> ## Documentation Index > Fetch the complete documentation index at: https://docs.physionlabs.ai/llms.txt > Use this file to discover all available pages before exploring further. # Run a failed evaluation again > Re-files a failed run's own request and returns the NEW evaluation, in `queued`. The original is left alone, with `retried_by` pointing at the successor — the failure stays on the record, because `error.code` is the only thing that can answer whether retrying was worth it. **This is the one submission in this API that is idempotent**, and it is idempotent on the run being retried rather than on your request. A burst of presses picks exactly one winner; every other press is handed that same winner. `POST /v1/evaluations` has no such guarantee — see its description. **It costs the ordinary price, and that is not paying twice.** A run that failed was refunded when it settled: in full on `failed`, and on `partial` exactly the detectors that did not land. The detectors that DID land were cached at the same moment, so the retry pays for the missing legs and gets the rest from cache. It consumes a `detect` rate-limit slot like any submission. Refused with `400` when there is nothing to retry, and the message says which case it is: the run has not finished, it did not fail, it delivered every detector it was asked for, it analyzed no stored clip (submit it fresh instead), it has already been attempted too many times, or it was already retried and that retry has since been deleted. ## OpenAPI ````yaml /openapi/galileo-v1.yaml post /v1/evaluations/{evaluation_id}/retry openapi: 3.1.0 info: title: Galileo API version: 0.1.0 summary: Evaluate generated videos with Galileo. description: | Submit generated videos for visual-glitch and prompt-misalignment analysis. Evaluation jobs run asynchronously. Retrieve a job until its status reaches `completed`, `partial`, or `failed`. contact: name: Physion Labs support email: support@physionlabs.ai license: name: Proprietary identifier: LicenseRef-Proprietary servers: - url: https://api.physionlabs.ai description: Production - url: https://api-dev.physionlabs.ai description: Development security: - bearerAuth: [] tags: - name: Platform description: Service health and account metadata. - name: Evaluations description: Submit videos and retrieve Galileo results. - name: Videos description: Upload a video once and reference it from evaluations. paths: /v1/evaluations/{evaluation_id}/retry: parameters: - $ref: '#/components/parameters/EvaluationId' post: tags: - Evaluations summary: Run a failed evaluation again description: > Re-files a failed run's own request and returns the NEW evaluation, in `queued`. The original is left alone, with `retried_by` pointing at the successor — the failure stays on the record, because `error.code` is the only thing that can answer whether retrying was worth it. **This is the one submission in this API that is idempotent**, and it is idempotent on the run being retried rather than on your request. A burst of presses picks exactly one winner; every other press is handed that same winner. `POST /v1/evaluations` has no such guarantee — see its description. **It costs the ordinary price, and that is not paying twice.** A run that failed was refunded when it settled: in full on `failed`, and on `partial` exactly the detectors that did not land. The detectors that DID land were cached at the same moment, so the retry pays for the missing legs and gets the rest from cache. It consumes a `detect` rate-limit slot like any submission. Refused with `400` when there is nothing to retry, and the message says which case it is: the run has not finished, it did not fail, it delivered every detector it was asked for, it analyzed no stored clip (submit it fresh instead), it has already been attempted too many times, or it was already retried and that retry has since been deleted. operationId: retryEvaluation responses: '201': description: A new evaluation, queued. headers: X-RateLimit-Limit: $ref: '#/components/headers/RateLimitLimit' X-RateLimit-Remaining: $ref: '#/components/headers/RateLimitRemaining' X-RateLimit-Reset: $ref: '#/components/headers/RateLimitReset' content: application/json: schema: $ref: '#/components/schemas/Evaluation' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '402': $ref: '#/components/responses/InsufficientCredits' '404': $ref: '#/components/responses/NotFound' '429': $ref: '#/components/responses/RateLimited' '500': $ref: '#/components/responses/InternalServerError' '501': $ref: '#/components/responses/NotImplemented' components: parameters: EvaluationId: name: evaluation_id in: path required: true description: Evaluation identifier. schema: type: string format: uuid headers: RateLimitLimit: description: Maximum submissions in the current sliding window. schema: type: integer minimum: 0 RateLimitRemaining: description: Submissions available before the next refusal. schema: type: integer minimum: 0 RateLimitReset: description: Unix seconds when the next rate-limit slot becomes available. schema: type: integer minimum: 0 RetryAfter: description: Seconds to wait before retrying. schema: type: integer minimum: 1 schemas: Evaluation: type: object required: - id - object - model - model_version - output_schema_version - created - status - input - usage - result properties: id: type: string format: uuid object: type: string const: evaluation model: $ref: '#/components/schemas/ModelId' model_version: type: string output_schema_version: type: string created: type: integer minimum: 0 description: Unix timestamp in seconds. status: $ref: '#/components/schemas/EvaluationStatus' input: type: object required: - prompt - video properties: prompt: type: string video: $ref: '#/components/schemas/VideoInfo' usage: $ref: '#/components/schemas/EvaluationUsage' result: