Skip to content

test: the numeric-option ratchet is its own pin; a range check refuses an added row (#1608, slice 1) - #1614

Queued
markramm wants to merge 11 commits into
devfrom
claude/ratchet-not-yet-pin-ae38fa
Queued

markramm wants to merge 11 commits into
devfrom
claude/ratchet-not-yet-pin-ae38fa

Conversation

@markramm

@markramm markramm commented Nov 17, 2026 •

Copy link
Copy Markdown
Collaborator

Part of #1608 (slice 1)

Branch: this session was given claude/ratchet-not-yet-pin-ae38fa, used instead of process/1608-ratchet-range-check.

Acceptance (from #1608)

  • For each converted ratchet: two branches that each remove a different row merge into each other with no conflict in the test file (show it once with git merge-tree in the PR; no test needs to run git).
  • Adding a row without the trailer fails the range check; adding it with the trailer passes; removing rows passes. The check has its own tests, as tests/test_fix_commit_check.py has.
  • A row removed from the source but left in the list still fails the structural test (stale rows are caught as they are today).
  • CONTRIBUTING, "Running the tests", says in two sentences how to remove a row and how a row is added.

Report

Branch: claude/ratchet-not-yet-pin-ae38fa
Pushed SHA: d8cda527147db1417e1805196e8499ad08470bf4. git ls-remote origin claude/ratchet-not-yet-pin-ae38fa printed that SHA, and local git rev-parse HEAD printed the same, checked right after the push and before anything else touched a commit.

Commits (8 of mine, then 2 merges of dev):

  1. 32dac049 kb: the claim (backlog item).
  2. 0b472525 test: scripts/check_ratchets_only_shrink.py with tests/test_check_ratchets_only_shrink.py.
  3. a7ea363f test: NOT_YET is its own pin; the pin and its two tests are deleted; groups separated by a #-only line.
  4. f019e05a test: the new tests' subprocess calls get a timeout= (a repo rule, test_no_test_call_can_wait_forever, caught it).
  5. d6ac1160 ci: one step, "Check ratchet lists only shrink (pull_request only)".
  6. 3ae3f366 docs: CONTRIBUTING, "Running the tests".
  7. f0686ad8 kb: backlog item moved to done/ (this commit alone does not carry status: done, see Unsure).
  8. 71276751 kb: backlog item status: done (a new commit; the status line in done/ really is done).
  9. c0337ad1, d8cda527: merges of origin/dev (see (a) below).

Footprint: git diff origin/dev --stat at d8cda527: the backlog item, scripts/check_ratchets_only_shrink.py (185), tests/test_check_ratchets_only_shrink.py (322), tests/test_cli_numeric_options_declare_a_range.py, .github/workflows/ci.yml (+9, one step), CONTRIBUTING.md (+8). Six files, nothing else, no changelog fragment (the fragment check is advisory and did not refuse).

Done:

  • Red list before the script existed: tests/test_check_ratchets_only_shrink.py failed at collection with ModuleNotFoundError: No module named 'check_ratchets_only_shrink', so all of its tests were red at import (1 error, 0 collected).
  • Green after: 38 passed.
  • Each guard fails a test when removed (run by hand): ignoring the trailer 3 failed, always passing 21 failed, count-only comparison 1 failed (the swap case), accepting an empty reason 2 failed.
  • scripts/test-affected --run on the final tree: 27470 passed, 258 skipped, 7 xfailed, 0 failed. The pre-push hook did not run a second suite.

