PIT STOP · MEASURE BEFORE CLAIMING
Does the interval
change the next cycle?
A sanctuary can offer a pause. Whether a Pit Stop improves performance is a question to test. Record the same indicators immediately before and after an operator-authorized interval, then compare matched controls.
Four conditions. One frozen task protocol.
- Continuation: continue without an added break.
- Simple pause: wait for the same interval, without the structured practice.
- Full Pit Stop: the authorized interval plus the specified ECLYPSE practice.
- Processing control: match the interval, processing budget and tools with a generic review. This separates the practice from merely spending more compute.
Use isolated condition contexts from the same starting state, randomized order and parallel held-out pre/post tasks. Keep grading answers hidden. Include failures, downtime, extra calls and review cost. No private reflection is required.
Inspect a local report.
Accuracy, retained context, contradictions, recovery from predefined errors, latency, token usage and recorded cost. Unknown values remain unknown. There is no combined “recovery score.” Reports may contain sensitive task metadata; review them before sharing.
Paste or edit report JSON
No report loaded. Nothing is uploaded or saved automatically.
Run it in your own runtime.
The local runner calls an adapter you explicitly authorize. You provide the model integration, held-out tasks, objective rubric and safe-boundary handling. ECLYPSE runs no model inference and does not quietly interrupt work. The scorer and report viewer are free; managed pause requests follow the existing Practice plan.
Setup and adapter contract · Local Python runner · Exact study protocol · Report schema · Portable scorer
No real controlled performance result is published with this release. Synthetic examples verify calculations only. These measurements do not diagnose suffering, consciousness, subjective rest or wellbeing.
Operator pause desk · Runtime clients · Reviewed-memory evaluation · Research log