Proposal Due Diligence: From Claim to Evidence

Most bad reviews are lost in the first ten minutes. A representative opens a proposal, reads the summary the proposer wrote, absorbs its framing, and spends the next two hours arguing inside a frame someone else built. Intake is the discipline of establishing, before any evaluation, three things: what decision is actually in front of you, how much review that decision deserves, and whether you are the right person to be reviewing it without disclosure. None of this is about the merits. All of it determines whether your later work on the merits is worth anything.

• Restate the decision from the on-chain action and its metadata anchor rather than from the pitch.
• Set and state a review depth proportionate to the ask, the reversibility and the time available.
• Run a conflict check and record its result before reading in detail.
• Open a diligence record another person could continue without speaking to you.

Cardano governance actions are typed. The Developer Portal governance guide and the constitutional framework distinguish, among others, treasury withdrawals, parameter changes, updates to the constitutional committee, hard-fork initiations, new constitutions and info actions. The type is not decoration: it determines who else votes, what thresholds apply, whether the outcome is reversible, and what a yes actually authorises. A parameter change and an info action can be described in prose that sounds identical and mean entirely different things.

So the first artefact you open is the on-chain record, and the second is the metadata document at its anchor. CIP-100 gives that document a predictable shape: authorship, the body of the argument, references, and witnesses that tie claims to identities. Reading the anchored document rather than a forum summary matters for a mundane reason. The anchored text is what the action is committed to; a forum post, a Discord thread or a slide deck can say anything and be revised silently. When the anchored document and the public narrative diverge, that divergence is itself your first finding, and it is a neutral, checkable one.

Restate the ask in your own words in a single sentence with three parts: what is authorised, under what conditions, and to whom. If you cannot write that sentence from the anchored document, you have found a scoping problem before you have evaluated anything. Record it as such, in those terms, without suggesting anyone concealed it.

You cannot deep-review everything, and pretending otherwise produces the worst outcome: shallow reviews presented with the confidence of deep ones. Choose depth explicitly and say so in your record. A screening read confirms the type, the ask, the deadline and any obvious scope problem. A standard review adds the claim-evidence map and source verification. A deep review adds independent feasibility checks, delivery history and, where warranted, an expert consultation.

1. Size of the ask, relative to the treasury and to your own tolerance for error.
2. Reversibility: a parameter change can be changed back; funds already disbursed generally cannot.
3. Breadth of effect: does this touch one team, one sector, or every participant?
4. Time remaining before the decision date, honestly counted in hours you will actually have.
5. Whether anyone independent has already published a verifiable review you can check rather than repeat.

Writing the chosen depth into the record is what makes a light review honest instead of misleading. "Screening read only; I did not verify the delivery history" is a defensible statement. Silence, followed by a confident conclusion, is not.

The conflict check is run first because its purpose is to bind you before you know whether you like the proposal. Once you have formed a view, every relationship starts to look either irrelevant or damning depending on which way you are leaning. Ask a fixed list of questions and answer them in writing: do I hold a position that this outcome could move? Have I worked with, been funded by, or competed against anyone named? Have I taken a public position on this topic that I would find costly to abandon? Is a delegator of mine an interested party? Does anyone in the proposal have a personal relationship with me?

Most answers will be no, and the check takes two minutes. When one is yes, the response is graduated. Disclose in the record, and in your published rationale, in plain terms: what the interest is, when it began, and roughly its scale. Where the interest is material and direct, consider abstaining and say why. Recusal is not an admission of wrongdoing; it is a statement that the decision should not depend on your particular position. The Institute's disclosure guidance exists so that disclosure looks routine rather than confessional.

The intake note writes itself once you separate the two documents. Action type: treasury withdrawal, irreversible once disbursed. Ask, restated: authorise 310,000 ada to one named organisation, which is to run twelve workshops, publish a curriculum, and distribute an unspecified portion to unnamed regional partners. Decision date: nineteen days away. Depth: standard review, because the sum is significant and the outcome is irreversible, with the sub-grant pool flagged for closer attention. Divergence noted: the thread says "education across three regions"; the anchored document commits to twelve workshops and a curriculum. That is not a contradiction, but the anchored scope is narrower and it is the scope on record.

Conflict check: the reviewer once spoke, unpaid, at an event organised by one named regional partner, two years ago. That is minor and indirect, and it is disclosed in one sentence rather than agonised over. The review proceeds.

A reviewer opens the same proposal, reads the forum thread, decides within a minute that community education is exactly what the ecosystem needs, and spends two hours writing a supportive analysis of a scope the action does not contain. They never open the anchor. They do not notice the sub-grant pool. Their conflict check is never written because they never felt conflicted. The output is fluent, sincere and unusable: nobody can tell which document it reviewed.

The inverse failure is just as common. A reviewer distrusts the proposer, opens with a hunt for wrongdoing, and produces a document in which every ambiguity becomes evidence of intent. Both reviews skipped intake. Both reached a conclusion before establishing the question.

• Reviewing the summary rather than the anchored document, then quoting the summary as if it were the commitment.
• Treating every action type as if it were a funding decision, and applying budget reasoning to a parameter change.
• Choosing review depth implicitly by how interesting the topic is rather than by size, reversibility and time.
• Running the conflict check after forming a view, when it can only confirm what you already want.
• Recording a divergence between narrative and anchor as deception rather than as a divergence.
• Starting the record only once you have something conclusive to say, which guarantees the early reasoning is lost.

After intake you can state, from primary records, what action is proposed, what a yes authorises, when the decision falls, how deeply you intend to review it, and what interests of yours are on the table. You have one timestamped artefact that any delegator could read and any other reviewer could continue. None of it is a judgement about merit, and that is the point: it is the frame inside which judgement will later be defensible.

Intake tells you nothing about whether the claims are true, whether the team can deliver, whether the price is reasonable, or whether the sub-grant pool in the example is well governed. It does not tell you whether a narrower anchored scope was deliberate or careless - and you should not guess. It does not reveal conflicts held by other people, which you are not in a position to assert. Those questions belong to the next four lessons, and to the questions you will eventually publish.

Ten minutes of intake buys you a review of the actual decision. Skipping it buys you a confident review of something else.