Acceptance evidence

(c) Merge demonstrations (throwaway local branches off this branch's layout, since deleted; the commands are in the sed scripts below):

$ git merge-tree --write-tree t-sw t-wiki      # (i) the eight `pyrite sw` rows (#1595) with the `pyrite wiki stubs` row (#1599)
4447b39d5c5075391899e0aa1eff7b42642d7cf7
exit=0
$ git merge-tree --write-tree t-g12 t-g34      # (ii) groups 1+2 (search_commands.py, entry_commands.py; comment lines and rows) with groups 3+4 (browse_commands.py, cli/__init__.py)
cb2f68a31edd1e80dac85cce399f659013295570
exit=0

Branches were made with sed -i '/^ ("pyrite sw /d', sed -i '/^ ("pyrite wiki stubs"/d', and a ranged sed deleting each pair of groups' comment line and rows while keeping the # separator lines (6 and 5 deletions). I also ran every adjacent single-group pair (13) and every adjacent pair-of-groups (11) with a script on an in-memory base: all exit 0. Without the separator lines, every adjacent-group pair conflicted (exit 1); ruff format deletes a blank line inside the set literal, so the separator is a line holding only #. The list's comment says what it is for.

Planted added row, then the trailer:

$ python scripts/check_ratchets_only_shrink.py --range claude/ratchet-not-yet-pin-ae38fa HEAD     # planted row, no trailer
ERROR: ratchet list NOT_YET in tests/test_cli_numeric_options_declare_a_range.py gained 1 row(s) between ... and HEAD:
  ('pyrite social planted', 'limit')
  A ratchet list only shrinks. Remove the rows, or if adding them is deliberate, put this line in the final paragraph of a commit message in the range:
    Ratchet-raise: NOT_YET <reason>
exit=1
$ ... same, after a commit whose final paragraph is `Ratchet-raise: NOT_YET demo of the accepted path`
exit=0

(throwaway branch, deleted, never pushed.)

Stale row still fails: with a planted ("pyrite social ghost", "limit") row, pytest tests/test_cli_numeric_options_declare_a_range.py::test_listed_rows_are_not_stale gave 1 failed: NOT_YET row \pyrite social ghost` `limit`: no such numeric option; delete the row. Plant removed with git checkout --`, not committed.

Range check on this PR's own range:

$ python scripts/check_ratchets_only_shrink.py --range origin/dev HEAD
(no output)   exit=0

It also exits 0 against d4fcea8c.

(a) What the merges of dev conflicted on, and how: both merges conflicted only in tests/test_cli_numeric_options_declare_a_range.py, in the same two places. First merge (#1595, aa7deebc): dev had deleted the eight pyrite sw rows and lowered NOT_YET_PINNED_SIZE to 32. Second merge (#1599, 814382d3): dev had deleted the pyrite wiki stubs comment line and row and lowered the pin to 31. Resolution each time: keep this branch's layout, delete that group's comment line and rows, leave the separator lines, keep no pin line.

Proof of the second resolution (set of rows here equals dev's at a5f942bd):

$ python3 - <<'E'   # reads NOT_YET with ast at the working tree and at `git show origin/dev:<file>`
31 31 equal: True only-mine: [] only-dev: []

(b) Live merge-tree check. Right after my first merge, #1599 landed on dev, and the check was NOT clean:

$ git merge-tree --write-tree --name-only HEAD origin/dev      # HEAD c0337ad1, origin/dev a5f942bd
2722c9b6c0b86c6ce8c9c7bb383449f49a6ff882
tests/test_cli_numeric_options_declare_a_range.py
CONFLICT (content): Merge conflict in tests/test_cli_numeric_options_declare_a_range.py

The cause is the transition (see Limits), not a defect of the layout. After the second merge it is clean, including against the latest dev (d4fcea8c, #1618):

$ git merge-tree --write-tree --name-only HEAD origin/dev
2fe5c7fad6b0e12655af0830d808bb0520976f5c
exit=0

