Editorial assessment

Assessments turn your editorial rules into evidence-backed judgments. Supply the rules inline; they are frozen with the job. No customer-specific preset or profile setup is required.

Submit your criteria

Write each criterion as a requirement the footage should satisfy. pass means the evidence supports that requirement; fail means it contradicts it; uncertain means the evidence is insufficient.

curl https://api.story0.com/v1/assessments \
  -H "Authorization: Bearer $STORY0_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: editorial-storm-001" \
  -d '{
    "input": {"url": "https://dev.story0.com/footage/storm-rushes-720.mp4"},
    "playback_policy": "sound_off",
    "criteria": [
      {"key": "clear_action", "description": "The central action is understandable without hearing the audio.", "severity": "major", "remediable": false},
      {"key": "concise_opening", "description": "The clip begins with useful action rather than extended empty footage.", "severity": "minor", "remediable": true},
      {"key": "no_graphic_injury", "description": "The footage contains no visible graphic injury.", "severity": "fatal", "remediable": false}
    ],
    "categories": [{"key": "weather", "description": "Weather and its visible effects"}],
    "media_requirements": {"min_width": 640, "min_height": 360},
    "billing": {"max_charge": {"currency": "USD", "amount": "0.75"}}
  }'

Wait for the returned job, then retrieve GET /v1/assessments/{result_id}. Up to 30 criteria and 20 optional categories are supported. Criteria keys must be unique. A criterion description can contain 3–1,000 characters.

Disposition rules

The server derives disposition from validated criterion outcomes. It is not an unconstrained model verdict. These rules apply in order:

A failed deterministic media check changes the result to needs_review, unless it already rejects safety. A proposed usable_after_edit result without a validated edit suggestion also becomes needs_review.

Safety criteria always require human review. usable is a recommendation under the submitted criteria; it is not publication approval, a rights determination or a guarantee that no harmful content exists.

Inspect the result

criteria preserves each criterion's severity and remediability beside its outcome, reason, evidence and confidence. Confidence is explicitly kind: "model_estimate", calibrated: false; do not read it as a measured probability.

media_facts and media_checks contain deterministic properties such as measured dimensions and the presence of an audio track. analysis contains the underlying timed video observations. categories supplies optional routing hints. Keep these distinctions in your own interface.

edit_plan only contains validated cut candidates from the analysis. Candidates overlapping protected keep moments or the payoff plus its reaction interval are removed. The result does not render those cuts automatically. A producer can inspect the evidence, revise their plan, and compose it with the story workflow.

Keep policies reusable

Version criteria in your application and store your version in metadata. The result's config_hash identifies the frozen configuration. Editing your policy later does not change past results.

Use cheap screening to decide which clips merit assessment, or call assessment directly when every clip needs the fuller check. Each submitted operation is explicit, estimated and charge-capped. There is no hidden paid workflow triggered by reading a result.