Run Inspector & Run Doctor
Debug a failed sync from the browser — the call timeline, quarantine, and AI-assisted diagnosis.
The Run Inspector is how support debugs a sync without a terminal. It reads the persisted run and shows exactly what happened; the Run Doctor adds an AI-assisted diagnosis on top.
What the inspector shows
- Stats — pages, records in, skipped, records out, quarantined, errors.
- Call timeline — every HTTP request/response, redacted, grouped by
pipeline stage (
fetch:list,enrich:detail, …) with status, duration, and attempt. - Run errors — categorized (
auth,rate_limit,http,transform,validation,delivery,network,timeout). - Quarantine — records that failed validation or transform, with their errors — and a one-click requeue.
- Replay — re-run to reproduce a failure after a spec fix.
Because the call log is captured inside the durable journal, it survives replay — the inspector shows the calls that actually committed the run.
Run Doctor (AI)
Run Doctor sends the run's shape — the resource spec, stats, categorized errors, a sample of failing calls (with redacted response previews), and quarantine samples — to a model and gets back a structured diagnosis:
- a one-line summary and a category
(
auth·extraction·transform·pagination·delivery·rate_limit·no_issue·unknown), - a confidence (high/medium/low) and an explanation,
- an optional suggested spec patch (a JSON-path + the proposed value + rationale),
- concrete next steps.
The suggestion is grounded in pipeline-stage heuristics (a $.value recordsPath
against a bare-array response reads as an extraction bug; a run of 401s reads as
auth; and so on).
Degrades gracefully
Run Doctor is optional. Without an AI credential configured it simply isn't offered — the rest of the inspector works unchanged.