GeezerVison

Reviewed 0b0159dc0cd42ae4bc5f93f81c601e26f9f3f7a2 against confirmed main/merge-base faa1c694c66a3732aaabbce93bdae140e521a67d.

Standards

No documented-standard findings.

Spec

The original #3861 single-wildcard case is fixed in my CPU checks. Both newly committed regressions pass here and fail when run against main’s implementation.

Requesting changes for one new P2 in the inline comment: when a concrete registration has no lifecycle hooks, hook lookup can select a hooked wildcard registration even though scheme lookup, requirement construction and settlement enrichers select the concrete implementation. An aborting wildcard hook incorrectly blocks verification or settlement before the in-process facilitator is called.

Independent checks on macOS arm64 / Go 1.24.5:

• Core and HTTP committed suites with -race -count=1: 233 top-level tests pass, 1 existing test skips; baseline 231 pass, 1 skips. The two new tests are included, not additional to that total.
• 120 directed selection controls across three fresh runs: head 105 pass / 15 fail; baseline 82 pass / 38 fail. The 15 head failures describe the single hook-isolation regression across three registration arrangements, not 15 bugs.
• Six minimal abort variants (verify/settle, both registration orders and exact re-registration): all fail here and all pass at main in three runs. Unrelated network/scheme controls pass.
• Builds, vet and changed-file gofmt/whitespace checks pass for the reviewed packages.

Scope: local fixture schemes and facilitator doubles, not valid signatures, live payment/settlement/RPC, production, full Go SDK/monorepo or Linux/Go 1.26.6 verification. Full golangci-lint/goimports were not run.

Rechecked 25eba09956152c5238b0441364eac3f0796f445f against the confirmed main/merge-base faa1c694c66a3732aaabbce93bdae140e521a67d and the prior reviewed head 0b0159dc0cd42ae4bc5f93f81c601e26f9f3f7a2.

Standards

No documented-standard findings.

Spec

The hook-isolation issue from my previous review no longer reproduces in the covered cases. Resolving the registration in s.schemes before reading its hook slot keeps an exact registration without lifecycle hooks from inheriting wildcard hooks. Both registration orders and replacement of a hooked exact implementation with an unhooked one pass. Hook dispatch and settlement payload/response enrichers select the same exact implementation in these controls; the single-wildcard behavior from #3861 remains working.

Independent CPU checks on macOS arm64 / Go 1.24.5:

• Core and HTTP committed suites with -race -count=1: 234 top-level tests pass, 1 existing skip. The new regression and its three cases are included, not added again to this total.
• The same 120 selection controls in three fresh runs: 120/120 pass here; the previous head still gives 105 pass / 15 fail. Six minimal verify/settle abort variants pass 6/6 here and fail 6/6 on the previous head. Two unrelated-registration controls also pass.
• The author’s new exact-without-hooks regression fails in all three arrangements when run against the previous implementation. The two original wildcard regressions still fail against main’s implementation.
• Exact/wildcard precedence, partial/no-hook isolation, re-registration, failure/cancellation dispatch, requirement construction, settlement enrichment ownership and caller-payload preservation pass in the directed controls. Builds, vet and changed-file gofmt/whitespace checks pass.

No new blocking finding in this scope. These checks use local fixture schemes and facilitator doubles, not live signatures, payment/settlement/RPC, Linux parity, full SDK/monorepo or an overlapping-wildcard specificity guarantee. Full golangci-lint/goimports were not run.

Approving 25eba09956152c5238b0441364eac3f0796f445f following my follow-up review. The prior hook-isolation finding is addressed: an exact registration without hooks no longer inherits wildcard hooks. I have no further blocking findings in the reviewed scope.

Prior exact-head validation on 2026-10-22 (macOS arm64 / Go 1.24.5): core/HTTP race suites passed 234 top-level tests with one existing skip; 120 selection controls and six minimal abort controls passed in three runs. Builds, vet and changed-file formatting checks passed. These used local fixture schemes and facilitator doubles, without live payments/RPC or full-SDK/Linux validation. No tests were rerun for this approval.

CI attempt 2 remains red on merge snapshot 73ed6ec06085361bbd5fd845a190de0d238fa5d3: Go compilation fails at eip3009.go:109/112, and two Next default-response assertions fail. Read-only comparison confirmed that the failing Go file and the entire TypeScript subtree match base c2a8924eb5dfe53806352d73c46a87d2a9e34038, outside this PR's five-file Go diff. Those CI failures still need maintainer attention before merging.