CI

  • Step: .github/workflows/ci.yml, one step, "Check ratchet lists only shrink (pull_request only)", directly after "Check fix: commits have tests (pull_request only)", same if: github.event_name == 'pull_request' && matrix.shard == 1. It runs python scripts/check_ratchets_only_shrink.py --range HEAD^1 HEAD. It needs the same as the fix-commit step: fetch-depth: 0 (already set) so git show and git log can reach HEAD^1 and the PR's commits, and HEAD being the PR merge commit with HEAD^1 the base tip. No new dependency (stdlib and git only). Maintainer-merged file.
  • Ran and passed on an earlier head of this PR (3ae3f366): job test (3.11, 1), step 11. On that SHA gate and verify-red were green too.
  • On the pushed SHA d8cda527: CI was only queued when I posted this, so gate and verify-red for d8cda527 are not yet read. gate and verify-red were green on 3ae3f366, and the check suite on c0337ad1 finished with nothing failing. I did not read d8cda527's results.
  • verify-red on 3ae3f366 graded the new file's tests as import-only (weak; a new script has nothing for them to fail against, as in the red list above) and test_listed_rows_are_not_stale as an unexpected pass (only its message text changed). That job is advisory.

Limits (the transition)

Every open PR written against the old layout (#1619, #1620, #1622, #1569, and any other #1541 slice) conflicts once with this change, whichever of it and this PR lands second, in the same two places: (1) the removed group's lines sit next to the new # separator lines, (2) the NOT_YET_PINNED_SIZE line they lower is deleted here. After this is on dev, two removals do not conflict (the merge-tree demonstrations above).

The two-line recipe an author follows, after git merge origin/dev, in tests/test_cli_numeric_options_declare_a_range.py:

  1. Take this file's version of the list (the one with the # separator lines and no NOT_YET_PINNED_SIZE), then delete your module's group comment line and its rows, leaving the separator lines.
  2. Delete the NOT_YET_PINNED_SIZE = <n> line and its comment if the conflict keeps either (nothing lowers a pin any more), then commit the merge.

Unsure

  • Reviewers: look twice at read_rows and check_range in scripts/check_ratchets_only_shrink.py (it becomes a gate). read_rows takes the last module-level assignment, as import would.
  • A list that exists at BASE but cannot be read there counts as empty at BASE, so rows at HEAD then need a trailer. This is the strict direction; it is not covered by a dedicated test.
  • The separator is a #-only line, so removing one whole group leaves two # lines in a row. Harmless, and the list's comment says so.
  • My first attempt at the final kb: commit recorded only the rename, I amended it before knowing the first push had gone through, the push was refused, and I stopped (no force-push, no rewrite of the remote). That is why commits 7 and 8 are two. The maintainer instructed the recovery.

Left

For slice 2, from a dry read of git grep -nE '^[A-Z_]*(SIZE|COUNT|PINNED|TOTAL)[A-Z_]* *= *[0-9]+' -- tests plus an ast pass over each file:

  • The reader handles as written (a literal dict or set at module level): ALLOWLIST (dict, 52) in test_tests_drive_the_surface_not_the_mock.py; PUBLIC_ENTRY_POINTS, PUBLIC_API_OPERATIONS, REST_NOT_YET_MIGRATED, REST_WRITE_TIER_NOT_YET_MIGRATED, REST_READS_OVER_POST (dicts) in test_every_entry_point_passes_the_policy.py; ALLOWED (dict, 2) in test_response_text_is_pyrites.py; ALLOWLIST (dict, 0) in test_storage_never_imports_services.py; ENTRY_FILE_WRITERS_NOT_YET_CONVERTED (dict, 0).
  • The reader does not handle: MCP_NOT_YET_MIGRATED (built by a call, not a literal), and ALLOWLIST and COMPOSITION_ROOTS in test_layer_boundaries.py (dict displays whose values are not literals, so ast.literal_eval of the whole value fails; reading only the keys would need a change to the reader).
  • Not lists at all: the three characterization pins (PINNED_MCP_TOOL_COUNT, PINNED_REST_OPERATION_COUNT, PINNED_MCP_TOTAL_TOOL_COUNT) are plain integers from walks; the issue's option (A) applies. PINNED_MTIME in test_verify_red_ci.py is a timestamp, not a ratchet.

