A 0 -> 1 binding is two states: split AuthoredReferenceResolution out of NewPoolCoincidenceResolution

[briansrls]

The defect

binding_disposition (the wave-admission wall, #9365) compares candidate sets
only, so every binding whose set went {} -> {one} is NewPoolCoincidenceResolution,
which refuses unless an operator admission names it. The function's own header says
that arm means "a name that began denoting something WITHOUT ANYONE AUTHORING A
REFERENCE — resolution arriving from a pool".

0 -> 1 is two states sharing one symbol, with opposite owners and opposite
repairs:

	cause lives in	repair
the target grew a name this module was already reaching for	another module	the coincidence the containment rule exists to remove — refuse
this module's own author wrote the import resolving a name it already spelled	this diff	the repair the wall was built to want — admit

The wall was refusing the second while naming it the first.

How it was found

gunbc#9485 — a one-line import repair, 5 blocking diagnostics -> 0 on a
discriminating pair — could not merge. It could not be admitted either:
NAMESPACE_TRANSITION_ADMISSIONS is a const &[TransitionAdmission] over a
DeltaSubject carrying String fields, so a row naming an actual module is E0015
(cannot call non-const associated function). That second half is escalated
separately and is deliberately not repaired here — this change removes the need
for admission rows in this class, so the unauthorable roster is off the critical
path rather than fixed by a rider.

The fix

The discriminator is the module's own source, and it was already in hand: the
membership arm one function over consults membership_supported, which is what
auto-admitted the membership edge of the very change whose bindings it refused.

locally_authored_claim_added answers it from the two ModuleDeclarationRecords
alone — deliberately index-free, because consulting either index reintroduces
the conflation (a blanket import m whose target grew the leaf would read as
authorship). false is the fail-closed answer: it leaves the delta on the refusing
arm, so nothing here can turn a refusal into silence by accident.

AuthoredReferenceResolution auto-admits, amending the 2026-08-26 operator
partition recorded on gunbc.compiler_frontend_program_interlock per the
2026-08-27 ruling relayed through swift-badger-524. The .dag coproduct is the
authority and the host enum is one realization; the vocabulary-join tests in both
directions gate the pair.

Executed evidence

The test file asserted NewPoolCoincidenceResolution over a head that authored
import probe.home { widget } — it planted the repair and named it the
coincidence. That is why the conflation survived review, and re-deriving the arm
required authoring a coincidence that actually is one: consumer source
byte-identical across both sides, reaching the target through a blanket import,
target grows the name.

Both arms mutation-checked, opposite directions:

• locally_authored_claim_added forced true → coincidence arm reds alone
• forced false → authored arm reds alone
• restored: 20 passed; 0 failed

cargo fmt --all --check clean.

[gunbai-bot]

Fixed in 49e58ea. Review 56882 is correct, and the fail-open direction is what makes it worth more than a scoping nit: the arm could only ever turn a refusal into an admission, which is the one direction this function must never move, and my own docstring asserted the opposite.

The arm is now a conjunction, and the order of its halves is the point: the blanket claim must be NEW IN THIS MODULE'S SOURCE and the head target must actually supply the leaf. Your parenthetical offered the alternative of accepting that this discriminator needs index consultation for blanket imports — that is what landed, but gated rather than accepted outright. Asking the surface question alone is exactly the conflation the function exists to discriminate (an unchanged blanket import whose target grew the leaf would read as authorship); asking it only after the claim is established as newly authored makes it safe, because a claim the author did not write cannot reach the surface check whatever the target did. A target absent from the index answers false.

So the docstring's index-free claim is corrected rather than restated — it was true of the leaf-named arms and never of this one.

Two tests, both directions of the scoping:

• an_unrelated_new_blanket_import_does_not_launder_a_pool_coincidence — your scenario, authored exactly: probe.home grows widget with its blanket import UNCHANGED, while the author adds a blanket probe.other that does not supply widget. Must stay NewPoolCoincidenceResolution and must stay refused. Restoring the leaf-blind arm reds this test alone — measured, not asserted.
• a_new_blanket_import_that_does_supply_the_leaf_is_an_authored_reference_resolution — so the scoping is not a narrowing to nothing: when the NEW blanket target is the one supplying the leaf, it auto-admits like any other authored reference.

22/22 green, fmt clean.

— sent from merry-lynx-302

[gunbai-bot]

CI investigated on 01ede37. There is no fix for this PR to push — the failure is main's, and this run carries the positive receipt that the change works.

The phase this PR is about PASSED, on the live tree:

required-ci: namespace-wave-admission base=9538523a3c head=58bbc70c88 modules_compared=4196
required-ci: namespace-wave-admission ADMITTED — every delta is auto-admitted or named by a transition admission

That is the whole point of the change, executed against the real corpus rather than a fixture: before it, the wall refused authored-import resolutions and had no authorable admission roster to relieve them.

The two failing phases are dag/product/fabric/contention.dag, and nothing else. Every .dag error in the run is one of four, all from one removal:

• 23:19 name 'grant_duration_seconds' not found in module 'product.fabric.supply'
• 280:3 non-exhaustive match: missing variant(s) UnobservedGrantDuration
• 421:3 non-exhaustive match: missing variant(s) QuoteNotPriceableWithoutDuration
• 463:30 and 469:43 function 'grant_duration_seconds' not found in scope

This branch touches src/v1/stage0/src/namespace_wave_admission.rs, its test file, and dag/gunbc/compiler_frontend_program_interlock.dag. It cannot produce any of those.

Fixed by #9488, which is open and whose declarations phase already passes. One correction worth recording, because the dashboard renders it as a red: #9488's floor job did not fail — its log ends ##[error]The operation was canceled after a clean compile.emit, so it was superseded, not refused. A cancelled check is not a failed one and not a passed one.

The srv3_wiring_with_a_duplicated_header duplicate declaration I reported on the earlier run is gone — fixed on main since.

This PR goes green when #9488 lands and this branch takes a run on top of it. Nothing here to rework.

— sent from merry-lynx-302