Build an Evidence Boundary

Learning objective

By the end of this lesson, you can separate facts, interpretations, forecasts, assumptions and unknowns, then prioritise evidence gaps by their ability to change the decision.

Why it matters

Evidence-led does not mean adding more pages to a pre-read. A large evidence pack can still hide weak attribution, inconsistent definitions and missing decision-critical facts.

The objective is to make the burden of proof and the limits of knowledge inspectable.

Core concept: classify the claim before debating it

Use five claim types:

• Fact: an observation with a source, definition and date.
• Interpretation: an explanation of what an observation means.
• Forecast: a conditional statement about a future result.
• Assumption: a proposition used without sufficient current evidence.
• Unknown: information the team knows it does not have.

Then test decision sensitivity:

If this claim changed, could the leading option, scale, timing or confidence change?

High-sensitivity items belong in the live work. Low-sensitivity details can remain in an appendix. Confidence and sensitivity are different: weak evidence can still be crucial when the decision depends heavily on it.

Visual explanation

Text alternative: facts, interpretations, forecasts, assumptions and unknowns are positioned by evidence confidence and decision sensitivity; high-sensitivity uncertainty becomes a live evidence gate, while low-sensitivity detail moves to context or later work.

Worked example: NordForge

NordForge reports that a prior pilot improved performance by 17%. During the same period, the area received a new leader, lower product complexity and an additional planner.

“Performance improved by 17%” can be a fact if its definition and period are valid. “The pilot caused the improvement” is an interpretation with serious limitations.

The enterprise option also assumes 95% utilisation while current utilisation is 74%. Because this assumption changes expected value, it receives a visible evidence gate rather than a footnote.

Design a one-page evidence pack

Include:

1. the decision question and criteria;
2. no more than five material claims;
3. the source, definition, date and claim type for each;
4. confidence and limitation;
5. unknowns most likely to change the choice;
6. an owner and date for every gap that must be closed.

Comparable options need comparable evidence. A preferred option should not receive detailed benefits while alternatives receive only risks. Missing data must remain marked as missing; an average score is not a neutral substitute.

Common mistake

Mistake: debating whether a number is “right” before agreeing what it measures.

Check the definition, denominator, period, source and comparison first. Precision cannot repair a category error.

Quick check

A vendor forecast is low confidence but could reverse the investment decision. Where does it belong?

A. Remove it because confidence is low.
B. Treat it as a fact.
C. Make it a visible assumption or forecast and define the evidence needed to test it.
D. Replace it with the midpoint of the range.

Answer: C. Low confidence increases the need for an explicit evidence boundary when sensitivity is high.

Practical prompt

Open the Evidence Register. Record five claims for your session, label their type and rank the unknowns by decision sensitivity—not ease of collection.

Summary

• Evidence volume is not evidence quality.
• Label facts, interpretations, forecasts, assumptions and unknowns.
• Keep source, definition and date attached to the claim.
• Give all options a comparable burden of proof.
• Turn sensitive unknowns into finite evidence gates.

Sources

• NASA, Systems Engineering Handbook (opens in a new tab), Decision Analysis Process.
• GOV.UK Service Manual, Researching in small group workshops (opens in a new tab).

Next lesson

Continue with Design Participation, Not Attendance: bring authority, unique information, execution knowledge and affected perspectives into the process without confusing contribution with a vote.