Calibrated uncertainty for earthquake mechanisms

Upload first-motion polarity picks or a focal-mechanism catalogue. The platform returns calibrated polarity probabilities, mechanism sets with a stated coverage level, and uncertainty carried into stress inversion. In this prototype the file check runs live; the example screens show cached results from saved runs.

Upload polarities or a mechanism catalogue

Pick level

Calibrated P(up) and a label set {U}, {D} or {U,D} for every pick

Event level

Mechanism sets with a stated coverage level, drawn as a cloud of nodal planes

Stress

SHmax and axis uncertainty with and without mechanism error

Pick or mechanism file→named reference set→calibrated sets→stress inversion

A · First-motion polarity picks

.csv

column	meaning
event_id	Events are the calibration unit (a pick set is judged jointly)
role	calibration (label known) or test
station, azimuth_deg, takeoff_deg	Position on the focal sphere (takeoff from the downward vertical)
polarity	Reference label U / D (catalogue or reader)
p_positive	Your model's raw probability of an up first motion

B · Mechanism catalogue

.csv

column	meaning
event_id, group_id	Views of one physical earthquake share a group_id
role	calibration (reference known) or prediction
pred_strike/dip/rake	Your point mechanism (degrees)
ref_strike/dip/rake	Named reference mechanism; blank for prediction rows
scale_deg	Optional: your own spread, for variable-radius sets

Try the format: example files cut from the saved runs. Drop one above and the check runs live in this browser.

B · mechanisms (628 rows)A · polarities (2,010 rows)

Example datasets loaded

cached results from saved runs · not computed nowexploratory

Polarity picks, calibrated

cached model output

Set coverage level90%

–

picks in this event

–

picks with a two-label set {U,D}

–

catalogue labels inside their set

–

event jointly covered

Most ambiguous picks

Reliability (59 test events)

grey ring = raw · blue dot = after T · size = picks

Mechanism set for one event

cached model output

Set coverage level90%

This event, view A

All 250 held-out Amatrice events

black line = stated level · live from saved scores

Synthetic benchmark · 120 trials, 6 station networks

Many events, one calibration

Amatrice sequence · cached

Set coverage level90%

calibration events

constraint cut42°

Held-out events shown

Uncertainty carried into stress inversion

California · 177 cells · cached

Change inside the 95% radius

Selected cell

Ridgecrest, before vs after the M7.1

Download results

demo · downloads inactive

Always attached to a result

Preview of the per-event table

computed live from the cached result

Calibrated uncertainty for earthquake mechanisms

Pick level

Event level

Stress

A · First-motion polarity picks

B · Mechanism catalogue

Example datasets loaded

Polarity picks, calibrated

Most ambiguous picks

Reliability (59 test events)

Mechanism set for one event

This event, view A

All 250 held-out Amatrice events

Synthetic benchmark · 120 trials, 6 station networks

Many events, one calibration

Held-out events shown

Uncertainty carried into stress inversion

Change inside the 95% radius

Selected cell

Ridgecrest, before vs after the M7.1

Download results

Always attached to a result

Preview of the per-event table

What the numbers do and do not mean