Paper-claim go/no-go: apply the pre-registered criterion to the projection result

Part of #19

Blocked by: #22

## Question

**Paper-claim go/no-go — the destination.** Apply the **pre-registered criterion (T2)** to the late-window projection's prod result (T3): does the â-reliable, same-word-end perceptual projection clear its pre-registered bar against the population null?

- **GO** ⇒ the integration section becomes **context-gated reactivation of the perceptual (not acoustic) code**. Graduates the fog: CLAUDE.md/ADR reconciliation write-up (+ a new ADR documenting the late-projection method), the integration-response figure plan, and characterization of the complementary acoustic-context subset.
- **NO-GO** ⇒ integration section retreats to **negative claims only** (not lexicality / mismatch / surprisal; not extended acoustic-or-perceptual response to the word-initial sound) — map #15's fallback.

Decide strictly against the pre-registered rule (confirmatory); note any exploratory secondary readouts separately without letting them move the headline. Use `/grilling`.

First resolve the blocker in #22, then use `/grilling` to apply the pre-registered T2 criterion to the late-window projection's T3 prod result. Done means recording a strict GO or NO-GO against the population null, separating exploratory readouts, and following the stated integration-section outcome; a GO also requires the CLAUDE.md/ADR reconciliation, a new ADR, the figure plan, and acoustic-context characterization.

Written by the indexing model from the issue text.