Declare the dependency gunbc.code_change_workflow and its witness already had (13 effect-summary-incomplete floor errors)

[gunbai-bot]

The defect

The floor refusal on PR #11907 (run 35555630169, job 106198684855) carried 13 effect summary incomplete errors over test.claim.code_change_workflow_witness_test calling gunbc.code_change_workflow.decide_code_change_transition and .classify_github_pr_terminal_anchor — each saying the callee's identity is established but names no registry row.

The chain, re-derived (DESIGN §6b)

The 13 are the symptom link. The same run carries, upstream, two errors nobody had joined to them:

dag/gunbc/code_change_workflow.dag:17:12: unresolved type 'ProcessNodeId'
dag/test/claim/code_change_workflow_witness_test.dag:307:32: unresolved type 'NotReady'

• ProcessNodeId is declared by gunbc.process_algebra; NotReady by gunbc.pr_digests.
• Neither module declared a dependency on either. Both carried no import header at all, so they resolved through the loader's bare-reference channel (bare_reference_pull_paths_for_source), which is applied only to import-free sources. That channel did not pull gunbc.process_algebra for this reference, so the type never resolved. Why it did not is not established here: I probed the obvious hypothesis — that an alias-shaped type X = NonEmptyStr where brand(...) reference is not "pullable" — against test.claim.extdeps_version_base, an import-free module whose only edge to extdeps.version is the same alias shape, and it compiles clean. So that hypothesis is falsified and the discriminator is unidentified. It is a separate question about the loader's bare channel, not a premise of this repair.
• With its type unresolved, gunbc.code_change_workflow is not compile-clean, so it drops out of the compiled closure into the name census only ([census] … indexed modules outside the compile-clean closure enter the name census only (not compiled)).
• The name census still establishes the callee identity; the item registry is built only over realized modules, so it has no row. unjoinable_callee_edges (v1.compiler.infer_service) then reports ResolvedCalleeRegistryRowAbsent — once per calling witness, 13 times.

The repair

At the earliest unjustified boundary: both modules now declare the dependencies they always had. Nothing is silenced, no module is excluded from the closure, and no witness stops calling either function.

Evidence — discriminating pair

Same invocation both sides, over a subject that reaches the witness module:

gunbc compile --output-dir <tmp> --source-root dag --source-root src/v2 \
  --entry dag/test/claim/code_change_workflow_witness_test.dag

	blocking errors	effect-summary-incomplete over this witness
control (main)	22	13
head	7	0

The 22 on main are 13 effect-summary + the 2 unresolved types above + 7 pre-existing in dag/gunbc/runner/microvm_controller_app_key_converge.dag; head keeps only those 7, which this change does not touch (see below). gunbc run --function all_code_change_workflow_witnesses evaluates to true on head, so the witnesses still discriminate (the run then refuses only because the fold returns Bool rather than ProcessExit).

modules_excluded=1, as asked

Not a silent drop. The floor's exclusion roster is floor_prepared_subject_exclusions() — four rows, all designed non-members: test/probe/ (must-not-resolve probes, consumed by class through gunbc.compile_diagnostic_census), test/fixture/meta_exec_confinement_scan/, test/manual/ownership_movable_test.dag, and the wet case4_expansion_carrier_splices receipt. Exactly one module in #11907's index matched one of those rows. The count is subject-relative (PR #11932's floor reports modules_excluded=0 over its own 1960-module subject), and this PR adds no module and removes none, so it cannot change which rows match.

Floor evidence: not obtainable on current main

The brief asked for the pair on the floor itself. My run (35561659566) could not produce one, and the reason is upstream of this change:

required-ci: parse FAIL …/dag/gunbc/runner/microvm_controller_app_key_converge.dag:677:3:
  source annotation sits inside a declaration body …            (x7)
required-ci: floor refused: REQUIRED-FLOOR REFUSAL cause=ArmSetConsumerPlanningUnavailable
  no parse-phase declaration index was lent to the floor …

The parse phase refused on that main-side file, so no declaration index was lent, so the floor refused at planning and never prepared a subject. The floor did seed the right modules first — [floor-phase] phase=touched-entry-compile-subject seeds=2 modules=["gunbc.code_change_workflow", "test.claim.code_change_workflow_witness_test"] — so the subject would have reached the witness. The entry-compile pair above is the same compiler over a closure that does reach it. (PR #11935's run at an adjacent main reports parse OK 6522 file(s); I have not identified what differs, and I am not claiming the parse failure is universal — only that it is the reason my run has no subject, and that the file it names is unchanged by this PR.)

Flagged, out of scope

dag/gunbc/runner/microvm_controller_app_key_converge.dag (landed 2026-09-21 in #11902) carries 7 blocking source annotation sits inside a declaration body errors. They do not appear in run 35555630169 — that subject did not reach the module — but any closure that does reach it refuses. Not touched here; scope for this PR was the effect-summary class only.

🤖 Generated with Claude Code