One HTTP method set, homed at its upstream: extdeps.ietf.http_semantics

[gunbai-bot]

De-fork program (node adhoc-badbfc4d-1ab), first wave.

Decision record: HttpMethod

Pair. std.types HttpMethod (7 arms) and v2.std.network HttpMethod (9 arms). The Phase 0 census first called the pair IDENTICAL, but that compared body text. The arm sets differ: v2 adds CONNECT and TRACE. So the v2 set is a strict superset, and the std set was an incomplete transcription, not a narrower concept.

Dominant. Neither. The request-method set is an IETF fact, so under DESIGN §3 (external upstream decomposition) its home is an extdeps module that cites the spec. There are no re-exports: both std declarations are deleted, and the one declarer is extdeps.ietf.http_semantics HttpMethod. It cites RFC 9110 §9 and, separately, RFC 5789 for PATCH, which is not an RFC 9110 method. WebDAV and the other IANA-registry extension methods are not modeled; a wire label outside the set still refuses by name (extdeps.http.server parse_http_method_label → UnknownMethod). DFS first: no existing module (extdeps.http.server, extdeps.http.client, extdeps.bmc.http) already carried the set.

What the two new arms mean at each consumer that matched over the seven:

consumer	TRACE	CONNECT
extdeps.http.server parse/label	known method	known method
extdeps.transports.rest rest_method_verb	Read (safe, RFC 9110 §9.3.8)	Write (not safe, and the tunnel's traffic is invisible to the grant)
gunbc.auth.approval_broker decision_method_admission	retrieves only	unsupported
std.effects derive_effect_shape	ReadEffect	typed refusal: MethodHasNoEffectShape → derive_op_effect answers MethodHasNoOperationEffect

CONNECT names no resource operation, so no EffectShape arm describes it. Any shape there would be invented, and the idempotency obligations downstream would trust it (DESIGN §5). derive_effect_shape now returns MethodEffectShape.

Also removed. std.types TransportRequest / TransportResponse. They were the only std.types use of HttpMethod, they have no consumers anywhere, and keeping them would have created an import cycle (std.types → extdeps.ietf.http_semantics → std.decl_ref → std.types).

Importers. Every explicit import of the method names from std.types or v2.std.network now points at the one declarer. Files that use a method arm without importing it are left alone: they resolve through corpus-wide lookup, and that lookup now sees one declarer where it saw two. Explicit-import hygiene belongs to the namespace-cut program, and a sweep here would be the file-by-file import sweep the program brief rules out.

Evidence

• Control: test.claim.http_method_set_witness. Its discriminating case is connect_and_trace_parse_as_known_methods: on main the parser answers UnknownMethod for both labels. It also covers the round trip of all nine labels, the verbs, and the CONNECT refusal vs. the TRACE read in derive_op_effect.
• Approval broker: test.claim.approval_broker_witness_test trace_is_refused_as_a_safe_method (exactly RedemptionMethodNotCommitting) and connect_is_refused_as_unsupported (exactly RedemptionMethodUnsupported). At 52e08d7 on srv1 (claim_batch built at that head, --hermetic): 15 PASS, 0 FAIL. Mutation controls in decision_method_admission, each restored afterwards: TRACE → DecisionMethodCommits turns exactly trace_is_refused_as_a_safe_method red, and CONNECT → DecisionMethodCommits turns exactly connect_is_refused_as_unsupported red.
• Stage0 mirrors are regenerated, not hand-edited (partition list, the std-core root, std_types, std_effects, v1_compiler_effect_derivation, emitted_population, and the new extdeps_ietf_http_semantics). A second --required-regen is byte-identical; the full receipt is in the PR comment. At that regen (0be933b) the four affected test files (http_method_set_witness, effects_witness, approval_broker_witness, http_serve_route_witness) gave 34 PASS, 0 FAIL.

Fixes the queue surfaced (separate commits)

• extdeps.github.rulesets: rule.parameters is optional (RulesetRuleParametersWire?) but was passed to the required-status-checks and merge-queue missing-field checks, which require the record. An absent parameters object now reports ...ParametersIncomplete { missing: ["parameters"] }. The floor refused both call sites once this PR brought the module into its checked set.
• gunbc.runner_guest_image: it called Filesystem.Write(...) without naming a source, and Filesystem is declared by both std.resources and extdeps.filesystem.filesystem_io. That ambiguous bare read (AmbiguousBareNameRead, surfaced as PureProducerShareRowModuleUnframeable on bash_command_fold) is what dequeued this PR. It now imports extdeps.filesystem.filesystem_io { Filesystem }. This fix has no control, and none can be written today. It is correct by reading the call: .success and .error exist only on the service's Write; the std.resources resource has a lowercase write returning written. A test that goes red on the wrong import can't be written on any route yet. On the seed, the service-receiver stamp matches by bare name (the defect: a wrong import to std.resources type-checks and emits green). On v2, there is no resource production and namespace_graft doesn't place service declarations. This is recorded as the fourth specimen of gunbc.recurring_failure_mode surface_shorthand_preempts_resolved_identity, with both capabilities as triggers: (a) the seed stamp matches owner module plus name, which is being dispatched as its own lane with a two-declaration seed fixture control; (b) v2 service grafting plus a resource production.

🤖 Generated with Claude Code

[gunbai-bot]

Regeneration receipt, at head 0be933b (run on srv1 by neat-boar-16):

• Installed only this change's mirrors: gunbc_stage0_crate_partition_generated.rs, extdeps_ietf_http_semantics.rs (new), the std-core lib.rs (via --emit-partition-crates --write, written=1), emitted_population.rs, std_types.rs, std_effects.rs, v1_compiler_effect_derivation.rs. The other drift a --required-regen shows on main (std_integer, std_machine_constraints, v1_compiler_infer_resolve, v1_compiler_emit_rust) belongs to other lanes and is not in this PR.
• The partition .dag was regenerated by generated_artifact_gate main_wet_one (ci auto-heal e8a3417 had committed the identical change).
• A second --required-regen is byte-identical for every installed file.
• claim_batch --hermetic over test.claim.http_method_set_witness, effects_witness, approval_broker_witness and http_serve_route_witness: 34 PASS, 0 FAIL.
• The rulesets change (c06d84c) fixes a latent optional-into-required argument that the floor refused once this PR brought the module into its checked set.