OPERATIONS / EXECUTION
Work, with a paper trail.
Real jobs, worker receipts, and durable events. No synthetic eval scores.
Token stays in this page’s memory and is sent only in the Authorization header to this origin. Use HTTPS on remote deployments.
DOCUMENT HILLCLIMBING
Compare on a frozen set
Select uploaded sources from document runs. A frozen set keeps exact document versions; every comparison uses one pinned grader.
Connect to manage document comparisons.
EXECUTION MACHINES
Worker capacity & draining
Capacity is last advertised available capacity, not a live health check. Draining stops new assignments and lets existing work report.
EVALUATIONS / PDF TO DOCX
Convert, render, measure
Fixed-position editable baseline. Fidelity measurements do not guarantee semantic structure or Microsoft Word compatibility. Conversion and grading use separately pinned images.
Connect to view document runs.
EVALUATIONS / CUSTOM TEXT
Compare real candidate runs
A small custom-text benchmark, not RULER or SWE-bench. Cases and candidates freeze when submitted. Expected answers stay with the coordinator; workers receive only input.
Connect to run and view evaluations.
Queue
Current coordinator projection. Assigned work may still be running or awaiting a report.
Connect to view jobs.
Enqueue a smoke command
Runs a trusted native command on an available worker. Resource requests are admission accounting, not isolation limits.
A submission error can leave the outcome uncertain. Retry with the same ID and unchanged fields to preserve duplicate protection.
Event history
Ordered facts from the durable log. Inspect past events without changing live queue state.
No events loaded.
Connect to inspect event envelopes.