Does the model improve?
Compare a defined cohort against a reference and ground truth.
CURRENT PUBLIC LEVEL
EVIDENCE
Results need
scope, version and limits.
THOR 42 LAB separates measured evidence from operational claims and from economic value. This page states the current public proof level we are prepared to stand behind.
THOR FLIGHTS / CURRENT PUBLIC EVIDENCE
P2 · Validated cohort.
Predictive evidence within the stated validation scope.
On the current 95-flight forward cohort, the reported mean absolute error is 7.98 minutes for THOR and 14.95 minutes for the reference. THOR performed better on 65.3% of flights in that cohort, over a 0-2h arrival horizon.
What this establishes. Predictive performance on the stated cohort and comparator.
What this does not establish. Guaranteed live-service outcomes, customer workflow utility, financial ROI, safety-of-flight suitability or independent third-party certification.
7.98 minTHOR · MAE
14.95 minREFERENCE · MAE
65.3%FLIGHTS THOR OUTPERFORMED REFERENCE
95FORWARD FLIGHTS
0-2hVALIDATION HORIZON
EVIDENCE LADDER
Prediction is one gate.
Value is another.
Does the signal change work?
Measure whether operators watch, prioritise or act differently in controlled customer conditions.
TO BE VALIDATED
Does that change create value?
Measure time, cost or service outcomes against an agreed customer baseline.
TO BE VALIDATED
PROOF DISCIPLINE
Evidence before adjectives.
The public website does not publish proprietary model parameters or supplier-sensitive implementation details. It publishes the performance claim, its scope and the boundary of what remains unproven.
Defined cohort
Know which observations are included.
Defined comparator
Know what THOR is being measured against.
Versioned run
Keep model and data versions attributable.
Bounded claim
Do not promote P2 into P3 or P4 without evidence.