Expand description
The error envelope and the closed set of error codes.
Every rejection the hub makes has the same shape:
422 { "errors": [ { "path": "results[0].metric", "code": "metric_id_invalid", "hint": "..." }, ... ] }Validation collects all errors before returning, so a producer fixes a
record in one round trip rather than one error at a time. path is a
JSON-pointer-like location in the submitted record; code is one of the
variants below and is listed as an enum in the OpenAPI document; hint
is prose for a human and is not part of the contract.
| Code | Status | Meaning |
|---|---|---|
schema | 422 | JSON Schema violation (unknown key, wrong type, missing required) |
number_too_large | 422 | integer outside ±2^53 |
counts_missing | 422 | results non-empty but no counts |
counts_inconsistent | 422 | attempted < completed + failed + skipped + errored |
attachment_ref_unknown | 422 | samples_ref / calls / artifacts[] / error.log not in attachments[].path |
attachment_path_invalid | 422 | duplicate, .., or absolute attachments[].path |
metric_id_invalid | 422 | results[].metric, run_results[].metric or a run’s metrics key not of the form {ns}/{name} |
ext_key_invalid | 422 | a key of ext (top-level or in a facet) not of the form {ns}/{name} |
type_mismatch | 422 | query: operator applied to a path of another type |
not_indexed | 422 | query: operator needs an index the path does not have |
unknown_path | 422 | query: path not in the schema |
runs_moved | 422 | an evalhub.eval/2.0 header carries runs; runs are written apart |
run_status_detail | 422 | error present iff status is error, skip_reason iff skipped |
run_id_invalid | 422 | run_id empty, containing /, or over 200 bytes |
run_id_mismatch | 422 | the run_id in the body differs from the one in the path |
run_result_value_or_label | 422 | a run_results[] element has neither value nor label |
run_results_eval_unknown | 422 | run_results[].eval is not the record of any core/uses_eval |
run_unknown | 422 | a run_id in the used set has no row in that Eval |
run_not_in_used_set | 422 | run_results[].run_id outside the Card’s used set for its eval |
too_many_run_results | 422 | more run_results[] than the configured limit |
batch_duplicate_run_id | 422 | the same run_id twice in one batch |
attachment_missing | 409 | attachments[].sha256 not uploaded and confirmed |
label_in_use | 409 | @{label} already names another version |
namespace_in_use | 409 | the organisation slug is already a user or an organisation |
registry_entry_exists | 409 | that registry address is taken; entries are immutable |
run_deleted | 409 | writing a run_id that was deleted; a deleted id is never reused |
batch_too_large | 413 | more runs in one batch than the configured count |
body_too_large | 413 | request body over the configured size in bytes |
A 413 carries this body only with the two 413 codes above. Other
statuses carry no body of this shape: 404 for anything the caller
may not know exists (including private records), 403 for a token
without the scope, 401 for no or an invalid token, 400 for a body that
is not JSON.
Codes are added, never renamed or reused. A client switching on code
must treat an unknown code as a generic 422. The enum is
#[non_exhaustive] for the same reason: a release may add a code, and a
match in another crate must have a fallback arm for it.
Structs§
- Error
Entry - One rejection.
- Error
Envelope - The body of every
422and409response.
Enums§
- Error
Code - The closed set of error codes.