integration/wave-1 is superseded and must not be merged to main

[ChelseaKR]

integration/wave-1 is still on origin, 12 commits "ahead" of main. It looks like unmerged work. It is not, and merging it would regress main. This records the check so the next reader does not have to redo it.

It carries nothing main lacks

$ comm -13 <(git ls-tree -r --name-only origin/main | sort) \
           <(git ls-tree -r --name-only origin/integration/wave-1 | sort)
(no output)

Its two once-unique files, src/exitdrill/atomic_write.py and tests/test_atomic_write.py, landed on main byte-identically as #127. Everything else on the branch is an earlier variant of work main now has, usually under a different test name:

on integration/wave-1	on main
test_partial_scoped_shortfall_is_counted_by_the_fail_closed_floor	test_restoration_only_dimensions_report_exactly_the_shortfall (#117)
test_is_sha256_hex_accepts_what_the_hashers_above_produce + rejects	test_is_sha256_hex_accepts_only_what_the_hashers_emit (#119)
the bound renderer / discovery / binding trio	test_every_declared_bound_constant_appears_in_the_bounds_tables and the threat-model bound tests (#121)
test_the_json_recursion_arms_are_named_with_the_depth_that_still_reaches_them	test_the_json_recursion_arms_are_still_present_and_still_reachable (#136)
test_the_subcommand_binding_rejects_a_name_no_document_carries	same name, ported to main's own matcher (#137)
atomic_write.write_bounded_file and its four tests	identical, via #127
the closed-key validator (#91), the digest predicate (#92)	contracts.py (#116), canonical.is_sha256_hex (#119)
scripts/ coverage + the three floors (#86)	#125
the 3.12/3.13/3.14 matrix (#90)	#118, plus an aggregate verify job the branch does not have

Merging it would take things away

main is 13 commits ahead of the merge base on its own side. The branch predates, and so would fight with:

• src/exitdrill/contracts.py — the branch does not have this file at all;
• the aggregate verify job that keeps the branch ruleset's required check name under a matrix (ci: verify every interpreter requires-python declares (#90) #118);
• CI job timeouts, dependency caching and run cancellation (ci: add job timeouts, dependency caching and run cancellation #107);
• CodeQL on pull requests and the weekly schedule rather than on push (ci: run CodeQL on pull requests and the weekly schedule, not on push to main #108);
• the README link fix (docs(readme): point the closing link at a page that exists #114), the validate-exercise docs and subcommand binding (docs: describe validate-exercise and bind every subcommand to a document #115), the published input bounds (docs: publish every declared input bound and bind each to its constant #121);
• scripts/ coverage measurement, the per-scope floors, and EXITDRILL_REQUIRE_GATE_TOOLS in CI (ci: measure scripts/ coverage, floor each scope, and require the gate tools in CI #125).

It would also land Claude attribution in permanent history

Seven of its twelve commits carry a Co-authored-by: Claude trailer and a Claude-Session: line:

b9bbf67 04174616 338e78d4 9111ac79 68e51e24 217dc097 2ceb2318

The standing rule is that no Claude attribution appears anywhere in git history. A squash merge composes its message from the branch's commits and appends co-author trailers it finds there, so merging this branch — by any route — would add all seven to main at once. History is never to be rewritten to remove them afterwards, so the only control is not adding them.

What to do

Owner's call, and nothing is urgent:

• Delete the branch, since main now contains everything it was for; or
• Keep it as a record of the wave, with this issue as the reason it is not merged.

Either way: do not merge it, and do not cherry-pick from it without diffing against main first — most of what looks missing is present under another name.

Prepared with AI assistance; reviewed before submission.

[ChelseaKR]

Re-checked independently against origin/main at ae9e274 (2026-09-06), because this issue is the only thing standing between the branch and someone merging it.

The finding holds, and it has got stronger since it was written.

$ comm -13 <(git ls-tree -r --name-only origin/main | sort) \
           <(git ls-tree -r --name-only origin/integration/wave-1 | sort)
(no output)

$ git diff --stat origin/main origin/integration/wave-1 | tail -1
34 files changed, 975 insertions(+), 1273 deletions(-)

The branch carries no path main lacks, and merging it would now remove about 300 more lines than it adds. main has moved on again since this issue was filed: test_release_versions.py (#139) does not exist on the branch and shows up in that diff as a 268-line deletion, alongside the ones already listed here.

The two neighbouring branches are in the same state and were checked the same way. refactor/evaluator-inert-guards (#84, #96) is superseded by #117, #120 and #126; ci/coverage-scope-and-python-matrix (#86, #90) is superseded by #118 and #125, which also added the aggregate verify job the branch does not have. Neither carries anything main lacks either.

Nothing was deleted or merged. The disposition is still the owner's call, and the "do not merge it" instruction is why this issue stays open.

Prepared with AI assistance; reviewed before submission.