Timers 1.4.2: ProPresenter digits no longer flick back a second

ProPresenter reports running timers as truncated whole seconds, and the
poller re-anchored the tiles' local count at the bottom of that second on
every 500 ms poll — so the displayed digit ran up to a second ahead and
got yanked back a step whenever a fresh poll sampled the next flip later
than the last one did. Tiles flickered ...12:33 → 12:34 → 12:33... about
once a second.

phase.js turns the quantized reports into one stable anchor per timer:
the only information in a poll is a FLIP (the report stepping one second
in the direction of travel), which dates the true boundary to the
midpoint of the two polls that straddle it. While reports stay consistent
with the anchor (±450 ms) the anchor holds, so tiles count along one
steady clock; a report the anchor can't explain (reset, seek, resume,
accumulated drift) re-anchors. Timers carry the anchor as a per-timer
asOf the client extrapolates from (optional in the provider contract,
documented in the guide); the client additionally refuses to tick a
running card backwards over sub-1.5 s jitter.

phase.selftest.js (66 checks) drives a simulated ProPresenter — truncated
reports, jittered polls — through the tracker and the client's render
math: digits never step backwards, dwells hold 1 s (the ~2 s zero-crossing
dwell is trunc's honest behavior), anchors stay within tolerance of the
true value, resets snap, pause/resume re-locks, 0.2% clock drift is
absorbed. Verified live against tools/pp-timers-mock.js: 78 s of 100 ms
DOM sampling across four running cards — every step ±1 in the card's own
direction, dwells 998–1002 ms, zero regressions; upstream kill/restart
still recovers.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

modules/propresenter-timers/phase.js

@@ -0,0 +1,113 @@
+'use strict';
+
+/**
+ * Phase lock for whole-second timer reports (the ProPresenter poller).
+ *
+ * ProPresenter reports a running timer as whole seconds — trunc(value):
+ * "12:34" means the true remaining time is anywhere in [754 s, 755 s). The
+ * old poller re-anchored a tile's local count at the bottom of that interval
+ * on every poll, so the displayed digit ran up to a second ahead and then
+ * got yanked back a step whenever a fresh poll happened to sample the next
+ * flip later than the last one did — the poll clock and ProPresenter's
+ * second boundaries drift through each other, so tiles flickered
+ * "…12:33 → 12:34 → 12:33…" about once a second.
+ *
+ * This tracker turns the quantized reports into ONE stable anchor
+ * { valueMs, at } — the continuous value at a fixed instant — which the
+ * client extrapolates from. Rules:
+ *
+ * - The only real information in a poll is a FLIP: the report stepping one
+ *   second in the direction of travel. At the flip instant the true value
+ *   equals the interval boundary just crossed, and the flip happened between
+ *   the last two polls, so their midpoint dates it to within half a poll.
+ * - While a poll's report matches what the anchor predicts (within TOL_MS
+ *   around the reported second's interval), the anchor HOLDS — the same
+ *   numbers are republished, so consumers count along one steady clock
+ *   instead of a freshly re-phased one per sample.
+ * - A report the anchor can't explain re-anchors: at the flip boundary when
+ *   this poll saw a clean one-second flip, else mid-interval (first sight,
+ *   an operator reset/seek, a resume, or clock drift past TOL_MS).
+ * - Before the first flip the anchor sits mid-interval: at most ±½ s off,
+ *   and the first flip (≤1 s away) locks it — instead of the old guaranteed
+ *   whole-second bias one way.
+ *
+ * trunc() semantics also cover overrun: a countdown at −12.4 s reports
+ * "-12", so a negative countdown's interval is (W−1 s, W] and its flip
+ * boundary is W itself, where a positive countdown's are [W, W+1 s) and
+ * W+1 s. Around zero ("00:00:00" covers (−1 s, 1 s)) the flip in from above
+ * crosses +1 s.
+ */
+
+/** How far a report may sit outside the anchor's prediction before it means
+    something real. A flip is dated to the midpoint of two polls, so its own
+    error bound is ±½ poll (~±300 ms at the 500 ms cadence, more when a poll
+    runs long); the tolerance must sit ABOVE that or consecutive flips
+    re-anchor each other forever, and safely below a full second so a real
+    one-second seek still registers. */
+const TOL_MS = 450;
+
+/** The continuous-value interval implied by a whole-second report. */
+function intervalFor(kind, wholeMs) {
+  if (kind === 'countdown') {
+    if (wholeMs === 0) return { lo: -1000, hi: 1000 };
+    if (wholeMs < 0) return { lo: wholeMs - 1000, hi: wholeMs };
+  }
+  return { lo: wholeMs, hi: wholeMs + 1000 };
+}
+
+/** The continuous value at the instant a running timer's report flips TO wholeMs. */
+function boundaryFor(kind, wholeMs) {
+  if (kind === 'countdown') return wholeMs <= 0 ? intervalFor(kind, wholeMs).hi : wholeMs + 1000;
+  return wholeMs;
+}
+
+/**
+ * One tracker per timer. Feed one observation per poll; get back what to
+ * publish: `{ valueMs, asOf }` — a continuous value at a fixed instant,
+ * stable across polls while the reports stay consistent with it. Not
+ * running (or no value): the report is returned verbatim, dated at the poll.
+ */
+function createTimerPhase() {
+  let anchor = null; // { valueMs, at }
+  let kind = '';
+  let lastWhole = null;
+  let lastAt = 0;
+  let wasRunning = false;
+
+  function observe({ kind: rawKind, running, wholeMs, pollAt }) {
+    const k = rawKind === 'countdown' ? 'countdown' : 'elapsed';
+    const dir = k === 'countdown' ? -1 : 1;
+    const flip = Boolean(running) && wasRunning && k === kind && lastAt > 0
+      && lastWhole !== null && wholeMs !== null
+      && wholeMs - lastWhole === dir * 1000;
+    const prevAt = lastAt;
+    lastWhole = wholeMs;
+    lastAt = pollAt;
+    wasRunning = Boolean(running);
+    if (k !== kind) anchor = null;
+    kind = k;
+    if (!running || wholeMs === null) {
+      anchor = null;
+      return { valueMs: wholeMs, asOf: pollAt };
+    }
+    const { lo, hi } = intervalFor(k, wholeMs);
+    // A mid-interval anchor is provisional — a guess up to ±½ s off that the
+    // tolerance check may fail to catch for a long time (its error only
+    // shows at a poll landing more than TOL_MS past a flip). The first
+    // clean flip is strictly better evidence (±½ poll), so it always wins.
+    if (anchor && !(anchor.provisional && flip)) {
+      const predicted = anchor.valueMs + dir * (pollAt - anchor.at);
+      if (predicted >= lo - TOL_MS && predicted <= hi + TOL_MS) {
+        return { valueMs: anchor.valueMs, asOf: anchor.at };
+      }
+    }
+    anchor = flip
+      ? { valueMs: boundaryFor(k, wholeMs), at: (prevAt + pollAt) / 2, provisional: false }
+      : { valueMs: (lo + hi) / 2, at: pollAt, provisional: true };
+    return { valueMs: anchor.valueMs, asOf: anchor.at };
+  }
+
+  return { observe };
+}
+
+module.exports = { createTimerPhase, TOL_MS };