Free AIRA Tool · 08
The 20-Minute AI Assumption Register
Before an AI-assisted plan becomes a decision, find the claims carrying its weight. Give each one evidence, an owner, and a small test.
01
Expose the assumptions
Identify the claims that would actually change the decision if they were wrong.
02
Attach evidence
Record the source, date, owner, and status for each claim.
03
Choose small tests
Check existing records or ask the accountable person before changing live work.
04
Write the decision note
Separate the safe next step from what must remain on hold.
The decision gap
A plausible plan can still depend on an unverified claim.
For example, a draft rollout plan may assume the team can export a particular data field, that a reviewer is available on Friday, or that a customer-facing change is reversible. The register makes those dependencies visible before anyone acts.
Find the load-bearing claims
Start with one real plan and its source material. Do not ask the model to improve the prose yet.
Prompt 01 · Find assumptions
Review this AI-assisted plan for a founder or operator. Do not rewrite it yet. Decision the plan supports: [DECISION] Plan: [PASTE] Known source material, with dates: [PASTE OR SAY NONE] List up to eight load-bearing assumptions. An assumption is load-bearing if changing it would change the decision, cost, timing, customer promise, or operating risk. Quote the exact plan passage, state the assumed fact in plain language, and explain which decision depends on it. Separate facts stated in supplied sources from inferences and unknowns. Do not invent a citation or a probability.
Build the evidence register
A source, date, and owner matter more than a persuasive explanation.
Prompt 02 · Attach evidence
Build an evidence register from the load-bearing assumptions below. Assumptions: [PASTE] Available sources with owner and date: [PASTE] For each assumption return: assumption, decision affected, source/evidence link or NONE, source date, evidence status (verified / partial / unverified / contradicted), what would falsify it, and the person who can verify it. A confident sentence is not evidence. If a source is stale for the decision, mark it partial and explain why.
Test the decision edges
Use the smallest safe check that could change the decision.
Prompt 03 · Choose tests
Choose the smallest safe test for this assumption register. Register: [PASTE] Decision deadline: [DATE] Available people and systems: [LIST] For the five assumptions most likely to change the decision, give one test each: question, existing record or person to consult, owner, timebox, pass/fail observation, and what to do if evidence remains missing. Prefer checking an existing record or asking the accountable owner before running an experiment. Do not recommend a customer-facing, financial, or irreversible action as an automatic test.
Write the decision note
Keep a hold condition visible when a critical claim is still open.
Prompt 04 · Decision note
Create a one-page decision note from this assumption register and test results. Register and tests: [PASTE] Return: decision being considered; what is verified; what remains unverified or contradicted; the five most consequential assumptions and owners; safe next step; explicit hold condition; review date. Do not silently fill a gap. If a load-bearing assumption is still unverified, recommend a bounded next step rather than treating the plan as approved.
What to take away
A usable register has no blank owner for a consequential claim. If evidence is missing, the outcome is not “approved by AI”; it is a bounded next check and a clear hold condition. Review the register whenever the plan, source, or decision deadline changes.
Need this in a live workflow?
AIRA can build the evidence and review path into the process.
We connect the real source of truth, accountable owners, and a review checkpoint before a plan moves into execution.
Explore custom workflow builds