A refusal means Deloosh could not release a trustworthy draft under the evidence and safety rules that applied to the request. The refusal rationale names the failed check so you can remove the claim, supply the missing proof, or correct the source material.
Refusals from the never-invent gate
The gate can return these exact reason codes and messages:
unsupported_performance_claim— “Performance/effectiveness claims require an explicit proof_asset claim.”unsupported_deliverability_fact— “Deliverability facts require an explicit proof_asset claim.”unknown_provenance_claim— “Draft provenance referenced a claim_id that was not present in the request context.”recipe_as_observed— “Only observed_artifact claims with claim_status=observed may be used as observed creative.”unknown_claim_used— “Unknown exemplar fields must remain coverage gaps, not generated copy or observed claims.”tier_c_effectiveness_claim— “Tier-C gallery exemplars are craft references only and cannot support effectiveness claims.”coverage_gap_not_preserved— “Draft provenance must preserve unknown fields and retrieval coverage gaps.”
For the two common claim classes, see DR-3, Performance claims need proof and DR-4, Deliverability facts need proof. DR-5, What is a proof asset and how do I add one? explains the evidence that can support them.
Other refusal paths
An image-only draft can be blocked; see DR-6, Image-heavy and image-only emails. A non-library or unsafe image is refused under the library-only guard; see DR-7, Why images come only from your brand library. If the drafting model fails to return a trustworthy result, use DR-8, Draft generation failed — what now?.
A red sending-health light stops drafting before content generation. See SEND-16, The sending-health light.
Edits have their own refusal surface: DR-11, “This block can’t take that change” covers block conflicts, and DR-12, Why was my edit blocked? covers safety findings on changed blocks.