Frontier out-of-sample validation

The deferral-fraction frontier used fixed-factor recovery for screening. This validation retests every candidate with the same resampled future cost, productivity, and capacity factors.

Five future paths per episode, 100 paths per target. Future years are re-solved with the rolling persistence forecast and a two-year lookahead. Completion remains from the fixed-factor calendar scan; this validation covers risk recovery and cost.

Results

Target

Paths

Final gap P50

Final gap P90

Recovery P50

Within 0.005

Reopt cost P50

0%

100

0.0000

0.0000

0.0000

100.0%

295,899

5%

100

-0.0010

0.0072

0.0112

87.0%

295,043

10%

100

-0.0010

0.0072

0.0214

87.0%

291,878

15%

100

-0.0010

0.0072

0.0312

87.0%

289,012

20%

100

-0.0010

0.0072

0.0421

87.0%

285,927

25%

100

-0.0010

0.0072

0.0516

87.0%

283,018

The final-gap distribution is the same across targets because the same future path set is used and reoptimization closes the gap in most paths. Recovery and cost vary with the deferral fraction.

All tested fractions recover most or all of the fixed-future risk gap under resampled futures. The 25% point has the highest recovery and lowest median reoptimized cost, but also the largest fixed-future risk loss. The table does not select a risk appetite.

Limits

  • First-year deferral is fixed.

  • Future factors are independent draws; no serial correlation or calibrated forecast model.

  • The rolling policy is a persistence-forecast heuristic, not a stochastic dynamic program.

  • No recurrence, weather, outage, safety, or asset-owner priority model.