openai.fine_tuning_job.succeeded: the succeeded contract

Prove that the fine-tuning job finished and named a model. The job reached the status succeeded and the member fine_tuned_model holds a name. The level does not prove that the model answers a request. No published operation states that.

Contract identity.
Fact	Value
Contract id	`openai.fine_tuning_job.succeeded`
Version	`1.0.0`
Hash	`13b56dac6222a5f4`
Completion level	`succeeded`
Publisher	Built and signed by Provely.
Certification	Provisional
Skill	OpenAI Platform API 0.1.0
Valid for provider API versions	`spec_2_3_0`

What is the intent?

Fine-tune the base model on the training file.

What is the subject and the action?

Member	Value
Subject type	`openai.fine_tuning_job`
Subject identity	fine_tuning_job_id = `$action.result.id`
Canonical effect	`ai.fine_tuning_job_start`
Provider operation	`POST /fine_tuning/jobs`
Idempotency	none, retry is not safe

How does the evidence correlate with this operation?

Strategy	Assurance	Keys	Required
`resource_id`	strong	fine_tuning_job_id from `$action.result.id`	yes
`fingerprint`	weak	training_file from `$input.training_file`	no

Which evidence does the contract require?

Minimum evidence level E2. An independent channel is required. Minimum channels: 1.

Channel	Level	Independence	Verifier	Description
`fine_tuning_job_readback`	E2	provider readback	`http`	Read the fine-tuning job from the provider. A condition path is $observed.fine_tuning_job_readback.status, and a second one is $observed.fine_tuning_job_readback.fine_tuned_model.
`fine_tuning_job_succeeded_events`	E3	provider event	`webhook`	The fine_tuning.job.succeeded events of this job. The payload carries the identifier of the job and no other member of it, so the name of the model comes from the read-back. The webhook verifier returns {events, count, latest, earliest, types, duplicates_dropped}.

Which conditions must all hold for VERIFIED?

Condition	Meaning	Path	Operator	Expected	Evidence
`fine_tuning_job_present`	The provider holds the job with the returned identifier.	`$observed.fine_tuning_job_readback.id`	`eq`	`$action.result.id`	`fine_tuning_job_readback`
`fine_tuning_job_status_succeeded`	The job reached the status succeeded.	`$observed.fine_tuning_job_readback.status`	`eq`	`succeeded`	`fine_tuning_job_readback`
`fine_tuning_job_named_a_model`	The member fine_tuned_model holds a name. The schema states that the value is null while the job runs.	`$observed.fine_tuning_job_readback.fine_tuned_model`	`exists`	`undefined`	`fine_tuning_job_readback`
`fine_tuning_job_training_file_matches`	The job trained on the file that the intent named.	`$observed.fine_tuning_job_readback.training_file`	`eq`	`$input.training_file`	`fine_tuning_job_readback`
`fine_tuning_job_created_in_window`	The provider created the job after the operation started. OpenAI writes created_at in seconds since the epoch.	`$observed.fine_tuning_job_readback.created_at`	`time_after`	`$operation.created_at`	`fine_tuning_job_readback`
`fine_tuning_job_finished_in_window`	The provider finished the job after the operation started. The schema states that finished_at is null while the job runs.	`$observed.fine_tuning_job_readback.finished_at`	`time_after`	`$operation.created_at`	`fine_tuning_job_readback`
`fine_tuning_job_succeeded_event`	At least one fine_tuning.job.succeeded event names this job.	`$observed.fine_tuning_job_succeeded_events.count`	`gte`	`1`	`fine_tuning_job_succeeded_events`

Which conditions give CONTRADICTED?

Condition	Class	Reason	Path	Operator	Expected
`fine_tuning_job_on_wrong_training_file`	wrong subject	The job trained on a different file than the intent named.	`$observed.fine_tuning_job_readback.training_file`	`ne`	`$input.training_file`
`fine_tuning_job_predates_operation`	pre existing state	The fine-tuning job is older than the operation. It proves nothing.	`$observed.fine_tuning_job_readback.created_at`	`time_before`	`$operation.created_at`

Which observed states map to a verdict before completion?

Rule	Match	Verdict	Reason
`terminal_failure`	`$observed.fine_tuning_job_readback.status` eq `"failed"`	FAILED	The training did not finish. The member error of the job states the cause.
`cancelled`	`$observed.fine_tuning_job_readback.status` eq `"cancelled"`	CONTRADICTED	A person or an agent cancelled the fine-tuning job. Ask a person before a second run starts.
`still_running`	`$observed.fine_tuning_job_readback.status` in `["validating_files","queued","running"]`	PENDING	The job has not reached a terminal status. The training may take hours, so the runtime observes again later.

How long does the runtime observe?

Timing member	Value
Initial delay	1000 ms
Poll interval	30000 ms
Backoff	exponential factor 2, max 1800000 ms
Maximum attempts	360
Timeout	604800000 ms
Stale read window	60000 ms
On timeout	UNVERIFIABLE (`evidence_unavailable_before_timeout`), escalated to a person

Where do these rules come from?

openai.openapi#/paths/~1fine_tuning~1jobs/post: openai.openapi, retrieved 2026-09-09
openai.openapi#/paths/~1fine_tuning~1jobs/post/responses/200: openai.openapi, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/status: openai.openapi, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/fine_tuned_model: openai.openapi, retrieved 2026-09-09
openai.docs.fine_tuning#fine-tuning-jobs/the-status-of-a-fine-tuning-job: openai.docs.fine_tuning, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/id: openai.openapi, retrieved 2026-09-09
openai.openapi#/paths/~1fine_tuning~1jobs~1{fine_tuning_job_id}/get/responses/200: openai.openapi, retrieved 2026-09-09
openai.docs.fine_tuning#fine-tuning-jobs/the-members-that-a-running-job-leaves-empty: openai.docs.fine_tuning, retrieved 2026-09-09
openai.docs.fine_tuning#fine-tuning-jobs/the-fine-tuned-model: openai.docs.fine_tuning, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/training_file: openai.openapi, retrieved 2026-09-09
openai.openapi#/components/schemas/CreateFineTuningJobRequest/properties/training_file: openai.openapi, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/created_at: openai.openapi, retrieved 2026-09-09
openai.openapi#/components/schemas/FineTuningJob/properties/finished_at: openai.openapi, retrieved 2026-09-09
openai.events#/events/fine_tuning.job.succeeded: openai.events, retrieved 2026-09-09
openai.openapi#/components/schemas/WebhookFineTuningJobSucceeded/properties/data: openai.openapi, retrieved 2026-09-09
openai.docs.webhooks#webhooks/the-event-types: openai.docs.webhooks, retrieved 2026-09-09

Can openai.fine_tuning_job.succeeded return VERIFIED from the action response alone?

No. The minimum evidence level is E2. The action response is E1. The completion conditions read fine_tuning_job_readback and fine_tuning_job_succeeded_events.

What happens after the timeout?

The verdict is UNVERIFIABLE with the reason evidence_unavailable_before_timeout. The operation goes to a person for review.