Define or learn the baseline
Use expected-signal limits or let the monitor learn a stable baseline before it starts judging excursions.
Long-duration evidence
Catch the PCB fault that disappears when you look for it
YProbe monitor mode watches a rail or signal over time, records one timestamped event per excursion and carries the representative trace into the same diagnostic record as your deliberate bench measurements.
A single capture can prove that a fault exists now. It cannot prove that a rail was healthy during the five minutes before the reset. Intermittent diagnosis needs time, stable contact and a record that distinguishes one long excursion from hundreds of identical frames.
From continuous captures to useful events
Watch for a real crossing
Droop, overshoot, dropout and glitch conditions are evaluated on fresh captures while the instrument remains free-running.
Debounce the excursion
One event records onset, worst value and recovery instead of creating a new row for every capture during the same failure.
Carry it into diagnosis
The event summary and representative trace join the session evidence so later hypotheses can explain what happened over time.
Events the monitor is built to preserve
Brownout or droop
A driven rail falls below its expected level under load, during radio activity or when another subsystem wakes.
Dropout and recovery
A supply collapses for milliseconds or seconds, then returns before a manual capture can be armed.
Glitch
A narrow excursion or missing cycles appear rarely enough that a normal single frame has little chance of containing them.
Thermal or time-dependent drift
Mean level or ripple changes gradually as the board warms, a load pattern changes or a marginal component degrades.
Why probe-contact detection matters more over time
A loose or lifted probe can look exactly like an intermittent board failure. YProbe classifies individual captures and uses cross-capture agreement before accepting a contact state. At the blocking confidence threshold, floating or intermittent contact cannot produce a PASS/FAIL, enter the session record or reach the AI as board evidence.
The classifier also avoids a common mistake: a noisy driven rail is not automatically a floating probe. Cross-capture mean wander is the decisive signal, and a floating input must remain physically plausible near ground rather than being inferred from noise alone.
- A refused capture still displays its trace so the technician can see what the instrument acquired.
- The stability tracker votes across captures instead of averaging them into a waveform that never existed.
- Contact metadata accompanies accepted evidence into later diagnostic prompts.
- The built-in simulator includes periodic dropout, brownout, thermal drift and intermittent glitch scenarios.
An example event record
14:02:11MONITOR+3V3_PHY · baseline learned 3.31 V (8 stable captures)
14:05:43DROOP+3V3_PHY · −310 mV for 1.8 s · worst 2.99 V
14:05:45RECOVERY+3V3_PHY · back within limits after 1.8 s
14:11:02DROPOUT+3V3_PHY · rail collapsed to 0.4 V for 220 ms
14:11:03RECOVERY+3V3_PHY · restored · event added to session evidence