Endpoint frame
Settings and sensor channels at the final observed test cycle. Nearness under a declared metric is not exact state identity.
Recorded operational history / hidden state
Looks close now.
Has a different horizon.
Two simulated engines end on unusually similar observed frames, yet NASA supplies 145 remaining cycles for one and 50 for the other. Their recorded prefixes show how much the snapshot leaves unresolved.
01 / Source-locked result
A snapshot can hide a very different remaining horizon
The cohort contains all 100 truncated FD001 test trajectories. The selected pair is near under one declared current-frame profile; the selection itself deliberately uses the supplied outcomes and is therefore a demonstration, not an accuracy test.
02 / Flagship pair
Test unit 25 / test unit 72
Settings and 21 sensor channels form the endpoint record. Unit ID, cycle count, observed-history length, and provided RUL do not enter the current-frame distance.
03 / What history adds
The pair stops looking exceptional when the window grows
Lower rank means closer among all 4,950 endpoint pairs under that exact profile. Compare ranks within a profile; do not compare raw distances across windows as if their scales were interchangeable.
04 / Observed prefixes only
Read the path before the endpoint
Select one anonymous sensor channel. Both lines use the same raw-value scale and actual cycle axis. The chart contains only published test-prefix rows; it does not invent the missing future or expose simulator health.
05 / Observation to outcome
Provided outcome, not prediction
NASA distributes one RUL value per test endpoint in a separate vector. Onto2D may display that value after analysis, but it cannot feed it back into the distance that chose what “looks close.”
06 / Operating context control
One condition does not mean no context
FD001 contains one documented Sea Level operating condition and one HPC degradation fault mode. All three setting columns remain in the record; two vary in training data and enter the combined distance, while the zero-range third setting remains context only.
07 / Historical Load
Undefined is the honest result
—NOT EVALUATED
null, not zero08 / Reading discipline
Five layers stay separate
Operational prefix
Ordered rows from cycle 1 to the endpoint. No future test row is included or synthesized.
Latent health
The simulator health index was not supplied to participants. A derived mean cannot become a direct observation.
Remaining life
The separate RUL vector is a held-out outcome. Here it helps explain the case and makes pair selection biased.
Prediction
No RUL model, prediction, score, or claim of generalization appears in this release.