Calibration
DD6 improves through real-world usage.
Teams should calibrate scoring based on observed discovery outcomes.
What to track
Useful feedback signals include:
- Number of discovery sessions actually needed vs predicted
- Spec quality after discovery (measured by CIRK accuracy or agent success rate)
- Rework caused by insufficient discovery
- Cases where discovery was skipped but should not have been
- Cases where full discovery was run but was unnecessary
Calibration goal
The goal is not perfect prediction.
The goal is better alignment between estimated discovery depth and observed need.
"Proportionality matters more than precision."
A model that consistently avoids both over-discovery and under-discovery is more useful than one that perfectly predicts session count.
What calibration can improve
With enough usage data, teams can:
- Refine what qualifies as I1/I2/I3 for their domain
- Adjust depth thresholds for specific intake types
- Identify which dimensions are most predictive for their projects
- Improve template matching accuracy
- Compare estimated vs observed discovery depth
Long-term direction
Over time, DD6 can support:
- AI-assisted scoring (the model classifies intakes automatically)
- Per-project dimension weighting
- Predictive depth recommendations
- Empirical discovery investment optimization