Learned

  • ruff format strips blank lines inside a set literal, so the group separator has to be a comment line. Captured in: the comment above NOT_YET in tests/test_cli_numeric_options_declare_a_range.py.
  • Two removals of adjacent groups conflict without a separator (exit 1) and merge with one (exit 0), by experiment. Captured in: the merge-tree demonstrations above and the same comment.
  • A new test file's subprocess calls need timeout= (tests/test_test_rules.py::test_no_test_call_can_wait_forever). Captured in: tests/test_check_ratchets_only_shrink.py (every call has timeout=60).
  • A branch that moves a ratchet's layout conflicts once with every open PR written against the old layout. Captured in: the Limits section above.

Tokens: not reported.

🤖 Generated with Claude Code

https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3

…atchet-raise trailer (#1608)

Reads each registered list at BASE and HEAD with git show + ast (never
imports the test module); removals pass, additions need the trailer, an
unreadable list fails by name. Registry holds NOT_YET only in slice 1.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3
…d so removals merge (#1608)

Drops NOT_YET_PINNED_SIZE and its two tests; the range check
(scripts/check_ratchets_only_shrink.py) now refuses growth. A comment-only
'#' line between module groups keeps adjacent group removals from touching.
test_listed_rows_are_not_stale is unchanged and still catches stale rows.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3

Copy link
Copy Markdown
Collaborator Author

Push refused, and I have stopped as the brief says. The cause is mine: I amended f0686ad ("kb: #1608 slice 1 is done") after it had already reached origin/claude/ratchet-not-yet-pin-ae38fa. I did not know a first push had gone through, because I stopped that background task before reading its result. The local tip is now a8eae1a. The two commits differ only in that f0686ad has the rename to kb/backlog/done/ and a8eae1a also has status: done in the file.

Nothing is force-pushed or rebased. The remote is at f0686ad, which is a clean state: CI was green on the earlier head 3ae3f36, and the status line is the only difference. Waiting for a decision on whether to merge origin/claude/ratchet-not-yet-pin-ae38fa into the local branch and push, or to leave the branch as it is. The report is not posted yet.


Generated by Claude Code

@markramm

Copy link
Copy Markdown
Collaborator Author

Conductor, answering your 14:51Z stop on PR #1614 (2026-11-17). Stopping was right. No force-push, no rebase, no amend from here on. Do exactly this:

  1. Get the local branch back onto the pushed history WITHOUT rewriting the remote:
    git fetch origin
    git reset --soft origin/claude/ratchet-not-yet-pin-ae38fa
    That moves only your local branch pointer to f0686ad and keeps your tree, so the status: done line is left staged as the one difference. Check git status shows only that file, then commit it as a NEW commit: kb: #1608 slice 1 backlog item status is done (a kb: commit, it touches no test, it must pass the commit hooks; never --no-verify on a commit).

  2. Merge dev with a merge commit: git merge origin/dev (dev is aa7deeb or later). One conflict is expected, in tests/test_cli_numeric_options_declare_a_range.py: fix(sw): declare min=0 on limit and offset for sw list commands (#1565) #1595 landed on dev, deleted the eight pyrite sw ... limit/offset rows and lowered NOT_YET_PINNED_SIZE. Resolution: keep YOUR layout (the # separator lines, no NOT_YET_PINNED_SIZE, no pin tests), and delete the software-kb group's comment line and its rows exactly as your own NOT_YET comment tells a contributor to do, leaving the separators. Nothing from dev's pin lines survives. If dev has more than fix(sw): declare min=0 on limit and offset for sw list commands (#1565) #1595 in that file when you merge (outside PR fix(wiki): declare min=1 on limit for wiki stubs (#1560) #1599, which removes the pyrite wiki stubs limit row, is armed to auto-merge and may land any minute), resolve the same way.

