weavegate

Deterministically replay the DB race, gate the deploy.

Why we build · What we build · Start here · Community

We build deterministic tools for application workflows whose database failures depend on execution order. Our first project, weavegate, controls the interleavings around transactional read → decide → write paths and lets rule-based SQL oracles judge the resulting state. The outcome is a saved schedule and concrete evidence that developers can replay against both the failure and its fix.

• Find the order · bounded exploration — expose the interleaving ordinary tests may miss.
• Force the race · named sync-points — coordinate application workers without timing sleeps.
• Prove the outcome · SQL oracles and artifacts — retain the verdict, state, trace, and replay command.

See weavegate’s documented limits for the boundary of these claims.

• Control · schedules — explore and replay release orders between user-placed sync-points.
• Judge · composable oracles — evaluate application invariants on real database state.
• Retain · CI evidence — write diagnostics, reports, traces, observations, and exit codes.

See the project README for current implementation, measured results, architecture, and planned boundaries.

• Run · Quickstart — build the CLI and reproduce the reference violation.
• Understand · Architecture — follow configuration, execution, judgment, and evidence flow.
• Verify · Measured results — inspect methods, environments, and reproduction commands.
• Contribute · Contributing — add fixtures, improve evidence, or propose a change.

See the documentation index for concepts, how-to guides, reference, and experiments.

• Questions · use GitHub Discussions.
• Defects · follow the routing in Support.
• Security · report vulnerabilities through the private route in Security.
• License · weavegate is available under Apache-2.0.

See the changelog for shipped work and milestones for planned work.