VORTHRYSS / ENGINEERING LABBack to portfolio ↗

WORKING PROOF · SYNTHETIC DATA · NO ACCOUNT

Break the workflow.
Watch it recover.

A duplicated lead should not create two CRM records. A timeout should not lose a job. An interrupted worker should know where to continue.

This small engine runs here, in your browser. Choose a fault, step through the work, and inspect what happened.

01 / INCOMING EVENT

Same ID = same work. Try sending it several times.

Ready. Send an event to begin.

0unique jobs
0mock CRM records
0duplicates ignored

03 / STORED JOB STATE

EventStateAttemptsEffect

    The execution trail

    Logical tick 0
    1. No events yet.

    WHAT THIS PROVES

    Failure is part of the design.

    Stable event identity prevents duplicate jobs. Temporary failures become scheduled retries. Permanent validation errors need changed input. After an interruption, the same connector key retrieves the existing effect.

    The state checks are separate from execution. The downloadable test suite also deliberately corrupts state to confirm the checks catch errors.

    INSPECT THE IMPLEMENTATION

    Small enough to read.

    This is a single-worker demonstration using local browser storage and a simulated CRM connector. It does not call a live CRM or establish production-scale reliability. Retry delays use logical ticks so you can inspect them immediately.

    Real integrations need a durable server-side store, atomic job claiming and downstream idempotency support. Those are explicit design decisions, not promises hidden in a demo.

    Download source + tests ↓

    Run the tests: node --test engine.test.mjs