Repository navigation
Re-check ADR 0002's feature choices on real pilot data #233
Description
Is your feature request related to a problem? Please describe.
ADR 0002's evidence comes only from synthetic pilots. The generator scales deviations multiplicatively, which is what a log scale assumes, so log winning shows it handles that structure, not that real pilots have it. The overlap constants were also set after trying a few settings.
Describe the solution you'd like
When real telemetry with targets exists, rerun the comparison from #231 on it and confirm or amend the ADR. Real pilots have no hidden tier, so compare silhouette and subsample stability instead of adjusted Rand index.
Describe alternatives you've considered
Treat the synthetic result as sufficient. The ADR already says it cannot establish this.
Additional context
Blocked by #213 (targets), #214 (pilot metadata) and #227 (client data handling).
Activity
jmacd867 commentedon Oct 15, 2026
Note: the first "real pilot data" is the client's Prepar3D export at about 1 Hz (see #228), not plugin data at 20 Hz. A re-check on it tells us about that data. Since
stddepends on the sample rate, the result may need repeating once X-Plane flights through the plugin exist.