Validation
How the model was tested
Calibration fits; validation tests. These are checks against evidence the model was not tuned to — an independent glacier, a satellite melt record, a held-out decade, and the geometry it must reproduce from observed thinning alone. Both the passes and the one documented failure are shown.
The ledger · six checks
Δh outline hindcast · geometry from observed thinning
Geodetic anchor residual · the documented residual
Stake fit · observed vs model
Snowline fit · observed vs model
Wolverine Tier-1 · regional coherence
SAR seasonal phenology · supporting check
Held-out 2000s · the failure we predicted
Provenance & caveats
Validation is separate from calibration: the model is exercised against evidence it was not tuned to. The Δh hindcast reproduces the 2000–2025 outlines from Hugonnet’s observed thinning with no fitted parameters; Wolverine Glacier, the SAR melt trajectory, and a held-out 2000s decade are independent checks. The geodetic-anchor residual (z ≈ +1.3) and the held-out failure are documented here alongside the passes.