docs: FEAT-588 record the arriving breath as a gate, not a diminished practice

[MP2EZ]

Closes FEAT-588

The deliverable is a ruling, not a change

FEAT-588 asked whether the Daily Loop's arriving breath presents itself as a practice when its job is to be a gate. philosopher and ux ruled independently and reached the same verdict: no change. The beat is a gate wearing session chrome, and its differences from the standalone practices are intent.

So what merges is the record. Nothing about the app's behaviour changes.

Why the ruling went that way

Five differences are the load-bearing evidence, all verified in source:

1. It starts itself. isBreathActive initialises to showBreath, where all three standalone practices gate on PracticeToggleButton's Begin → Pause → Resume. The absence of a Begin button is the gate — you do not enter a threshold, you are already in it.
2. It carries no name while it runs. The title is an instruction ("Take a moment to arrive"); the principle name appears only once you are through.
3. It ends in the next beat. breathCompleted flips and the same ScrollView re-renders — no completion screen, no quote, no Continue. A practice in this app always ends in a bounded artefact.
4. No session haptics, because the opt-in promises eyes-closed practice and this beat is eyes-open by construction (FEAT-565's recorded ruling).
5. It is not in STANDALONE_PRACTICES — a doorway, not a room.

One premise of the report is inverted

The strongest item on the "borrowed practice vocabulary" list turns out to point the other way. Timer's built-in Pause/Resume is not borrowed — it is orphan vocabulary. All three standalone practices pass showControls={false} ("using custom button below"); the only two live sites rendering Timer's own transport are this beat and DailyLoopCompleteScreen's coda. The practices refused this control. The beat is not a stripped-down practice; it is the raw Timer default, and the loop's two breaths share one internally consistent idiom.

Deliberately not changed

All considered, none owed by the ruling: dropping showControls (a real improvement both lenses liked, but it needs an accessibility sign-off on WCAG 2.2.2 — auto-starting motion over 5s — and it intersects DEBUG-468's ruling that Pause is this beat's stay-with-it affordance); showProgress; and 30000 → 32000 ms for four clean cycles, which is a refinement rather than a defect, would invalidate three shipped rationales written around 3.75 cycles, and would leave the coda's 15000 ms (1.875 cycles) inconsistent unless carried too.

Where the record lives, and why the style matters

The in-code half is a * -prefixed block in DailyLoopStepScreen's module header, plus a // anchor beside breathCompleted. That is deliberate and it is the difference between this PR costing nothing and costing a simulator gate.

practices/dailyloop/ is a Protected Path, so Phase 2.5's inert filter is what decides. It strips //, /**, * and */ — but not a JSX {/* … */} block, whose opener is {/* and whose body lines carry no *. FEAT-565's own record-only commit (7bba8f4c, +19/−0, pure prose) is written that way and left 17 surviving lines, so it paid a full gate run for a comment. Replaying the same filter over this diff gives 0 surviving lines in the .tsx; the docs bullet is outside app/, so it never enters the candidate set at all.

Verified, not assumed — the filter was run against both commits.

The crisis pass changed the record

crisis is required for any dailyloop/ diff, comment-only included — the inert filter exempts the Maestro half, never the agent half. It returned NEEDS_EDIT and all three corrections are applied:

• The record said "no session haptics". That is a term of art in this codebase naming only the sessionStart/sessionEnd anchors, but FEAT-565 declined the whole hook, both channels — so as written a reader could take scheduled breath cues to be open. Crisis-adjacent, not stylistic: wiring any haptic channel onto this beat would import DEBUG-587's recorded-and-unfixed 1–3 frame blur residual onto a surface that currently has none.
• The case against 30000 → 32000 ms listed only content and consistency rationales, so a reader who satisfied all of them would reasonably conclude the constant was theirs to move. BREATH_DURATION_MS is load-bearing in a crisis ruling: tenseMode.ts re-hosted quick depth's SUPPORT_LINE onto Sphere Sovereignty precisely because Aware Presence's reflection phase sits behind this 30s gate, and crisis review rejected the alternative as making quick's crisis affordance strictly less available than deep's. The header now says so.
• The docs bullet took the same correction so the two records cannot drift.

Both would have propagated silently — through a record written specifically to be quoted back at future readers. Confirmed independently: Aware Presence carries SUPPORT_LINE in no (depth, mode) combination, and the root FAB is live throughout the breath (DailyLoop is in IMMERSIVE_ROUTES, not SUPPRESSED_ROUTES), so the gate delays no 988 affordance.

philosopher ruled during planning.

Checks

Typecheck, lint baseline, 179 dailyloop tests, test:accessibility 456, test:integration 162 — all green. No executable line changed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CkZagJp7dzojvrjY7uYwkh