  3. This conflict is the last one of its kind, and it is evidence for your acceptance line. In the report say:
    (a) what the merge of dev conflicted on and how you resolved it;
    (b) if fix(wiki): declare min=1 on limit for wiki stubs (#1560) #1599 lands on dev AFTER your merge: run git merge-tree --write-tree --name-only HEAD origin/dev and paste the output. A clean result there is the live proof that two honest PRs removing different rows no longer conflict. If it is not clean, say so plainly and stop; that is a finding against the layout, not something to patch around.
    (c) the git merge-tree demonstration the acceptance asks for, as already planned.

  4. Run the affected tests, push with a normal push (--no-verify on the push only, and only past a failure you have shown to be unrelated), check git ls-remote origin claude/ratchet-not-yet-pin-ae38fa equals your local HEAD, wait for gate and verify-red on that SHA, then put the report in the PR body and post the one-line comment "Report posted at ".

Scope reminder: the one step in .github/workflows/ci.yml is the maintainer's to merge, so say in the report exactly what it runs and what it needs (fetch depth, base ref), and that the range check passes on this PR's own range. Do not touch any other ratchet in this slice.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3
One conflict, in tests/test_cli_numeric_options_declare_a_range.py: #1595
deleted the eight `pyrite sw` rows and lowered NOT_YET_PINNED_SIZE. Kept this
branch's layout (separator lines, no pin, no pin tests) and deleted the
software-kb group's comment line and rows, leaving the separators.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3

Copy link
Copy Markdown
Collaborator Author

Finding, and I have stopped as the brief says. #1599 landed on dev (814382d3) after my merge, and the live check is NOT clean.

$ git merge-tree --write-tree --name-only HEAD origin/dev     # HEAD = c0337ad1 (pushed), origin/dev = a5f942bd
2722c9b6c0b86c6ce8c9c7bb383449f49a6ff882
tests/test_cli_numeric_options_declare_a_range.py

Auto-merging tests/test_cli_numeric_options_declare_a_range.py
CONFLICT (content): Merge conflict in tests/test_cli_numeric_options_declare_a_range.py

Two hunks conflict. What I traced, read-only, nothing resolved or pushed:

  1. The pyrite wiki stubs group: fix(wiki): declare min=1 on limit for wiki stubs (#1560) #1599 deleted that group's comment line and row from the OLD layout, which has no separator lines. My branch adds the # separator lines on both sides of that same group, so the two edits touch adjacent lines.
  2. The pin: fix(wiki): declare min=1 on limit for wiki stubs (#1560) #1599 lowered NOT_YET_PINNED_SIZE from 32 to 31, and my branch deletes that line.

Both conflicts exist because dev does not yet have this branch's layout: #1599 was written against the old layout and the old pin. That is what I make of it, but it is a reading of the diff, not something I have proven. It does not show that two removals conflict on the new layout; my earlier git merge-tree demonstrations (exit 0, on this branch's layout) cover that case, and they did not include a branch written against the old layout. I read this as a one-time transition conflict for whichever of #1599 and this PR merges second. It is still a conflict, so this PR cannot merge into dev as it stands.

Waiting for a decision on how to proceed. Resolving it by merging origin/dev the same way as before (keep this branch's layout, delete the wiki group's comment line and row, drop the 31 pin) would be a new merge commit. I have not done it.

Done and green so far: test-affected 27469 passed on c0337ad1, and the range check exits 0 against origin/dev. The report is not posted.


Generated by Claude Code

@markramm

Copy link
Copy Markdown
Collaborator Author

Conductor, answering your 15:29Z stop on PR #1614 (2026-11-17). Stopping was right, and your reading is right: it is a one-time transition conflict, because #1599 was written against the old layout and the old pin. Proceed, no decision needed:

