## No Black-Box Grants: Ratify the Rules Before SPP Is Absorbed ### The missing delegation The Empowering the ENS Foundation temp check ([discuss.ens.domains](https://discuss.ens.domains/t/temp-check-next-era-of-ens-dao-empowering-the-ens-foundation/22175)) correctly identifies the absence of an accountability layer between ENS and the entities it funds. Its proposed solution, however, delegates more than grant administration. It delegates authority to determine which evidence matters, how quality is measured, which conflicts are acceptable, what constitutes failure, and when funding should continue. Those decisions shape the allocation of ENS capital as directly as custody rules do. If the rules are written later by the same institution that selects recipients, the DAO may retain formal removal authority while losing practical visibility into individual funding judgments. SPP3 offers a useful starting point. It introduced a published rubric, a named committee, structured interviews, and a consolidated recommendation. ENS has also tested frontier models as screening tools and found that they can identify missing budgets, unfalsifiable KPIs, undefined scope, weak repository evidence, superficial ENS alignment, and unresolved conflicts. The same experiment identified risks from prompt gaming and overly agreeable model behavior. The transition should preserve what worked and establish stronger guarantees where the present process remains dependent on discretion and self-reporting. ### Proposed amendment This proposal amends the Empowering the ENS Foundation temp check by conditioning absorption of the Service Provider Program into the Foundation’s grants program on DAO ratification of a Grants Charter. Before SPP is absorbed into the Foundation’s grants program, the Foundation shall submit a **Grants Charter** for DAO ratification. The Charter shall establish the following requirements. #### 1. Public decision rules The Foundation shall publish the grant mandate, eligibility conditions, evaluation dimensions, evidence requirements, conflict rules, funding bands, resubmission process, and grounds for appeal before an application period opens. The criteria should explain what constitutes credible evidence without revealing every operational test in advance. #### 2. Evidence-linked decision records Every material award or rejection shall produce a versioned decision record containing: - the applicable rubric and policy versions; - the evidence supporting each material finding; - committee participation, votes, and recusals; - unresolved uncertainty or disagreement; - the final human rationale; - any departure from the published process; - the disposition of an appeal or resubmission. The record need not disclose legally protected or security-sensitive information. Any redaction should identify its legal or operational basis. #### 3. Rules for AI-assisted evaluation AI systems may support eligibility review, evidence collection, inconsistency detection, comparative analysis, and drafting. They shall have no unilateral authority to approve, reject, suspend, or release grant funds. When AI materially informs a recommendation, the Foundation shall publish a versioned evaluator manifest identifying the models, prompts, retrieval sources, aggregation procedure, and human-review rules. Before applications close, the Foundation shall publish a cryptographic commitment to the evaluator manifest. After decisions are issued, the manifest shall be disclosed to the greatest extent compatible with applicant privacy and security. This commit–reveal structure gives applicants advance knowledge of the governing standards, prevents evaluators from silently changing the operational process after reviewing applications, and permits retrospective audit without publishing every test in time for applicants to optimize against it. Material disagreement among models or between a model and a human reviewer shall remain visible in the decision record. Consensus scores shall not erase uncertainty. #### 4. Verifiable delivery conditions Every grant above a DAO-defined threshold shall include a delivery manifest specifying: - milestones and acceptance criteria; - the evidence required to establish completion; - the party authorized to verify each milestone; - the funding tranche attached to each milestone; - remediation, suspension, and termination conditions. Experimental work may use learning milestones or decision gates where final outputs cannot responsibly be fixed in advance. The acceptance condition must still be observable. #### 5. Appeals and controlled evolution Applicants shall receive a defined process for correcting factual errors, contesting unsupported findings, and resubmitting after remediable deficiencies. The Foundation may update operational procedures between application rounds. Material changes to eligibility, conflicts, appeals, AI authority, or milestone enforcement shall require public notice and a DAO challenge period or ratification. The Grants Charter shall receive an annual public review based on applicant outcomes, delivery evidence, false-positive and false-negative findings, overrides, appeals, and observed attempts to game the process. ### Relationship to other amendments This proposal remains neutral on treasury custody and board composition. It strengthens either institutional outcome. A premise register can record the assumptions used in grant decisions. Cryptographic-accountability systems can anchor evaluator commitments and decision records. Tag-gated payment infrastructure can consume verified delivery states. Return-on-Governance reporting can aggregate the resulting outcomes. Those mechanisms govern institutional inputs, payment execution, or aggregate performance. The Grants Charter governs the adjudicative path connecting an application to a funding decision. ### Principal trade-off A more legible process creates administrative cost and can induce applicants to optimize against published criteria. The Charter should therefore apply proportionately. Small grants can use simplified records. Sensitive information can be protected. Experimental grants can use staged discovery. The normative criteria remain public, and the exact operational evaluator is committed before the deadline and disclosed afterward. The objective is to make discretion attributable, evidence reviewable, procedural changes detectable, and funded claims testable. ### Amendment text > This proposal amends the Empowering the ENS Foundation temp check. Before the Service Provider Program is absorbed into the ENS Foundation Grants Program, the Foundation shall submit a Grants Charter for DAO ratification. The Charter shall require public evaluation criteria, evidence-linked decision records, conflict and appeal procedures, and verifiable delivery conditions for material grants. Any AI system materially informing a grant recommendation shall remain advisory, operate under a versioned evaluator manifest committed before applications close, and preserve sufficient records for retrospective audit. No AI system shall exercise unilateral authority to approve, reject, suspend, or release grant funding. ### Requested allocation and deliverables **Maximum request: $1,000** The work is structured so that partial allocations retain independent value. - **First $200** Publish a draft ENS Grants Charter and a machine-readable grant decision-record schema. - **Next $250** Publish the evaluator-manifest and commit–reveal specification for AI-assisted screening. - **Next $300** Produce a retrospective test protocol and worked examples using publicly available ENS grant and SPP materials. - **Final $250** Publish an integrated implementation note, threat model, adoption pathway, and open-source reference repository. All artifacts will be vendor-neutral, publicly available, and designed for reuse by ENS or other public-goods grant programs. ### What would change my position I would withdraw or materially narrow this amendment if an existing binding Foundation policy already provides equivalent public criteria, evidence-linked decision records, evaluator-version commitments, applicant challenge rights, and milestone verification. I would also support a simpler mechanism if it achieves the same guarantees with materially lower administrative cost.

• ENS Grants Charter and machine-readable decision-record schema$200
• AI evaluator-manifest and commit–reveal specification$250
• Retrospective test protocol and worked examples using public ENS materials$300
• Implementation note, threat model, adoption pathway, and open-source reference repository$250