Reproducible vertical federated learning evidence
See what changes when the data stays distributed.
Explore VertiMosaic’s synthetic VFL benchmark across model quality, party contribution, entity overlap, party dropout, feature drift, communication, timing and memory — with provenance attached to every published run.
Research boundary: raw-feature locality is demonstrated; cryptographic privacy is not implied.
Model qualityROC-AUC, PR-AUC, F1 and calibration metrics.
Party contributionBank, Telecom, Insurance and Retail feature views.
RobustnessPartial overlap, availability and controlled drift.
ProvenanceSeed, source commit and system measurements.
—core benchmark rows
—fixed experiment seed
4synthetic party roles
NOraw feature pooling
CPUreference execution
What do you want to understand?
Jump directly to the experiment family that answers your question.
Interactive evidence explorer
Each view is generated from data.json; use the selectors to interrogate the run.
Loading provenance…
Loading benchmark evidence…
Reading the current
Reading the current
data.json published by the VertiMosaic pipeline.