Skip to main content

​
Predictive Stall Detection

​
Overview

Predictive Stall Detection watches for early warning signs that you might be losing momentum - before you actually stall. The system analyzes five engagement signals every four hours and alerts your sherpa when it detects patterns that historically lead to founders going quiet. This gives your sherpa time to reach out with support before you hit a full 21-day stall, keeping your venture moving forward.

​
Step-by-Step Guide

For Founders:
  1. No action required on your part - Predictive Stall Detection runs automatically in the background while you work on your venture.
  2. If your engagement drops, the system calculates a stall probability score by analyzing:
    • Your recent message activity with your sherpa
    • Your momentum trend over the last 7 days
    • Overdue action items from meetings
    • Missed standup responses
    • Meeting attendance patterns
  3. When risk crosses a threshold, your sherpa receives a proactive alert through their preferred channel (Slack, email, or WhatsApp) with specific context about what signals triggered the warning.
  4. Your sherpa reaches out with a timely check-in before you fully disengage. The alert includes specific talking points - like which action items are overdue or how many standups you’ve missed - so the conversation is productive.
  5. As you re-engage, the system tracks your recovery. When your activity returns to normal, the stall risk drops back to LOW and your sherpa sees the prediction resolved.
For Sherpas:
  1. Access your Sherpa Dashboard at /sherpacombineddashboard to see all founders you’re supporting.
  2. View stall risk indicators in the dashboard. Founders with MEDIUM or HIGH stall probability appear highlighted with their risk level and top contributing signals.
  3. Click into a prediction to see the full breakdown: momentum history chart, overdue action items list, standup participation record, and AI-generated recommended actions.
  4. Take action based on the recommendation. The system suggests specific interventions like “Follow up on 3 overdue action items - the oldest is 8 days past deadline” or “Schedule a brief check-in to remove blockers.”
  5. Document your intervention through normal sherpa channels. As the founder responds, the system automatically updates the stall probability in the next 4-hour scan cycle.
For Program Managers:
  1. Check the PM Dashboard at /pipelinedashboard each morning to see HIGH-risk predictions across your portfolio.
  2. Use the stall predictions API GET /api/v1/stall-prediction?riskLevel=MEDIUM,HIGH to pull a prioritized list of at-risk founders sorted by probability score.
  3. Review signal breakdowns for each prediction to understand whether the risk is communication-based, momentum-based, or task-completion-based.
  4. Coordinate with sherpas on HIGH-risk cases. The system automatically notifies you when a prediction escalates from MEDIUM to HIGH so you can provide backup support.
  5. Track resolution progress through the same dashboard. Predictions move to RESOLVED status when the founder’s engagement returns to healthy levels.

​
Common Questions

Q: How does the system know I’m about to stall?
The system combines five signals into a composite score: your message frequency compared to your normal pattern, your momentum trajectory over the last week, how many action items are overdue, standup participation, and meeting attendance. When this composite score crosses 0.30 (MEDIUM risk) or 0.60 (HIGH risk), it triggers an alert to your sherpa. These thresholds vary slightly by ICP category because different founder stages have different normal engagement patterns. Q: Will I get notified that I’m flagged as “at risk”?
No - predictions go directly to your sherpa, not to you. This is intentional. The goal is for your sherpa to reach out with a natural, supportive check-in rather than making you feel like you’re being monitored. You’ll simply notice your sherpa being proactive about staying connected. Q: What happens if I’m just busy for a few days - does that trigger a false alarm?
The system looks for patterns, not isolated events. Missing one standup or going quiet for two days won’t trigger an alert. The algorithms require multiple signals converging - like declining momentum plus overdue items plus missed standups - before crossing alert thresholds. A brief busy period without other warning signs stays in the LOW risk zone. Q: Can my sherpa see my stall probability score?
Yes - sherpas see the exact probability score (0.0 to 1.0) and the breakdown of all five contributing signals with their individual scores. This transparency helps your sherpa understand whether to focus the check-in on task completion, communication patterns, or momentum recovery. Q: How is this different from the existing stall detection system?
The existing system (Feature 069) waits 7, 14, or 21 days of complete inactivity before firing alerts - by day 21, you’ve been silent for three weeks. Predictive Stall Detection analyzes engagement patterns and warns your sherpa days or weeks earlier, while there’s still strong momentum to preserve. Both systems run in parallel for now.

​
Troubleshooting

Issue: I’m getting too many check-ins from my sherpa
This happens if your normal engagement pattern is highly variable - the system interprets dips as risk signals. Your sherpa can adjust your cadence settings in /rhythm/:jobId/cadence to better match your natural work rhythm. The prediction thresholds can also be tuned per ICP category by your Program Manager through PATCH /api/v1/stall-prediction/:icpCategory. Issue: My sherpa didn’t reach out even though I haven’t responded in days
If you have no assigned sherpa, MEDIUM alerts are suppressed (there’s no one to notify). HIGH alerts still create PM signals for portfolio visibility, but direct intervention requires sherpa assignment. Check your Sherpa Dashboard at /sherpacombineddashboard to confirm your journey has an active sherpa. Issue: The prediction says I have overdue action items but I completed them
The system reads from Meeting Action Items (Feature 072). Make sure you’ve marked items as COMPLETED in the meetings interface at /meeting/:meetingId/action-items. If an item shows PENDING or IN_PROGRESS, it counts as overdue once past its due date. The prediction updates within 4 hours of you completing tasks. Issue: My momentum dropped but I’m still actively working
Momentum (Feature 046) tracks visible progress signals like completing tasks, advancing stages, and generating artifacts. If you’re doing deep work that doesn’t create visible outputs yet, momentum may dip temporarily. Your sherpa understands this - the check-in is an opportunity to log what you’re working on so momentum recalculates accurately. Feature 069: Reactive Stall Detection - The existing system that triggers escalating alerts at 7, 14, and 21 days of complete inactivity. Predictive Stall Detection warns earlier, but the reactive system continues running as a backstop. Both appear in your sherpa’s dashboard, and both create PM signals for portfolio oversight. Feature 073: Standup Cadences - Configure your standup rhythm and response expectations at /rhythm/:jobId/cadence. Your standup participation rate feeds into stall predictions, so keeping your cadence aligned with your actual work pattern improves prediction accuracy and reduces false alerts. Feature 072: Meeting Action Items - Track commitments from planning sessions and check-ins. Overdue action items are a strong stall predictor, so staying on top of your action item list (visible at /meeting/action-items/mine) directly reduces your stall probability score and keeps your sherpa confident in your progress.