Composite scoring involves judgment — severity anchors, persistence, reach, orientation. GII manages it partly through blinded reviewer scoring and inter-rater analysis: reviewer agreement is measured with named statistics, not an undefined standard. Components that do not yet agree are flagged for anchor revision — no target score distribution is imposed. The pilot advances beyond beta only when agreement is acceptable (one of the pilot success criteria).
Computed live by @gii/calibration on an illustrative blinded rating set. GII-US is pre-pilot; these are not ratings of any real decision.
Agreement by scorecard component
Values are on the statistic's own 0–1 scale, so a bar's length is directly comparable across statistics. Each is graded against a pilot-learning floor and a higher production floor.
4.0626999999999995 components in total · Component detail
The named statistics#
| Statistic | Applies to |
|---|---|
| Krippendorff's α (interval) | E, R, P, S magnitude components and I, Q, F, B, T confidence components |
| Weighted κ / ordinal α | evidence-level and maturity assignment |
| ICC(2,k) | composite magnitude |
Calibration results#
Each component’s agreement, graded against the provisional thresholds a pilot-learning floor and a higher production floor:
| Component | Statistic | Agreement | Level |
|---|---|---|---|
| Magnitude — severity (E) | krippendorff-interval | 0.9823 | production |
| Magnitude — reach (R) | krippendorff-interval | 0.775 | pilot |
| Confidence — identification (I) | krippendorff-interval | 0.4518 | below pilot |
| Evidence level (A–F ladder) | weighted-kappa | 0.8654 | production |
| Composite magnitude (GII-M) | icc-2k | 0.9882 | production |
The estimators are standard (Krippendorff’s α, Cohen’s weighted κ, ICC(2,k)) and tested against documented reference datasets; the thresholds are provisional, calibrated in the pilot. Compute on your own ratings via POST /api/v1/calibration or gii calibrate <ratings.json>.