Request AI-suggested corrections for failed rows

Sends the roster's validation failures to an AI correction service and returns suggested values for the cells it can repair.

When to use: call this once validation has run and rows are sitting in VALIDATION_FAILED state. Saving drafts first is not required - Smart Fix reads the failed rows directly. Review what comes back, then apply the suggestions you accept with PATCH /roster-records/{rosterId}/draft and re-run validation.

Preconditions: supply the tenant-id header and the rosterId of an existing roster job. The roster must hold at least one VALIDATION_FAILED row carrying an error type eligible for correction - incorrect data type, invalid date, email, or format, a value outside the allowed set, an invalid NPI, or a missing required field. Rosters with more than 2000 failed rows are rejected with 400; resolve some rows and call again. Whether practitioner or facility rules are applied is derived from the roster's own job type and is not a caller-supplied parameter.

Cost: the failures are sent to the correction service in batches, so this call takes noticeably longer than a typical request on a large roster.

What you get back: one suggestion per correctable cell, each naming the source column, the original value, the system attribute it maps to, the suggested replacement, a confidence rating, and a status of auto_fixed, llm_suggested, or requires_human_review. Every suggestion carries the rowIndex, which is the row's rowId, so it can be matched back to the row it came from. The response also reports how many errors were submitted, fixed, and skipped, whether every batch succeeded, and any batch-level error messages. No rows are modified by this call - it reads and suggests, and nothing is stored until you save drafts. Calling it again on the same rows simply returns a fresh set of suggestions. Smart Fix can be turned off for the platform, in which case this endpoint returns 503 until it is re-enabled.