NEL2: From Radial Displacement to Claim-Gated Growth Forecasts: Identifiability, Telemetry, Uncertainty, Recovery, and Final Synthetic Closure

Research question

When do measurement identifiability, information available at forecast time, predictive skill, uncertainty, and recovery jointly justify a growth claim?

MODEL SCOPE

Public model description

The measurement model separates irreversible growth from reversible water deformation, temperature response, and instrument effects. Equal-input, different-growth constructions delimit identifiability; virtual component references add explicit information assumptions. Forecasts train on available algorithmic labels, while latent growth remains a separate evaluation target. Intervals, selective issuance, and recovery each have their own information and acceptance contract.

PROTOCOL

Declared evidence chain

NEL2.1–NEL2.12 progress from source identity and annual memory–climate reconstruction through identifiability, extraction, cadence, telemetry, future growth, uncertainty, abstention, shift/recovery, and sample sufficiency to frozen closure. Annual observations are reused retrospectively rather than presented as pristine holdout evidence. The final synthetic cohort contains 240 parents with 4,560 linked replays; the primary weekly comparison uses common accepted windows. Previously frozen components and negative gates are retained after the final evaluation.

Series: NEL2.1–NEL2.12 · 60 recorded annual series · 240 synthetic parents · 4,560 linked telemetry replays

• NEL2.1–NEL2.12
• 60 recorded annual series
• 240 synthetic parents
• 4,560 linked telemetry replays

The labels below preserve the public protocol identity. They are an index to the publication, not a substitute for its full methods or an open reproducibility bundle.

EVIDENCE

Results inside the model

1. On 9,777 accepted windows out of 12,800 scheduled primary weekly forecasts, latent-growth MAE is 7.065 µm versus 9.036 µm for Persistence on the same windows: a 21.81% reduction.
2. Exceedance risk at the fixed diagnostic tolerance is 35.43%; the weakest group has 75.44% growth coverage under nominal 90% intervals. None of 160 primary drift-challenge parents receives a timely first alarm. G4 and G5 fail, so the integrated conjunction remains negative.
3. Additional virtual instrumental information supports conditional correction, but blind-reference and repeated-change controls expose its limits. The separate automatic growth-claim layer issues zero claims under prior qualification; non-issuance has undefined error and coverage, not perfect performance.

BOUNDARY

CLAIM BOUNDARY

Numerical evidence is not physical validation. Transfer to a material, device, organism, environment, or operational service requires a separate validation chain whenever such a transfer is relevant.

Publication record