This document describes the corrected released analysis. It is not a preregistration.
At a fixed monitoring checkpoint, estimate when the measured battery capacity margin would reach the NASA 1.4 Ah boundary and quantify how that estimate changes as more evidence becomes available.
The former first 60% of final history rule depended on the eventual recorded observation horizon.
The released protocol replaces that retrospective rule with common fixed checkpoints at cycles 40, 60, and 80.
A discharge cycle nested within one of four battery cells.
For each battery and checkpoint:
- last observed capacity;
- current margin to 1.4 Ah;
- fitted OLS slope;
- projected threshold crossing cycle;
- observed crossing cycle when available;
- signed projection error;
- absolute projection error;
- forecast revision from the previous checkpoint;
- whether the projection lies inside the recorded horizon;
- false early warning flag for no crossing cells.
For each checkpoint:
- mean absolute crossing cycle error across cells with observed crossings;
- median absolute crossing cycle error;
- mean absolute forecast revision across all cells with a previous forecast;
- false early warning count.
B0007 has no observed 1.4 Ah crossing through cycle 168.
No artificial crossing cycle is imputed.
Aggregate MAE is never interpreted alone. Cell level forecasts and median error are reported because the mean can be dominated by a large individual error.
Forecast revision is interpreted as descriptive evidence of forecast stability. It is not a calibrated uncertainty interval or probability of failure.
- observed capacity trajectories with threshold and checkpoints;
- fixed checkpoint processing pipeline;
- cell level absolute threshold cycle errors;
- per cell forecast revision;
- evidence and claim boundary.
Any future change to the threshold, checkpoints, model family, crossing definition, error metric, or source cells should be released as a new analysis version rather than silently replacing this protocol.