The tier-2 event-feed conformance family assigns below-the-poll-seam obligations to Layer-1 adapter conformance (the README's row-15 note, schema.json's fixture-30 respond variant, and SPEC §23's zero-egress paragraph all name it), but that layer does not exist yet. Until it lands, prose in those places pointed at "G1b" — a connector-program milestone label defined nowhere in this repository, so the obligation was undiscoverable and unguarded.
The concrete first obligation: the adapter's own 302 test — a real generated PollEvents call meeting a real redirect against an adapter with automatic redirect-following disabled — proving zero requests to the refused URL with a sentinel that can actually observe one. Tier 2 structurally cannot see this (its poll lane is a seam; the redirect's refusal is exercised there via the shipped per-hop predicate, but "the adapter never follows" is a wire fact): until this test exists, the follow-cross-origin-redirect mutation is not killed at any tier.
The five prose sites that reference this obligation cite this issue and are enforced by the tracking-issue registry (spec/tracking-issues.yml), so closing this issue without landing the test will fail CI until the sentences are rewritten — which is the designed behavior.
The tier-2 event-feed conformance family assigns below-the-poll-seam obligations to Layer-1 adapter conformance (the README's row-15 note, schema.json's fixture-30 respond variant, and SPEC §23's zero-egress paragraph all name it), but that layer does not exist yet. Until it lands, prose in those places pointed at "G1b" — a connector-program milestone label defined nowhere in this repository, so the obligation was undiscoverable and unguarded.
The concrete first obligation: the adapter's own 302 test — a real generated
PollEventscall meeting a real redirect against an adapter with automatic redirect-following disabled — proving zero requests to the refused URL with a sentinel that can actually observe one. Tier 2 structurally cannot see this (its poll lane is a seam; the redirect's refusal is exercised there via the shipped per-hop predicate, but "the adapter never follows" is a wire fact): until this test exists, thefollow-cross-origin-redirectmutation is not killed at any tier.The five prose sites that reference this obligation cite this issue and are enforced by the tracking-issue registry (
spec/tracking-issues.yml), so closing this issue without landing the test will fail CI until the sentences are rewritten — which is the designed behavior.