LHR Hold ML
Predict only · phase/geofence owns the real stack · actual-hold labels train ML
Stage 1
Soft
Far out / weak margin — geometry prior. Hint only.
Stage 2
High
Stronger ML / sector. Still a prediction until a hold geofence fires.
Observed
Phase
Hold geofence sets
stack_* — phase owns the real stack, not ML.Learning
Actual hold
Label the hold it used. Strengthens predictions only — never overrides phase.
Accuracy
—
labelled outcomes
Labels
—
actual-hold buttons
Route patterns
—
learned keys
LHR inbound
—
live fleet
Soft / High
—
prediction stages
In hold
—
phase stack_*
Live stack loadBNN · LAM · OCK · BIG
BNN Bovingdon NW
LAM Lambourne NE
OCK Ockham SW
BIG Biggin SE
How to trainfeeds ML
ML only predicts which hold an inbound will use.
When it hits a hold geofence, phase sets
stack_* and owns the real stack.
Use Actual hold buttons to label outcomes and train the model — that never overrides phase.
Live LHR arrivals
| Callsign | Hex | Alt | NM | Pred | Stage | Conf | Method | Alts | Phase | Actual hold |
|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||||
| Callsign | Hex | NM | Alt | Pred | Stage | Conf | Method | Actual hold |
|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||
| Route key | Typical hold | Samples | Confidence | Mix | |
|---|---|---|---|---|---|
| Loading… | |||||
| When | Callsign | Hex | Predicted | Actual | Source | Result |
|---|---|---|---|---|---|---|
| Loading… | ||||||