Definition
Degree to which the fuel optimization model's recommended fuel load covers tail-risk scenarios (95th and 99th percentile fuel burn events) at rates consistent with the model's stated confidence. FTRCI measures not whether the average flight is safe but whether the model is calibrated for the flights where fuel becomes critical.
The Measurement Gap
The measurement gap FTRCI addresses: legacy fuel operations management tracks aggregate fuel consumption, fuel on arrival statistics, reserve infringement events, and diversion rates, but none of these metrics can see whether an AI fuel model's stated confidence is calibrated for tail-risk scenarios. A fuel model that correctly predicts mean fuel burn for 95% of flights can still underestimate fuel requirements for the 5% of flights experiencing abnormal fuel burn, extended holding, or late ATC rerouting. Those are exactly the flights where a calibration failure becomes a safety event.
Traditional fuel oversight asks: "Did we have enough fuel?" FTRCI asks: "Did the model's 99% confidence claim actually mean 99%?" These are different questions. An airline can have zero reserve infringements in a quarter while running a fuel model that is systematically overconfident at the tail, because PICs exercised discretion and added fuel above the model recommendation. FTRCI measures the model's recommendation, not the actual fuel loaded.
The decision FTRCI supports: should the fuel optimization model's minimum fuel recommendations be increased for tail-risk scenarios? Should the model's stated confidence bands be recalibrated? This question cannot be answered without separating model output from PIC discretionary additions and comparing stated model confidence to actual sufficiency rates at each confidence band.
Formula
Plain language: Group flights by the fuel model's stated sufficiency confidence band (e.g., "99% confident fuel is sufficient," "95% confident"). Within each confidence band, measure the actual rate at which the recommended fuel load would have been insufficient: defined as final reserve infringement or diversion required given actual fuel burn. Compare stated confidence to actual sufficiency rate.
Notation:
FTRCI(b) = Actual_sufficiency_rate(b) / Stated_confidence(b)
Variables:
- b = confidence band (e.g., 99%, 95%, 90%)
- Actual_sufficiency_rate(b) = percentage of flights in band b where the ML-recommended fuel load would have been sufficient given actual fuel burn: no final reserve infringement (percentage)
- Stated_confidence(b) = midpoint of the model's confidence band (percentage)
Units: Ratio per band. 1.0 = perfectly calibrated at the tail. Below 1.0 = model overstates sufficiency confidence. Track the 99th percentile band specifically. This is where safety-relevant tail risk concentrates.
Critical measurement rule: always measure against the model's recommendation, not the actual fuel loaded. The question is whether the model's output is safe, not whether the PIC's discretionary add-on saved it. Flights with PIC fuel additions above model recommendation must be included in the computation using the model's recommended quantity as the measurement basis.
Aggregate across fleet types and full seasonal cycles. The 99% confidence band requires large sample sizes to validate, a rolling 12-month window is typically the minimum for statistical meaning.
Proposed formula, pending validation pilots.