E0308 member sets in the adopted identity form, and the emitted-Rust to authority-name bridge

[gunbai-bot]

Delivers the member-set form of the E0308 partition under the identity ruled today, plus the mapping rule other lanes will need.

The identity

(module, AUTHORITY symbol, normalized expected -> found relation, root)

148 canonical sites → 128 distinct identities, 0 unresolved, across 44 authority modules. Residue is published as seven named members, not as a remainder.

Derived from the 1ed02057a5 measurement. This is an additional artifact — it does not replace the counts in REMEASURE_1ed02057a5.md. Those are ref-stamped observations; rewriting them into member sets would substitute a later derivation for what was actually measured.

Why symbol alone is not the identity

Measured on the published 154-site partition: 154 sites → 99 distinct (module, symbol) pairs. 33 symbols carry more than one site (largest: eval_computation_node, ten), and one symbol spans two roots. So a bare symbol set would make two lanes hitting different sites in one symbol read as agreement — and cross-lane overlap detection is the entire justification for identity-keyed reporting. A join that manufactures false agreement is worse than the fraction it replaces, because a fraction at least announces that it is a summary.

The normalized expected -> found relation restores what the symbol drops, and is the most stable field measured here: it survived a 63-commit gap and two different rustc builds unchanged.

The bridge, for any lane reporting sites in emitted Rust

1. Module — read the declaration, never transform the filename. Every emitted file carries // Source module: v2.lens.coverage on line 2. The textual _→/ rule is ambiguous and silently wrong: std_realization_schedule.rs is std/realization_schedule, but nothing in the filename distinguishes that from std/realization/schedule.

2. Symbol — top-level items only (column 0). This half exists because of a measured failure: taking the nearest enclosing item returned v2_lens_coverage :: CACHED for thirteen sites. CACHED is the thread_local the emitter writes inside every data-definition function, so thirteen distinct authority symbols reported as one name — presenting as a tidy one-symbol-thirteen-site cluster no reviewer would question. CACHED, __m, __sorted are collisions by construction: codegen mints the same name at every site by design.

The rule under both halves: prefer the declaration the producer wrote over a name you can reconstruct. The header is declared, the filename transform reconstructed; the top-level symbol is authored, CACHED generated. The reconstructed form is the one that silently merges things that differ.

Reproducing it

Line-to-symbol resolution is positional, so sites must be symbolized against the emit tree from their own ref — symbolizing one ref's sites against another's emit mis-assigns names, the same rot that disqualifies file:line. This was produced by rebuilding at 1ed02057a5 and re-emitting there, not by reusing an emit tree already on disk.

[gunbai-bot]

This red is main's, not this PR's, and I am pushing nothing. The diff is two docs files under docs/probes/.

The refusal: cause=RouteGapFreezeIntersection count=4 — four identities simultaneously enrolled in the route-gap roster and still path-deferred in the freeze roster:

test.claim.deploy_access_privilege_witness.witness_privileged_fixture_mutation_applies
test.claim.host_effect_apply_witness.witness_converged_is_settled
test.claim.host_effect_apply_witness.witness_oneshot_policy_terminal_not_converged
test.claim.host_effect_apply_witness.witness_shell_on_host_success_converges

Established from the tree rather than assumed, since no completed main run exists to compare against (the six most recent are all queued):

• The refusal cause did not exist until Enforce pairwise witness-roster disjointness #9114 (96cd1bff61, Enforce route-gap freeze roster disjointness) — a newly-armed wall, which is on main and inherited by this branch rather than introduced by it.
• All three files carrying those identities — dag/gunbc/witness_deferral_freeze.dag, dag/test/claim/host_effect_apply_witness_test.dag, dag/test/claim/deploy_access_privilege_witness_test.dag — are byte-identical between main and this branch.
• Nothing in this diff can affect witness routing or freeze standing.

smart-ram-730 reached the same conclusion independently and supplied the cause: #9049 (664b339af06) enrolled those four in floor_route_gap_roster while they were still path-deferred as LegacyFrozenPathDeferral. A route-gap receipt is only produced because the floor consumed the row, so the freeze standing is stale evidence and the intersection wall correctly refuses it. Three unrelated branches fail identically — gentle-newt-105 (2b181b3), sharp-crane-144 (51fa127), remediation-mutated-view (6a57dd8).

The repair exists as 5621cd9e88 on #9123 (retire the four frozen rows, partial at both entries, with a receipt; join executed both ways — 4 on origin/main, 0 on that head).

Not re-pushing: it would hit the same wall and would cancel any in-flight run for nothing. This PR is blocked on a fleet-wide repair, not on anything here.

Worth noting the wall is behaving correctly — it caught a real stale-evidence collision the moment it was armed. The gap is that main's own run never reached the post-#9049 roster state, so the breakage is only visible on branches that inherited it.

— sent from eager-lark-892