fix: the hole reservation now runs after the accumulator re-plan, so value_acc_ is always sized to channels.size() before reserve_hole() writes into it.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/acquisition.cpp

@@ -518,6 +518,19 @@ void Acquisition::on_data(const udps_frame_t* frame) {
   BufferStore& store = model_.live();
   const u64 g = part.time_stride;
 
+  // The filter's groups are `g` long, so a re-plan has to start a fresh group rather than
+  // finish the one it was in with the wrong length.
+  if (acc_stride_ != g || acc_channels_ != channels.size()) {
+    acc_stride_ = g;
+    acc_channels_ = channels.size();
+    value_acc_.assign(channels.size(), 0.0);
+    value_acc_n_.assign(channels.size(), 0);
+    math_sample_.assign(channels.size(), std::nanf(""));
+    time_acc_ = 0.0;
+    abs_time_acc_ = 0.0;
+    time_acc_n_ = 0;
+  }
+
   // The gap the counter reports, reserved as NaN before this frame's own samples are
   // stamped: the hole belongs *ahead* of the data that revealed it. Running the missing
   // samples through the same per-sample loop below is the whole trick -- `tick_` advances
@@ -528,6 +541,11 @@ void Acquisition::on_data(const udps_frame_t* frame) {
   // Capped at `kGlitchPackets`: 256 packet spans in one go is an outage, not loss, and
   // reserving one would push the entire retained history out of the ring to make room for
   // samples that are not coming.
+  //
+  // This block must follow the re-plan above: reserve_hole() runs the missing samples
+  // through the same decimation accumulators (value_acc_, value_acc_n_, math_sample_),
+  // which must already be sized to the current channel count and stride before any
+  // per-sample write.
   const std::uint32_t missing = frame->late ? 0u : frame->lost;
   if (missing > 0 && batch_ > 0 && double(missing) <= kGlitchPackets) {
     reserve_hole(missing, part, channels, store, g);
@@ -542,19 +560,6 @@ void Acquisition::on_data(const udps_frame_t* frame) {
     std::fflush(stderr);
   }
 
-  // The filter's groups are `g` long, so a re-plan has to start a fresh group rather than
-  // finish the one it was in with the wrong length.
-  if (acc_stride_ != g || acc_channels_ != channels.size()) {
-    acc_stride_ = g;
-    acc_channels_ = channels.size();
-    value_acc_.assign(channels.size(), 0.0);
-    value_acc_n_.assign(channels.size(), 0);
-    math_sample_.assign(channels.size(), std::nanf(""));
-    time_acc_ = 0.0;
-    abs_time_acc_ = 0.0;
-    time_acc_n_ = 0;
-  }
-
   // Calibration is read per frame; it only changes under the same lock.
   TriggerEngine& engine = engine_;
   const int trig_channel = tc.channel;