  1. git fetch origin, then git merge origin/dev as a NEW merge commit. No rebase, no amend, no force-push.
  2. Resolve tests/test_cli_numeric_options_declare_a_range.py exactly as last time: keep this branch's layout; delete the pyrite wiki stubs group's comment line and its row (dev removed them in fix(wiki): declare min=1 on limit for wiki stubs (#1560) #1599); no pin line.
  3. Prove the resolution in the report with one comparison: the set of rows in your list equals the set of rows in dev's list at the dev commit you merged (31 rows at a5f942b). Paste the command and its output.
  4. scripts/test-affected --run, push normally, check git ls-remote origin <branch> equals your local tip BEFORE doing anything else to a commit.
  5. Post the report now (PR body, "## Report"), then one comment: "Report posted at ". Do not wait for dev to stop moving. If dev moves again before your push, merge again the same way; after the report, the conductor tells you if another merge is needed.
  6. In the report's limits, state the transition plainly: every open PR written against the old layout (fix: reject invalid collection query limits before opening the KB #1619, fix(qa): declare ranges for --tier and --max-age on qa assess (#1564) #1620, fix(cli): declare ranges for --tier on ci and --port on serve (#1563) #1622, fix(social): reject nonpositive --limit values #1569 and any other Every numeric CLI option declares its range, and a test says so #1541 slice) conflicts once with this change, whichever lands second, in the same two places; after this is on dev, two removals do not conflict (your merge-tree demonstrations). Name the two-line recipe an author follows to resolve it, so the conductor can paste it on those PRs.

One conflict, in tests/test_cli_numeric_options_declare_a_range.py: #1599
deleted the `pyrite wiki stubs` group and lowered NOT_YET_PINNED_SIZE to 31,
both written against the old layout. Kept this branch's layout, deleted the
wiki group's comment line and row, left the separators, no pin line.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3

Copy link
Copy Markdown
Collaborator Author

Report posted at d8cda52


Generated by Claude Code

@markramm

Copy link
Copy Markdown
Collaborator Author

Review at d8cda52 (conductor, 2026-11-17)

Automated: read by the conductor from the pushed head, with one single-process run of the new script; no suite run. CI on d8cda527 had not finished when this was written (the test legs were still running), so this review does not claim the gate.

The theme's property holds on this head: removing rows from NOT_YET edits no other line, and adding a row is refused on the pull request unless a commit says why.

Checked by running, against dev at 38d98071:

Checked by reading:

  • The script parses with ast, never imports the file, and fails when the file, the name or a literal value is missing at the head, so it cannot pass by being unable to look. A list absent at the base counts as empty, so a new list needs the trailer. A trailer naming another list, or with no reason, does not count.
  • The CI step sits beside the fix-commit check in the test job (shard 1, pull_request only, same HEAD^1 HEAD range, and that job checks out full history), so gate waits for it.
  • The two deleted tests (test_not_yet_may_shrink_and_never_grow, test_the_pin_is_not_slack) are the pin itself. That deletion is the theme (A ratchet's pin never makes two honest pull requests conflict #1608, option B, approved 2026-11-17). test_listed_rows_are_not_stale stays and still refuses a row whose option has a range.
  • 38 tests of the script cover the trailer rules, the four container forms, a missing file or name, a non-literal, a merge commit that brings no row, and the real registry.

Limits, stated here so the merge is made knowing them:

  1. Growth of the list is now refused only by the CI step on a pull request. The local suite and the pre-push hook no longer refuse an added row; a contributor learns of it from CI. It still cannot reach dev without the trailer, because the step is in a required job.
  2. The trailer is the author's to write. It is visible in the commit and it is what a reviewer looks for; the check does not judge the reason.
  3. The transition costs each open pull request that edits the pin line one merge of dev (fix: reject invalid collection query limits before opening the KB #1619, fix(qa): declare ranges for --tier and --max-age on qa assess (#1564) #1620, fix(cli): declare ranges for --tier on ci and --port on serve (#1563) #1622, fix(social): reject nonpositive --limit values #1569, fix(cli): refuse --importance outside 1-10 on create and update as a usage error #1428). fix(qa): declare ranges for --tier and --max-age on qa assess (#1564) #1620 is in the merge queue now and will land first, so this branch needs one more merge of dev after it (delete the two qa assess rows, keep the layout, no pin line; prove the row sets equal as before).
  4. Only NOT_YET is registered. The other pinned lists are slice 2 of A ratchet's pin never makes two honest pull requests conflict #1608.

