Event-Triggered Residual Reporting¶
Report-by-exception protocol — instantiates Predictive Residual Processing
Stay silent while the world matches prediction; speak only when a precision-weighted residual crosses a bar worth someone's attention.
If the receiver already predicts the routine, then routine updates carry no information and shouldn't be sent at all. Event-Triggered Residual Reporting is the report-by-exception protocol: it withholds as-predicted updates and puts a message on the channel only when a residual crosses an information, consequence, or novelty threshold — and the crossing test is precision-weighted, so a deviation counts for more when the signal producing it is reliable. Its lever is the number of transmissions: skip messages entirely rather than shrink them. That precision-weighting is what separates it from naïve thresholding (which treats a noisy blip like a trustworthy one) and its message-skipping is what separates it from delta encoding (which still sends something every cycle, only smaller).
Example¶
A battery-powered vibration sensor sits on a remote industrial pump. Streaming a reading every second would drain the battery in weeks and flood the network with "nominal." Instead the sensor predicts the pump's expected vibration signature and reports only when the residual — the deviation from prediction, weighted by the sensor's confidence in the measurement — crosses a threshold suggesting real change, such as the onset of bearing wear, rather than transient noise. Months of healthy operation produce almost no traffic; the first genuine sign of a fault produces an immediate report. The battery lasts years, and the one message that matters still arrives fast — capacity spent on surprise, not on confirmation.
How it works¶
- Predict the expected value the receiver would otherwise be told.
- Compute the residual and precision-weight it — scale the deviation by the reliability, consequence, or novelty of the signal, so trustworthy small deviations can outrank noisy large ones.
- Test against the threshold — cross it and emit on the channel; otherwise stay silent, where silence means "as predicted."
- Emit only the event — the channel carries threshold-crossing residuals and nothing else.
Tuning parameters¶
- Threshold level and criteria mix — how information, consequence, and novelty combine into the bar to report.
- Precision-weighting rule — how signal reliability scales a residual before the crossing test.
- Deadband width — the zone of "close enough to prediction" that stays silent.
- Heartbeat interval — a periodic "still alive" so that silence is unambiguous.
- Debounce — hysteresis to stop chattering when a residual hovers at the threshold.
When it helps, and when it misleads¶
Its strength is large traffic and energy savings while still delivering what matters fast — capacity is spent only on genuine surprise.
Its central hazard is that silence is ambiguous: no report can mean "all nominal" or "the reporter has died," and events occurring between reports can be aliased away entirely.[1] A mis-set threshold either chatters or goes mute. The classic misuse is raising the threshold to cut noise until real events fall below the bar and vanish without trace. The discipline: add a heartbeat so silence is verifiable, set the threshold against a measured budget rather than by feel, and precision-weight so that reliable small deviations still register instead of being lost among large noisy ones.
How it implements the components¶
residual_propagation_channel— it governs what actually reaches the reporting channel, admitting only threshold-crossing residuals and keeping the rest off the wire.precision_weighting_rule— its crossing test weights each residual by precision/reliability before deciding whether it's worth a message.
It does not apply the low-level per-error pass/block gate inside the signal path (that's Precision-Weighted Error Gate), route an emitted surprise onward to a defined action (that's Surprise-to-Action Bridge), or shrink the payload of each individual message (that's Delta or Differential Encoding).
Related¶
- Instantiates: Predictive Residual Processing — the "send only surprise" policy at the transmission layer.
- Consumes: Confidence Threshold Table — the governed thresholds it reports against; Precision-Weighted Error Gate — the upstream weighting it builds on.
- Sibling mechanisms: Surprise-to-Action Bridge · Delta or Differential Encoding · Confidence Threshold Table · Precision-Weighted Error Gate · Anomaly Detection Model
Notes¶
Without a periodic liveness signal, this protocol cannot distinguish "nominal" from "dead" — a silent reporter and a silent world look identical on the channel. A heartbeat is therefore not optional decoration but the thing that makes silence readable. It pairs naturally with Surprise-to-Action Bridge, which takes the events it emits and turns them into owned action.
References¶
[1] Send-on-delta / report-by-exception (and the deadband it implies) are the standard names for this scheme in control and telemetry; the well-known caveat is that absence of a report is not positive confirmation of health, which is why a heartbeat is added. ↩