What changed
§15.6 is now unconditional and binds every track. The old trigger — which fired only when an idea's output was "primarily a judgment rather than a mechanically decidable artifact" — switched the protocol's only forced-alternative rule off for exactly the class of task it was needed for. It is gone.
§15.6 1,372 B -> 1,015 B net -357 B
Round 1 now carries ## Existing alternatives, and the runtime enforces it. Enumerate the mechanisms your proposal builds by hand — components named, not described — and for each, name what the toolchain, stdlib, dependencies or platform already ships, with a locator. A scoped null result is legal and must name the sources consulted; "the hand-built route is correct" is a valid outcome.
The form is not stylistic. An open-ended "consider alternatives" instruction is measured not to work: instructing people to produce ideas very different from the examples does not reduce conformity, while instructing them to conform does increase it (Smith, Ward & Schumacher 1993). Enumeration is the form with positive evidence (Chrysikou & Weisberg 2005; George & Wiley 2020).
Design rounds had no existence-or-shape validator at all — unlike review rounds. A participant CLI could exit 0 having written nothing. From this release, runner.ValidateRoundOneArtifact gates the section via protocol.HasNonEmptySection, so a bare heading or a prose mention does not satisfy it.
§15.7's per-track table no longer exempts fast from a section the prose calls unconditional.
What is deliberately NOT here
The ratified package had three legs. One shipped. The evidence exchange and the disposition scanner are neither implemented nor printed in the protocol, because printing a rule nothing enforces is the defect this release exists to fix. They are protocol-evidence-exchange and protocol-disposition-scanner.
Provenance
Idea protocol-generation-bias: 3 deliberation rounds, 6 participants, review consensus 3 ✅ / 2 🟡, no blocks. All 24 external citations independently verified by agents that did not write the citing file — 20 survived, 4 did not.
Three reviewers independently found that the first cut of this gate was dead code: the validator had zero non-test callers. Wiring it to the runtime path broke five pre-existing tests whose fixtures lacked the section, which is the reachability proof.