No Data Is Not Neutral
No data is not neutral
When a system cannot tell whether a person answered, a sensor reported, or a
workflow actually observed something, neutral is a lie.
Unknown is its own state. Treating absence as "fine", "average", or "zero risk"
fabricates confidence and pushes the product toward the wrong action.
The rule
Omit missing signals from the score. Renormalize over what is actually known.
Surface the missingness anywhere it changes confidence, explanation, or
recommendation.
Life Coach OS example
The 2026-06-20 life-coach-os readiness incident made the failure mode plain.
The app let untouched subjective sliders sit at the default 3/3, then scored
that as if the user had explicitly said they felt neutral. That fabricated "I
feel fine" signal helped prop up readiness while the same user had ACWR
2.21, Form (TSB) -70, and self-reported being tired.
The bug was not just a weighting mistake. It was a truth mistake. The system
collapsed two different states:
- the user said "I feel neutral"
- the user never answered
Those are not interchangeable. One is evidence. The other is missing evidence.
The shipped fix did the right thing: only count subjective axes the user
actually touched, omit untouched defaults, and renormalize the score over the
remaining signals.
Why this matters
Missingness is often directional information.
- An untouched form default means "unknown", not "average".
- A missing telemetry stream means "we cannot tell", not "nothing happened".
- A stale or absent sensor reading means "reduced confidence", not "safe".
- An imported historical artifact can prove provenance without proving live
current state.
If you flatten those cases into neutral, the product stops being conservative
where it should be and starts sounding more certain exactly when it knows less.
Design check
Before a score, badge, or recommendation uses a value, ask:
- was this observed?
- was it explicitly entered?
- was it inferred?
- was it merely defaulted?
If the answer is "defaulted" or "missing", the UI should say so and the scoring
path should treat it as unknown.