Purpose and fit

Use this when a proposal sounds plausible but depends on statements such as “people will use it”, “the sources are complete” or “review will be quick”. Build an evidence plan around the assumptions that could overturn the proposal. This is earlier-stage decision work, before a full workflow evaluation.

The outcome is a short list of questions, checks and decision consequences. Avoid collecting convenient evidence that leaves the decisive uncertainty untouched.

Before you start

Provide the proposal or a short description of it, the next decision and any existing observations or constraints. No test results or prepared dataset are needed to design the evidence plan. Share authorised, anonymised excerpts and adapt the prompt to the proposal information you actually have.

Expect an assumption register and a feasible investigation plan. It should distinguish evidence from hypotheses, prioritise uncertainties that can change the next commitment, and leave unperformed checks and results blank.

Required inputs

  • The proposal, intended beneficiary, expected outcome and next commitment under consideration.
  • A description of how the proposed change is expected to produce that outcome.
  • Existing observations and their source, date, relevance and limitations.
  • Available people, approved data, time and boundaries for low-risk investigation.

Practical workflow

  1. Write the proposed chain of change: the intervention changes a task, the task improves an outcome and someone benefits. Underline every link that is assumed rather than observed.
  2. Turn each assumption into a specific question. Replace “users will trust it” with an observable question about whether intended users can identify the sources and accept an output under stated conditions.
  3. Prioritise by decision impact and uncertainty. Ask what would happen if an assumption were false. Investigate weak assumptions that would change the next commitment before polishing peripheral estimates.
  4. Choose the smallest appropriate check for each priority question. Options include an authorised source audit, an interview about a recent work item, a manual rehearsal or a bounded comparison using sanitised examples.
  5. Before checking, record what finding would support, weaken or leave the assumption unresolved. State the sample or observation scope and the limitations; avoid inventing numerical thresholds merely to make the plan look precise.
  6. Review the evidence with its owner. Update the proposal, gather more evidence, reduce scope or stop. Preserve mixed findings and record which decision actually changed.

Expected output

An assumption register and a short decision update. Copy this record for each important assumption:

Assumption: [specific statement] Why the proposal depends on it: [causal link] Current evidence: [source and limitation] Decision impact if false: [consequence] Check: [method and authorised scope] Owner: [person/role] Supporting finding: [observable condition] Weakening finding: [observable condition] Unresolved result: [what it would mean] Evidence collected: [result and reference] Decision change: [continue/change/stop/investigate] Review date: [date]

Hypothetical example: a proposal assumes that policy questions can be answered from one approved document set. A source audit finds that several common questions require local exceptions. The team narrows the proposed task before testing answer quality.

Human checks

The sponsor checks that a result could genuinely change the decision. Domain colleagues challenge whether the proposed observation addresses the assumption. Have someone who is not advocating for the proposal review contrary evidence. Keep participants’ identities and sensitive content out of AI inputs unless that use is authorised.

Limitations

A passed check increases confidence only within its scope. Interviews describe people’s reports, which may differ from later behaviour. Small or convenient samples can miss important variation. The register cannot replace an evaluation of the complete workflow when the proposal reaches that stage.

Starter prompt

Examine this proposal using only the sanitised material provided: [proposal, change logic, evidence and constraints]. The next decision is [specific commitment]. Extract the assumptions linking the proposed change to the intended outcome and distinguish them from supported observations.

Prioritise assumptions that are uncertain and could change that decision. Create a register with these fields: assumption; why the proposal depends on it; current evidence and limitations; decision impact if false; proposed check and authorised scope; owner; supporting finding; weakening finding; unresolved result and meaning; evidence collected; decision change; review date. Leave uncollected evidence blank. Suggest a proportionate check for each priority assumption and flag confirmation bias and missing perspectives. Do not manufacture evidence, claim validation, contact participants or execute a test.

Adapt the requested fields to the authorised inputs actually supplied. If the change logic is incomplete, label the proposed links as assumptions and ask what decision they affect. Retain mixed or contradictory evidence and specify what a check could leave unresolved; do not fabricate findings or probability scores. State the expected output and acceptance criteria before drafting. Ask for essential missing information; where a useful partial draft is possible, label its limits and assumptions. Do not invent facts, evidence, results or approvals.

Data, model variability and your responsibility

Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.