oneOf: - $ref: '#/components/schemas/EvaluationResult' - type: 'null' error: description: >- Why the run failed, on a `failed` one. NULL on a run that did not fail, and ABSENT on rows written before this field existed. Not in `required`, and that is a correction rather than a preference: it was, and the older rows in the store do not carry the key at all. A client that enforced the contract at runtime — the Python one does, since its models validate — raised on any page deep enough to reach them, while the TypeScript client passed because its types are erased before a response is ever seen. So the contract was not merely wrong, it was wrong in a way that only one of the two clients could report. Test for a VALUE, not for the key. oneOf: - $ref: '#/components/schemas/EvaluationFailure' - type: 'null' detectors: type: array description: Per-detector state. Older evaluations may omit this field. items: $ref: '#/components/schemas/DetectorState' video_id: type: - string - 'null' description: Stored video identifier when the evaluation used an uploaded video. attempt: type: integer minimum: 1 description: >- Which try this is. 1 for a run submitted directly; 2 or more for one produced by `POST /v1/evaluations/{evaluation_id}/retry`. There is a ceiling, so a clip that keeps failing under the same instructions stops being retryable rather than being retried forever. timing: description: >- How long this run took. NULL when nothing was measured -- which is every run settled before it was recorded, and is not backfillable. Null and never 0. A run whose latency nobody recorded and a run that took no time are different claims, and only one of them is true. NOT in `required`, for the same reason `error` is not: during a rolling deploy some tasks are still the older build, and a response from one of those carries no such key. A client that enforces the contract at runtime -- the Python one does -- would raise on those responses, and it would raise only during a deploy, which is the worst time to be debugging a client. Test for a VALUE, not for the key. oneOf: - $ref: '#/components/schemas/Timing' - type: 'null' metadata: type: - object - 'null' additionalProperties: true description: >- Whatever you passed on create, echoed back. NULL when you passed nothing -- the key is always present, its value says whether there was any. Only found by submitting through the API. Every evaluation created in the console carries metadata, so a contract checked against console traffic alone looked correct here. ModelId: type: string enum: - galileo - gemini default: galileo EvaluationStatus: type: string enum: - queued - processing - completed - partial - failed VideoInfo: type: object additionalProperties: false required: - duration_sec - width - height - fps - num_frames properties: duration_sec: type: number minimum: 0 width: type: integer minimum: 1 height: type: integer minimum: 1 fps: type: number exclusiveMinimum: 0 num_frames: type: integer minimum: 1 EvaluationUsage: type: object additionalProperties: false required: - video_seconds - billable_units properties: video_seconds: type: number minimum: 0 billable_units: type: number minimum: 0 EvaluationResult: type: object additionalProperties: false required: - glitches - summary properties: glitches: type: array items: $ref: '#/components/schemas/Glitch' summary: $ref: '#/components/schemas/EvaluationSummary' EvaluationFailure: type: object additionalProperties: false required: - type - code - message - request_id properties: type: type: string code: type: string message: type: string request_id: type: string DetectorState: type: object additionalProperties: false required: - detector - status properties: detector: $ref: '#/components/schemas/GlitchType' status: $ref: '#/components/schemas/DetectorStatus' error: $ref: '#/components/schemas/DetectorError' Timing: type: object additionalProperties: false required: - e2e_ms properties: e2e_ms: type: integer minimum: 0 description: >- Milliseconds from submission to the terminal answer, measured by the service. This is what you waited: it includes our queueing, fetching the video, and the model's own time. One number rather than a breakdown, deliberately. The finer measurements are the model server's own clock in its own units, and a number read from the wrong level is wrong by three orders of magnitude rather than plausibly close. ErrorResponse: type: object required: - error properties: error: $ref: '#/components/schemas/Error' quota: $ref: '#/components/schemas/RateLimitWindow' Glitch: description: >- One finding. Which fields it carries depends on `type`, so branch on that rather than probing for a field. A KEY THAT DOES NOT APPLY IS ABSENT, not null. This is a correction: the service used to send `severity: null` on a visual glitch and `prompt_segment: null` alongside it, and this schema said so. It no longer does — the response is now assembled from an allowlist per finding type, so a visual glitch has no `severity` key at all. Code written against the old shape kept working, because `x.severity == null` is true either way; code that tested `"severity" in x` did not. oneOf: - $ref: '#/components/schemas/VisualGlitch' - $ref: '#/components/schemas/PromptMisalignment' discriminator: propertyName: type mapping: visual_glitch: $ref: '#/components/schemas/VisualGlitch' prompt_misalignment: $ref: '#/components/schemas/PromptMisalignment' EvaluationSummary: type: object additionalProperties: false required: - num_glitches - has_visual_glitch - has_prompt_misalignment properties: num_glitches: type: integer minimum: 0 has_visual_glitch: type: boolean has_prompt_misalignment: type: boolean GlitchType: type: string enum: - visual_glitch - prompt_misalignment DetectorStatus: type: string enum: - pending - reused - completed - failed - skipped DetectorError: type: object additionalProperties: false required: - code - message properties: code: type: string message: type: string Error: type: object required: - type - code - message - request_id properties: type: $ref: '#/components/schemas/ErrorType' code: $ref: '#/components/schemas/ErrorCode' message: type: string request_id: type: string RateLimitWindow: type: object required: - limit - used - remaining - reset_at - reset_in_sec - window_sec - unlimited properties: limit: type: integer minimum: 0 used: type: integer minimum: 0 remaining: type: integer minimum: 0 reset_at: type: - integer - 'null' minimum: 0 reset_in_sec: type: - integer - 'null' minimum: 0 window_sec: type: integer minimum: 1 unlimited: type: boolean VisualGlitch: type: object additionalProperties: false required: - id - type - description properties: id: type: string type: type: string const: visual_glitch description: type: string source: $ref: '#/components/schemas/GlitchSource' region: $ref: '#/components/schemas/GlitchRegion' description: |- Where in the video this finding is, with per-frame boxes. Absent on a finding the detector localised no further than the clip. PromptMisalignment: type: object additionalProperties: false required: - id - type - description properties: id: type: string type: type: string const: prompt_misalignment description: type: string source: $ref: '#/components/schemas/GlitchSource' prompt_segment: $ref: '#/components/schemas/PromptSegment' description: The span of your prompt this finding is about. severity: type: integer minimum: 1 maximum: 5 description: >- How far a prompt requirement was from being realized: `6 - score`, so 1 is a minor mismatch and 5 means the requirement is absent entirely. It reads in the OPPOSITE direction from a score, which is the one thing worth getting right before you threshold on it. This is also the number our own reporting threshold reads, so a finding you receive is by definition one that cleared it. There is deliberately no `confidence` beside it. That is the model's certainty about its own answer -- a different axis, not stable across releases, and not calibrated for anyone outside to threshold on. Optional: a finding from an older pipeline may carry no score. ErrorType: type: string enum: - invalid_request_error - authentication_error - rate_limit_error - api_error ErrorCode: type: string enum: - invalid_body - unknown_model - missing_video - missing_prompt - prompt_too_long - invalid_glitch_types - video_too_long - invalid_video - not_found - not_implemented - insufficient_credits - missing_api_key - invalid_api_key - unauthenticated - rate_limited - concurrency_limit - internal_error - model_unavailable - model_output_invalid - model_timeout - run_abandoned GlitchSource: type: string enum: - model - human description: >- Who produced this finding. Absent means the model, which is the common case -- the field is only set explicitly where a human annotated something the model missed, so treat its absence as `model` rather than as unknown. Deliberately NOT declared with a `default`. A default reads to a code generator as "the server always sends this", and it does not: the value is omitted, not defaulted, and a generated type that made it required would be wrong on almost every finding. GlitchRegion: type: object additionalProperties: false required: - start - end - boxes properties: start: $ref: '#/components/schemas/TimePoint' end: $ref: '#/components/schemas/TimePoint' boxes: type: array items: $ref: '#/components/schemas/BoxKeyframe' PromptSegment: type: object additionalProperties: false required: - text - char_start - char_end properties: text: type: string char_start: type: integer minimum: 0 char_end: type: integer minimum: 0 TimePoint: type: object additionalProperties: false required: - frame - sec - timecode properties: frame: type: integer minimum: 0 sec: type: number minimum: 0 timecode: type: string pattern: ^[0-9]{2}:[0-9]{2}:[0-9]{2}:[0-9]{2}$ BoxKeyframe: type: object additionalProperties: false required: - frame - sec - box properties: frame: type: integer minimum: 0 sec: type: number minimum: 0 box: $ref: '#/components/schemas/BoundingBox' BoundingBox: type: object additionalProperties: false required: - xmin - ymin - xmax - ymax properties: xmin: type: number minimum: 0 maximum: 1 ymin: type: number minimum: 0 maximum: 1 xmax: type: number minimum: 0 maximum: 1 ymax: type: number minimum: 0 maximum: 1 responses: BadRequest: description: The request cannot be processed as sent. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' Unauthorized: description: The API key is missing, invalid, or revoked. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' InsufficientCredits: description: The account balance cannot cover this evaluation. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' NotFound: description: The resource does not exist or is unavailable to this account. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' RateLimited: description: The account reached a rate or concurrency limit. headers: X-RateLimit-Limit: $ref: '#/components/headers/RateLimitLimit' X-RateLimit-Remaining: $ref: '#/components/headers/RateLimitRemaining' X-RateLimit-Reset: $ref: '#/components/headers/RateLimitReset' Retry-After: $ref: '#/components/headers/RetryAfter' content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' InternalServerError: description: The API encountered an unexpected error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' NotImplemented: description: This deployment does not implement the requested operation. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' securitySchemes: bearerAuth: type: http scheme: bearer bearerFormat: Galileo API key description: Use a key created in the Playground. Keys begin with `gk_`. ````