It is very strange. There seems to be something unique to the rate coming from the wire chamber that causes the TDC to emit an otherwise valid "empty" set of data (ie. header+trailer word) without raising any kind of error. We know there should be at least one reference channel arriving on the front during that event -- it's just silently dropped.
This problem goes away if the input signal come from a (fanned out) pulser. With a pulser-driven input _no_ data are dropped, even under very high input rates and using an uncorrelated trigger.
We tried adding a long delay (200+ microseconds) immediately after
entering the interrupt handler just in case the the TDCs needed to
"settle" after receiving a trigger. No luck.