Ω-W1-T3: run the master meter on its own 60 FPS timer, and make a meter frame cost one rect

src/shell/instrument/CLAUDE.md

@@ -260,21 +260,31 @@ against a performance budget — they are there because `VoiceEngine::applyLiveT
   detector's 4-sample group delay INSIDE that budget, not on top), bounded by one 500 ms tick.
   Narrowing it further means a second deferral mechanism (a posted window message) rather than
   the tick — deliberately not built.
-- **The MASTER meter's ballistics ride the sync tick, and that tick is 500 ms.** They run
-  BEFORE the tick's in-flight-drag guard on purpose — a drag suppresses the reload poll, but
-  the bus keeps sounding. Elapsed time is measured (`GetTickCount64`), never assumed from the
-  timer's period, and the tick repaints only when `meterDrawEqual` says the picture changed.
-  **The published block state is therefore ACCUMULATED, not sampled**: at 48 kHz / 512 frames
-  ~47 blocks elapse per tick, so the processor folds a per-channel max and a min limiter gain
-  across them and `masterBusMeter()` clears the accumulators as it reads. A plain overwriting
-  store displayed one block in ~47 and lost the rest — the specified "a peak displays on the
-  first UI frame after it occurs" is what the fold restores. `masterBusMeter()` is CONSUMING,
-  so exactly one caller may hold it; the embed strip reads its own non-consuming
+- **The MASTER meter has its OWN 16 ms timer, beside the 500 ms sync poll.** It runs outside
+  that poll's in-flight-drag guard on purpose — a drag suppresses the reload poll, but the bus
+  keeps sounding. Elapsed time is measured (`steady_clock`, whose resolution the rate needs —
+  `GetTickCount64`'s ~15.6 ms quantized a frame's delta to 0 or 16), never assumed from the
+  timer's period, a zero delta skips the advance, and the tick repaints only when
+  `meterDrawEqual` says the picture changed. **The published block state is ACCUMULATED, not
+  sampled**: the processor folds a per-channel max and a min limiter gain across every block
+  since the last read and `masterBusMeter()` clears the accumulators as it reads. A plain
+  overwriting store displayed one block per window and lost the rest — the specified "a peak
+  displays on the first UI frame after it occurs" is what the fold restores. `masterBusMeter()`
+  is CONSUMING, so exactly one caller may hold it — which is why the meter block MOVED to the
+  meter tick rather than being copied there; the embed strip reads its own non-consuming
   `embedActivityLevel()`. **The tick's FIRST read is discarded**, because that caller is the
   only consumer: with no editor open the accumulators hold everything since the instance was
   created, and advancing off them would open the meter at the session's loudest peak. The clip
-  latch is not discarded with them — it is a latch the user clears. A meter-rate timer remains a
-  separate change and is not in.
+  latch is not discarded with them — it is a latch the user clears.
+- **A meter frame repaints the bar FIELD, not the client area**, which is what makes the rate
+  affordable: `paint` honours `ps.rcPaint` against the field rect `paintDeck` cached, composes
+  into a RETAINED back buffer (so the rest of the face is still there from the last full
+  compose) and blits the dirty region alone. The fast path is chosen GEOMETRICALLY, never by a
+  flag — Windows unions a meter invalidate with any other pending one into a single rcPaint, so
+  a flag would paint a meter frame over a face that had really changed. What stays outside the
+  field stays on the full path: the static numeral gutter (AA text re-blended onto itself every
+  frame thickens) and the GR lamp (drawn straight onto the deck group's gradient, which a
+  sub-rect fill cannot reproduce), so a lamp transition takes a whole-client repaint.
 - The bake's availability probe runs on the SAME tick that paints the button, so the
   control can never be enabled on one tick and refuse on the next. The bake Hold control's
   applicability (`resolveBakeHoldNeeded`) rides the same tick for the same reason, and