GET /v1/dev-reporting/ai-record returns a tenant's read-only feedback record. The request requires the verify:read scope.
from and to use YYYY-MM-DD. By default, from is the first day of the current UTC month and to is today in UTC. from must not be after to, and the inclusive range may be at most 366 days. Each invalid format, reversed range, or overlong range is refused with 400.
| Field | Description |
|---|---|
tenant_id | The identifier of the tenant whose record was read. |
from | The first included UTC calendar day in the requested range. |
to | The last included UTC calendar day in the requested range. |
source | The fixed source label for this record, feedback. |
figures | The four rate-bearing figures derived from feedback and judged criteria. |
disputed | The object containing the count of disputed feedback. |
by_model | The current model-split availability marker. |
test_split | The current test-split availability marker. |
escalations | The current escalation-figure availability marker. |
agreement uses feedback on verifications with a met, partial, or not_met verdict that was marked confirmed_correct or confirmed_wrong; its count is the confirmed_correct feedback.
false_met uses the same kind of confirmed feedback, limited to verifications with a met verdict; its count is the confirmed_wrong feedback. overturn uses the same sample as agreement; its count is the confirmed_wrong feedback.
abstention uses criteria judged in the range; its count is the criteria that returned cannot_determine. disputed feedback is a count with no rate. A verification counts once, using its newest feedback.
| Field | Description |
|---|---|
count | The number of events counted by this figure. |
sample | The number of eligible observations in this figure. |
rate | The count divided by the sample, or null when the rate is withheld. |
too_few_to_show | Whether the sample is below the minimum needed to show a rate. |
first | The earliest included observation time, or null when there is no sample. |
last | The latest included observation time, or null when there is no sample. |
When a figure has fewer than 30 samples, it still shows its count, but its rate is null and too_few_to_show is true. A withheld rate is never a zero rate.
The model split is not yet available: by_model is not_split_for_feedback. The test split and escalation figures are not yet available: test_split is not_yet_available and escalations is not_yet_available.
No figure changes a verdict, an escalation, or a consequence. No feedback reaches any training or calibration path.