Skip to main content

​
Predictive Stall Detection

​
1. Overview

Predictive Stall Detection monitors your momentum, engagement, and activity patterns to spot early signs of disengagement - days or even weeks before you actually stall. Instead of waiting until you’ve gone silent for three weeks, your sherpa receives proactive alerts when risk patterns emerge, giving them time to reach out, remove blockers, and keep you moving forward. This system combines communication frequency, momentum trajectory, action item completion, standup participation, and calendar engagement into a single risk score that runs every four hours across all active ventures.

​
2. Step-by-Step Guide

For Founders:
  1. Your risk score runs automatically - Every four hours, the system analyzes your last 14 days of activity and calculates a stall probability score from 0.0 (no risk) to 1.0 (critical risk).
  2. Early intervention when patterns emerge - If your score crosses a threshold (0.3 for MEDIUM risk or 0.6 for HIGH risk), your assigned sherpa receives an alert with specific context about what’s changing: momentum drops, missed standups, overdue action items, or reduced communication.
  3. Sherpa reaches out proactively - Your sherpa uses the alert’s AI-generated recommendations to contact you through your preferred channel (WhatsApp, Slack, email) with a targeted check-in focused on the specific signals that triggered the alert.
  4. Risk resolves when you re-engage - As you respond to standups, complete action items, and momentum recovers, your score drops below the threshold and the prediction status updates to RESOLVED - no further alerts fire for that cycle.
For Sherpas:
  1. Receive early warning alerts - When a founder crosses a MEDIUM (0.3) or HIGH (0.6) threshold, you receive an alert through your preferred channel within 60 seconds, including the founder’s name, job title, probability score, and top contributing signals.
  2. Review signal breakdown - Access the full prediction detail through the Sherpa Dashboard to see all five signal scores, momentum history over the last seven days, overdue action items, and AI-generated recommended actions.
  3. Act on specific recommendations - The alert includes actionable guidance like “Follow up on 3 overdue action items - the oldest is 8 days past deadline. Consider scheduling a brief check-in to remove blockers” rather than generic advice.
  4. Escalation at HIGH risk - When a founder’s score crosses 0.6, both you and the Program Manager receive alerts. The PM sees the prediction in their morning scan and can coordinate intervention if needed.
For Program Managers:
  1. Access portfolio risk dashboard - Navigate to the PM Dashboard and view all active predictions sorted by stall probability, showing which founders across your portfolio need attention.
  2. Filter by risk level - Use GET /api/v1/stall-prediction?riskLevel=MEDIUM,HIGH to see only founders who have crossed intervention thresholds, with each prediction showing founder name, job title, probability score, and top contributing signals.
  3. Review prediction details - Click into any prediction to see the full signal breakdown, momentum history, action item summary, recent activity timeline, and AI-generated recommendations for the assigned sherpa.
  4. Monitor resolution - Track predictions as sherpas intervene. When risk drops below thresholds, predictions update to status RESOLVED and disappear from the active risk list.

​
3. Common Questions

Q: What happens if I have no momentum history because my job just started?
A: The system returns a stall probability of 0.0 (LOW risk) when there’s insufficient data for prediction. You need at least three momentum snapshots over seven days before the system can detect trajectory changes. Until then, you’re considered low risk by default. Q: Will I get duplicate alerts from both the predictive system and the existing stall detection?
A: Yes, but they serve different purposes. Predictive Stall Detection warns your sherpa days or weeks early based on pattern changes. The existing stall worker (Feature 069) still enforces the 7/14/21-day escalation rules. You might receive a predictive MEDIUM alert at day 5, then a reactive CRITICAL alert at day 21 if the stall continues - this is intentional and allows for layered intervention. Q: What if I don’t have an assigned sherpa when my risk score increases?
A: MEDIUM alerts are suppressed when no sherpa is assigned (no recipient to contact). HIGH alerts still create signals visible to your Program Manager in the PM dashboard, ensuring someone with oversight sees the risk even without direct sherpa assignment. Q: How does the system know what actions to recommend to my sherpa?
A: The AI analyzes which signals contributed most to your risk score. If you have three overdue action items and declining momentum, the recommendation prioritizes following up on those specific items. If you’ve missed three standups but everything else is normal, it focuses on standup re-engagement. The system uses the actual data (item counts, days overdue, momentum magnitude) to generate specific, contextual guidance. Q: Can my stall risk be different across multiple ventures?
A: Yes. Each active job is analyzed independently. You could be LOW risk on one venture (strong momentum, attending meetings, completing tasks) and HIGH risk on another (momentum dropping, missed standups, overdue items). The system tracks each venture separately because your engagement patterns differ by project.

​
4. Troubleshooting

Issue: I’m receiving alerts but my sherpa says they aren’t getting them
Solution: Check channel preferences in the Sherpa Dashboard. Alerts respect quiet hours (22:00-07:00 local time) and have a 24-hour cooldown per risk level - if your sherpa was already alerted for MEDIUM risk today, they won’t receive another MEDIUM alert until tomorrow. HIGH alerts create a new notification cycle. Your sherpa should also verify their preferred channel (WhatsApp, Slack, email) is correctly configured. Issue: My risk score hasn’t updated even though I completed action items and responded to standups
Solution: The predictive worker runs every 4 hours, not in real-time. After you complete tasks or respond, the next scan (within 4 hours) will recalculate your score. If your score remains elevated after two scan cycles (8 hours), contact your sherpa - there may be other contributing signals like communication frequency or momentum trajectory that need attention. Issue: I see HIGH risk predictions in the PM dashboard but no sherpa alerts were sent
Solution: This occurs when a founder crosses HIGH threshold (0.6) but has no assigned sherpa. The prediction creates a PM signal for visibility, but MEDIUM-level alerts to sherpas are suppressed without an assignee. HIGH alerts attempt to notify both the sherpa and PM - if the sherpa field is null, only the PM sees it. Assign a sherpa to enable direct founder intervention. Issue: AI-generated recommendations are generic or don’t match the actual risk
Solution: If the AI recommendation fails (timeout after 15 seconds), the system falls back to template-based recommendations that may feel generic. The fallback uses signal data but lacks the nuance of AI analysis. Check your AI tier - recommendations use the EFFICIENT tier (Haiku). If you’re seeing consistent timeouts, your sherpa can still access the raw signal breakdown (momentum history, overdue items, missed standups) in the prediction detail view to craft their own intervention. Feature 069: Stall Detection - The existing reactive stall system that fires at 7, 14, and 21 days of inactivity. Predictive Stall Detection warns you early; Feature 069 provides the backstop escalation if early intervention doesn’t work. Together they create a two-layer safety net. Feature 070: Proactive Nudge Engine - Powers the actual delivery of stall alerts to sherpas through WhatsApp, Slack, email, or other channels. Predictive Stall Detection generates the risk signal; the Proactive Nudge Engine routes it to the right person through their preferred channel at the right time (respecting quiet hours and cooldowns). SupaSherpa Momentum Tracking - The momentum score is one of five signals feeding stall probability. If you want to understand why your momentum is dropping (and thus contributing to elevated stall risk), review your momentum history in the SupaSherpa dashboard. Momentum tracks your progress velocity, clarity of next actions, and engagement frequency - all inputs to the predictive model.