Skip to main content

​
PEM detection

Post-exertional malaise is delayed, so detection uses lag windows rather than same-day reactions.
1

Trigger detection

Spending significantly exceeds pool.
2

Lag window monitoring

System tracks 0-120 hours after trigger.
3

State drop check

Check-in state drops below recent baseline.
4

Confidence gate

Act only with enough evidence to distinguish from normal fluctuation.

​
Attribution window

WindowAttribution confidenceNotes
0-48 hours1.0strongest confidence
48-72 hours0.7covers common delayed onset
72-120 hours0.4catches longer-delay cases
>120 hoursnone automaticuser attribution accepted
Attribution confidence scales how strongly adjustments are applied.

​
User-reported crash attribution

When the system cannot confidently attribute a crash, a quiet link lets the user identify likely triggers from recent completed tasks.
  • Recorded attribution confidence: 0.6
  • Task cost estimate can be adjusted upward in personal library
  • Signal is fed back into future detection
This is optional and only shown in recovery context. It is never forced.

​
PEM response

  • Confidence -0.30
  • Recovery mode enabled
  • Recovery ends when state vector returns near pre-PEM baseline

​
Regime shift detection

A regime shift is detected when 4 of 5 consecutive days show normalized completion ratio below 0.5 or above 1.2. Response:
  • Confidence -0.50
  • Coefficient EWMA switches to fast mode (alpha = 0.35)
  • Stability indicator moves to unstable

​
Multi-day cumulative load

The system evaluates 3-day and 7-day cumulative spending versus cumulative pool to catch crash risk from repeated moderate overexertion.
Risk can emerge from stacked moderate days, not only one extreme day.

​
Persistent under-rating detection

If 14-day mean check-in rating remains below 2.0 without regime shift, show a gentle educational nudge clarifying that 3 means “about my usual.” Cooldown: 30 days.