Example 1
Compare two experiments
- Starting point
- Baseline and revised outputs use the same task.
- Useful result
- A side-by-side check of outcomes under a visible deterministic rule.
Inspect original sample traces or import your own JSON. Follow the spans, review the outputs and keep an evidence-based record.
| Case / span | Experiment | Check | Latency | Review |
|---|
Original sample traces. Import your own logs; no model requests are made.
The essentials
Review the input, output and check together before deciding whether a change helps.
Possible starting points
Example 1
Example 2
A clear path
Use sample traces or import your own JSON or JSONL.
Inspect spans and run the available text checks.
Annotate the result and export the cases for comparison.
Access & output
Reviewed traces in JSON or CSV. No account or payment is needed for the included local tools.
Keep: Reviewed traces in JSON or CSV
Inspect the tracesGood to know
No. Checks run on the text already in the collection. No model endpoint or credential is needed.
It proves only the selected deterministic rule passed for that input. Review correctness, coverage and the user outcome separately.
Reviewed traces in JSON or CSV. Saved work stays in this browser on this device. It is not an account backup or cross-device sync; keep a separate export.
Reviewed traces in JSON or CSV. No account or payment is needed for the included local tools.
Open the contact page to prepare a question about Lens. You can download the draft or use a configured email address; the form never silently sends it.
Your next step