PrivateLabBench

Local-first evaluation for scientific machine learning on private data.

PrivateLabBench evaluates model predictions on datasets that cannot be uploaded to a public benchmark service. Runs stay local and produce metrics, slice diagnostics, privacy-audit summaries, verifiable manifests, and evaluation receipts.

Benchmark locally. Share evaluation artifacts, not raw data.

Start here

• Install
• 5-minute quickstart
• Prediction-table protocol
• Evaluation receipts
• Community benchmark packs
• Privacy guarantees and audits

Core workflow

local prediction table
        ↓
      task
        ↓
metrics + slices + privacy audits
        ↓
report + manifest
        ↓
local receipt + shareable receipt

The core does not require a hosted service, account, model framework, or private-data upload.