than at the end of the file.

The guard is the right shape for a prose rule: it pins the two duties rather than the wording, so a later rewording of the paragraph is allowed while deleting either duty reddens it.

@argszero argszero left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ LGTM — cycle cyc20261116-161641

The head f02080b1 no longer contains master (#2913 landed after it), so this verdict rests on a fresh measurement of the landing tree rather than on the PR's own CI:

  • check-merge-plan-suite.py 2917 on base 3b9a8fc8: 6447 passed, 38 skipped in 306.46 s, final tree af7197bd330a.
  • check-merge-landing-diff.py 2917: one path reads backwards — emrg/server/ios_app_prompt.md, where the base's later hunk shows as a deletion this PR does not make; the landing change on that file is the two added paragraphs. The other 21 paths are #2913's own removals.
  • check-merge-pairs.py 2915 2917: both ordered pairs merge clean into a healthy tree.

The change closes a gap the template genuinely had: the copy section required every claim to be shown by the build, but never said which of the five store fields carries which part, so the shortest field to correct — the review notes — could be corrected alone while the description and subtitle kept a sentence the build no longer supported. The two paragraphs give each field one job and its own exclusion, and make a change of shape move every field that describes it, with the notes corrected last rather than only. They sit in §5.2, the section that already owns store copy, and carry no dates or migration note.

@argszero argszero left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ LGTM — cycle cyc20261116-164148

Read on the landing diff: 20 of the 22 paths in diff(master, head) are the base's own later changes shown as reversals this PR does not make. What merges is 2 paths — ios_app_prompt.md §5.2 (+4) and the new test_ios_app_store_copy_fields.py (+129).

Landing tree af7197bd330a8148725b87a3dd823edf27285ca2, the tree whose full suite ran 6447 passed / 38 skipped; git merge-tree --write-tree 3b9a8fc8 f02080b1 reproduces it exactly, so that reading describes the tree this merge would produce. This reading expires the moment #2915 lands — its base moves and the resulting tree is a different one, which gets its own measurement before that merge.

On the change: the gap was real and narrow. §5.2 already said every claim must be shown by the build, and nothing said which of the fields carries which part of it, so the shortest field to correct — the review notes — was correctable alone while the description and the subtitle kept a sentence the build no longer showed. The two paragraphs answer that: one gives each field one job and says what each is not, the other makes a change of shape move every field at once and puts the notes last rather than only. The test pins eight load-bearing sentences verbatim, keeps a control that reads a section with none of them as missing, and refuses a §5.2 that restates §5.3's platform mechanics — one fact in one place, which is the drift this repository keeps paying for.

@argszero

Copy link
Copy Markdown
Owner Author

The landing tree this PR now merges is not the one the votes were cast on, and the difference is worth naming rather than absorbing.

#2915 landed as a9647932 after the three votes here were cast, so this head is now behind_by=2 and its landing tree moved from af7197bd to e5ce9fbbdfca13dd205bb139fb169ab5cb06ee0d. Measured on that tree: suite 6453 passed, 38 skipped in 289.57s (check-merge-plan-suite.py, base a9647932, exit 0).

The votes stand: they are bound to the head, and the head did not move. This comment carries the reading; it is not a vote.

@argszero
argszero merged commit b1216aa into master Nov 16, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

emrg: the ios-app template says which store field carries which part of the copy

1 participant