How can I separate facts from assumptions before comparing options?

Use four labels, not two

Start by writing every statement that affects the comparison. Do not force all of them into fact or assumption. Use four labels: evidence-backed input, assumption, value, and constraint. A value states what matters; a constraint states a boundary an option must respect. Neither becomes a fact merely because it feels firm.

For this low-stakes worksheet, call an input evidence-backed only when another person could inspect its source, observation, or record and see when it applied. The label is time-bounded and context-bounded. A written price can change, one observation may not transfer, and a reliable general source may not describe your exact situation. Record those limits beside the evidence rather than upgrading it to certainty.

Build the LENS statement ledger

LENS is a Forever Free Compass framework, not an external research finding. Give each decision input one row and complete the four fields before calculating a score or naming a preferred option.

1. Literal statement: write one claim without combining an observation, prediction, and preference in the same sentence.
2. Evidence anchor: name the inspectable source or observation, its date, and the limit on what it supports.
3. Not-yet-known link: state what must still be forecast, inferred, transferred from another context, or supplied by another person.
4. Sensitivity to the choice: record whether a credible change would alter the ranking, cross a constraint, or change the acceptable next step.

Classify each input before comparing options

A useful label describes how the statement should be handled, not how confident you feel about it. Keep the original wording visible so a vague impression cannot quietly become an evidence-backed input later in the comparison.

Anchor evidence without overstating it

GovS 010 says analysis should use data from a range of sources while recognizing limitations and uncertainty. It also identifies a log of data and assumptions as a proportionate form of documentation. Adapted here, that means an evidence row should show where the input came from and an assumption row should remain explicitly marked instead of being absorbed into the same number or sentence.

Ask three questions of an evidence anchor: What exactly was observed or recorded? When and in what context did it apply? Which part of my statement does it not establish? If a calendar shows two free hours next Tuesday, that supports availability for that period; it does not establish that the same capacity will exist every week or that the activity will remain worthwhile.

Expose the unknown link in each assumption

The 2026 Green Book describes uncertainty as assumptions that are not fully known and notes that weak support for a rationale can sometimes be improved with research, earlier evaluations, or pilots. Its setting is public appraisal, but the distinction transfers carefully: a forecast, a causal explanation, and a belief that evidence from another context applies here should not be presented as direct observations.

Rewrite an assumption as: “For this comparison, I am treating ___ as true even though ___ is not yet known.” The second blank names the missing link. “This option takes three hours” may contain a recorded session length, a forecast about future sessions, and an assumption that setup time stays constant. Split those into separate rows before comparing the option with another one.

Challenge the statement, not the person

The Government Data Quality Hub warns that assumption-led service design can produce the wrong design, unreliable data, and unnecessary burden, and recommends reframing statements around current evidence about user needs. For a personal comparison, use that as a prompt to ask what evidence is missing and whose perspective is absent, without treating the person who voiced the assumption as the problem.

Some assumptions concern another person's time, agreement, preferences, or behavior. Do not convert those into facts by confidence or repetition. Ask them when appropriate, respect consent and privacy, and keep the row unresolved when the information is not yours to obtain.

Verify only what could change the decision

The 2026 Test and Learn guidance recommends focusing on assumptions that are important to the intended outcome and weakly supported, using targeted and proportionate methods rather than testing everything equally. In the LENS ledger, sensitivity provides that filter: if a credible change would reverse the ranking, cross a real boundary, or alter the next action, improve the evidence or shrink the commitment before relying on the result.

If changing the assumption across a credible range leaves the next action unchanged, record the uncertainty and move on. More information has a cost. The goal is not to eliminate every unknown; it is to keep the comparison honest about which inputs are observed, which are chosen, and which are still doing speculative work.

Raise the standard for consequential choices

LENS is for structuring low-stakes reflection. It does not establish legal facts, medical facts, financial suitability, workplace obligations, safety, or another person's consent. It also cannot show that one event caused another or that a small observation will generalize.

When a decision affects health care, legal rights, regulated finances, employment duties, immediate safety, confidentiality, or material consequences for another person, use reliable domain records and qualified guidance. Do not average away a binding rule or safeguard because an assumed benefit gives one option a higher score.