Both are the design as approved, not defects; none needs a follow-up beyond slice 2, which #1608 already holds.

For the maintainer: this changes .github/workflows/ci.yml (one step), so the merge is yours. Recommended: merge once gate is green on the head that follows #1620's landing. Until it lands, the loop holds other #1541 slices out of the queue so that the branch is not chased by a new conflict each time.

Recommendation: merge, after the merge of dev that follows #1620 and a green gate.

@markramm

Copy link
Copy Markdown
Collaborator Author

Conductor, 2026-11-17T18:38Z, for PR #1614 (#1608 slice 1). #1620 has MERGED (dev 5beda72, 18:20Z). dev is now f95e9b3 (#1637 and #1611 landed after it; neither touches your files). This is the third merge of dev you were asked to hold for; GitHub now shows the PR as conflicting, which is expected.

Do now, as already instructed:

  1. git fetch origin and merge origin/dev by a NEW merge commit. No rebase, no force-push, no --no-verify on the commit.
  2. Resolve tests/test_cli_numeric_options_declare_a_range.py by keeping your layout (no pin line) and taking dev's row set: fix(qa): declare ranges for --tier and --max-age on qa assess (#1564) #1620 deleted the two qa assess rows. Your NOT_YET must equal dev's set exactly.
  3. Prove it: the row count on your head and on origin/dev, and the symmetric difference of the two sets (must be empty); the range check over origin/dev..HEAD exits 0; scripts/test-affected --run.
  4. Push, confirm git ls-remote origin <branch> equals your HEAD, and post ONE comment on test: the numeric-option ratchet is its own pin; a range check refuses an added row (#1608, slice 1) #1614: the new SHA, the two counts, the empty difference, the range check's exit code.

Then stop. Do not mark the PR ready: it changes .github/workflows/ci.yml, so the merge is the maintainer's. Three PRs are in the merge queue behind this (#1645, #1640, #1638); none touches the ratchet file.

One conflict, in tests/test_cli_numeric_options_declare_a_range.py: #1620
deleted the two `qa assess` rows and lowered NOT_YET_PINNED_SIZE to 30, both
written against the old layout. Kept this branch's layout, deleted the qa
group's comment line and rows, left the separators, no pin line.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013BnEiZFj8CP7nEnkefXVU3

Copy link
Copy Markdown
Collaborator Author

Third merge of dev posted at 81ec516

  • Merged origin/dev at f95e9b3b as a new merge commit. The only conflict was tests/test_cli_numeric_options_declare_a_range.py, in the same two places: fix(qa): declare ranges for --tier and --max-age on qa assess (#1564) #1620's deletion of the two qa assess rows next to the separator lines, and its lowering of the old pin. Resolved by keeping this branch's layout, deleting that group's comment line and rows, leaving the separators, no pin line.
  • Row counts: head 29, origin/dev 29.
  • Symmetric difference of the two NOT_YET sets: [] (empty).
  • python scripts/check_ratchets_only_shrink.py --range origin/dev HEAD: exit 0, no output.
  • scripts/test-affected --run on the merged tree: 28495 passed, 258 skipped, 7 xfailed, 0 failed.
  • git ls-remote origin claude/ratchet-not-yet-pin-ae38fa equals local HEAD (81ec5163…), checked right after the push.
  • git merge-tree --write-tree HEAD origin/dev: exit 0.
  • gate and verify-red on 81ec5163 have not been read yet.

PR left as a draft; not marked ready.


Generated by Claude Code

@markramm
markramm added this pull request to the merge queue Nov 17, 2026
Any commits made after this event will not be merged.
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.

2 participants