Calibrate cards from actuals
Run the feedback loop by hand, seed it from history, and see how you compare to other firms.
Found in Card Library → Install Cards → Calibration. Available to Owners and Admins. This is where the feedback loop becomes visible: DarkBird compares every Install Card's estimated hours against the actual hours logged on completed projects and computes a per-card multiplier (1.3x means that card's estimates run 30% low).
- Recalibrate from history — manually re-runs the math against every completed project's actuals. Pure calculation, no AI cost. Cards with too few historical samples are skipped rather than calibrated on thin data.
- Explain [N] cards — an optional AI step that writes a plain-English rationale for each high-variance card (e.g. "Cabling density in walls and ceilings often exceeds template assumptions").
- Import historical data — seed calibration from project data that predates DarkBird via a CSV upload (cardName, estimatedHours, actualHours, optionally quantity and manufacturer).
- How you compare — once enough other organizations have calibrated the same card name, this panel shows your multiplier against the cross-organization median.
How the comparison works, in both directions. Your calibrations contribute to this anonymized benchmark, and other organizations' calibrations are what make yours comparable. What is shared is a card name and a variance multiplier — never your organization's name or id, never pricing, rates, clients, or project data. Nothing is published for a card until at least 3 distinct organizations have contributed a sample for it, and each organization counts once regardless of how many projects it has, so no single organization's numbers can be identified or inferred from this view.
Controlled by the Contribute to Cross-Org Benchmarks toggle in Settings → AI Preferences — a separate setting from Contribute to Global Brain, which governs catalog categorization signals only. On by default. Turning it off excludes your calibration data from future aggregation runs and replaces this panel with a locked notice — you can't see the comparison without being part of the pool that produces it.
Where this shows up elsewhere: calibration multipliers are applied automatically when the AI recommends cards during an interview, and the Estimate tab's confidence badge factors in how many of an estimate's cards have real calibration data behind them.