Calibration Dashboard

Entity-centric investigations with graph exploration

CONFIDENCE IS NOT TRUTH

Models don't fail when they're wrong. They fail when they're not corrected.

BRIER SCORE
0.138
lower is better
ECE
0.075
bucket avg gap
ACCURACY
77%
48 outcomes
DRIFT
0.082
within band
45% BRIER-CALIBRATED

DRIFT DETECTED

SEED FALLBACK Β· no live outcomes yet DRIFT Β· 0.082

Reliability Diagram

5 buckets Β· 48 outcomes

The dashed line is perfect calibration. Points above the line are under-confident; points below are over-confident. The doctrine treats a curve that sits below the line as a louder warning than accuracy alone.

Brier Score β€” Rolling Window

window: 6 Β· 48 points

Lower is better. 0.0 is perfect, 0.25 is the coin-flip baseline. A rising trend means the model is drifting and needs recalibration.

Recent Decided Cases

  • Acme Holdings 0001

    Acme Holdings s.r.o.

    proceed_with_conditions incorrect
  • Acme Holdings 0002

    Acme Holdings s.r.o.

    proceed correct
  • Acme Holdings 0003

    Acme Holdings s.r.o.

    proceed correct
  • Acme Holdings 0004

    Acme Holdings s.r.o.

    proceed correct
  • Acme Holdings 0005

    Acme Holdings s.r.o.

    proceed correct
  • Acme Holdings 0006

    Acme Holdings s.r.o.

    proceed_with_conditions incorrect
  • Acme Holdings 0007

    Acme Holdings s.r.o.

    proceed correct
  • Acme Holdings 0008

    Acme Holdings s.r.o.

    proceed_with_conditions correct
  • Acme Holdings 0009

    Acme Holdings s.r.o.

    proceed incorrect
  • Acme Holdings 0010

    Acme Holdings s.r.o.

    proceed_with_conditions correct

PRISMATIC

Reality-Enforced Decision Infrastructure