Every AI decision, on the record.

Output: Top-K ranked list of brand fits

Reason factors: Sport overlap, audience demo overlap, content category match, brand trust score

Confidence: Cosine similarity 0-1 + bucketed confidence (low/med/high)

Audit trail: Every match logged with the athlete-vector + brand-vector hash

Output: Plain-language summary + flagged clauses

Reason factors: Perpetuity, exclusivity, morality, IP grab, payment terms, termination, indemnification

Confidence: Deterministic pattern match + LLM second pass with disagreement flag

Audit trail: Original text + rule hits + LLM response stored for the athlete to download

Output: Dollar range + tier

Reason factors: Followers × engagement × sport multiplier × region multiplier × performance multiplier

Confidence: Each multiplier shown; user can override any input

Audit trail: Snapshot of all inputs saved with the estimate

Output: Visible / held / blocked

Reason factors: Pattern hits (scams, predatory tactics, sexualized content) + LLM classifier

Confidence: Pattern weights + LLM probability with disagreement flag

Audit trail: Public transparency log at /transparency

Output: 0-100

Reason factors: EIN verification, business age, principals' check, deal history, dispute rate, social proof

Confidence: Each component contribution shown

Audit trail: Brand can view + dispute their own score

Output: Likelihood the deal closes

Reason factors: Comparable deals, athlete tier, brand tier, brief alignment, season timing

Confidence: Bucketed (low/med/high)

Audit trail: Every prediction logged for backtest