| Criterion | Weight | What good evidence looks like | What weak evidence or an adverse finding looks like | Practical test |
|---|---|---|---|---|
| 1. Relevant demand and procurement route | 20 | Dated demand matching modality, topic, use and delivery; accountable purchasing route; limitations visible | Broad AI market claims, unrelated logos, no buying process | Ask what the next procurement decision is, who owns it and which requirement your archive satisfies |
| 2. Comparable execution and receipts | 15 | Permissioned comparable transaction evidence and a reference; collected cash distinguished from booked revenue | Unrelated deal, invoice only, unsupported pipeline amount | Trace one permitted example from engagement through delivery to remittance |
| 3. Rights discipline and use boundaries | 15 | Explicit use schedule, cohort exclusions, recipient/sublicensing controls and documented legal responsibility | “All AI uses” unexplained; assumed rights; impossible assurances | Walk one article/image from source agreement through proposed grant |
| 4. Net economics and fee transparency | 12 | Defined fee base, deductions, recoupment, taxes, costs and dates; reproducible calculation | Percentage without denominator; discretionary charges; headline guarantee without conditions | Recalculate a sample statement and a no-sale scenario |
| 5. Exclusivity, performance and exit | 10 | Limited scope and term; measurable obligations; practical release, renewal and survival provisions | Broad lock-in without outcomes, indefinite tail, unexplained retained rights | Test a missed target and termination scenario against the draft |
| 6. Sample security and downstream controls | 10 | Approved recipients, minimum sample, logged access, retention conditions and incident process | Uncontrolled bulk request, unclear recipients or undocumented subcontractors | Demonstrate the proposed release process using non-sensitive test material |
| 7. Delivery and preparation capability | 7 | Buyer-related specification, pilot quality measures, realistic update duties, accountable owners | Unpriced transformation, generic capability deck, no acceptance criteria | Run a representative agreed test and record exceptions |
| 8. Reporting, remittance and dispute handling | 5 | Report fields match contract; finance can reconcile amounts and pursue exceptions | Aggregated totals with no calculation trail or deadlines | Reconcile one reporting period or agreed synthetic example |
| 9. Identity, stability and conflicts | 4 | Current contracting entity, ownership, responsibilities and incentives disclosed; relevant checks completed | Brand ambiguity, undisclosed affiliate, conflicting representation or fee incentives | Match proposal, contract and payment details; document conflicts and mitigations |
| 10. Governance and decision usability | 2 | Named contacts, documented escalation, accessible evidence and change notices | Oral commitments, shifting scope or no accountable lead | Ask who decides when rights scope or requirements change |
| Total | 100 |
The concentrated demand and execution weights reflect this desk's partner-selection problem. A content owner can reasonably adopt different weights, but must freeze and explain them before comparing bids.
NIST AI 600-1 discusses third-party diligence and testing. This table is an original licensing framework, not NIST scoring.