Evidence threshold

When more than 40% of the weighted score rests on factors with no evidence behind them, Verity publishes no band at all.

Why a fallback is not a safe default

The no_input fallback scores 3. That sits below the top of every factor's observed range — so deleting a merchant's data strictly improved their score.

This is not hypothetical. Testing moved an entity from 3.24 / high / recommend reject to 2.86 / medium / no recommendation purely by deleting its registered activities. Missing data was rewarded.

The obvious repair — raise the fallback to the worst value — is equally wrong in the other direction: it punishes merchants for checks we have not run. Neither number is right, because the question "how risky is this merchant" has no answer when most of the input is missing.

So the honest response is to stop producing a band.

What counts as unevidenced

Both cases where the score came from the fallback rather than from evidence:

• The observation was absent — nothing to read
• The observation was present but matched no configured row, so it took the fallback anyway

The second matters more than it sounds. Reading a country from the registry and having no configured score for it is not evidence-based scoring, however real the observation was.

What happens at the threshold

The breakdown	Written and readable — it is the most useful thing on the page
The score	Computed and stored
The band	Not published
The entity	Keeps whatever it had, including "never assessed"
A gate floor	Still applies — evidence in its own right

The entity page states the score with the caveat attached, so nobody can band it themselves off a figure the assessment declined to band:

Scored 2.86 on a 1–5 scale, but no band was assigned: too much of that score rests on factors with no evidence behind them. 6 of 8 factors had no evidence behind them, which is 62% of the weight.

Reaching the threshold in practice

Of the 21 points in the default factor set, 4 can never be evidenced until a compliance team fills in the country risk and legal form tables — both ship empty on purpose. That leaves a maximum evidenced weight of 17 of 21, or a floor of about 19% unevidenced for even a fully checked merchant.

A merchant with commercial registration, screening, identity and portfolio linkage on file lands comfortably under 40%. A bare entity with only a registration date does not, and correctly gets no band.

Changing the threshold

Risk::Assessment::INSUFFICIENT_EVIDENCE_THRESHOLD. It is a constant rather than a setting because moving it changes what past assessments would have said; the stored insufficient_evidence flag is recorded per assessment so an old one keeps its verdict either way.