SLOP3255 · Week 09 · Evidence limits
Noise, gaps and uncertainty#
Clean sound can overstate certainty. An honest instrument preserves the limits of its evidence.
A value is not alone#
Rendering only 0.83 removes the information needed to judge how much confidence it deserves.
Missing is not zero#
| State | Meaning | Auditory treatment |
|---|---|---|
| 0 | a valid measured value | render using the normal rule |
| NA | no valid value exists | explicit gap marker |
| flagged | value exists but is suspect | retain value + quality cue |
| below limit | bounded non-detection | threshold cue + limit value |
Silence alone cannot distinguish these four states.
Make the quality pattern visible#
Colour and sound need a shared legend: gold = wider uncertainty; coral = missing or suspect.
Four uncertainty strategies#
| Strategy | Example | Best for |
|---|---|---|
| Boundary | low and high tones around estimate | reading an interval |
| Texture | more noise as confidence falls | continuous quality overview |
| Repetition | multiple samples from a distribution | variation and overlap |
| Layer | separate optional quality channel | expert inspection |
No strategy is inherently honest. Test whether the cue supports the intended uncertainty task.
Avoid the “bad data” instrument#
Punitive cue
Harsh distortion marks every uncertain value.
Uncertainty begins to sound like failure.Informative cue
A bounded, optional texture shows confidence without replacing the value.
Quality remains data, not judgement.The emotional character of a cue affects interpretation.
Test the uncertainty task#
A cue that merely sounds “fuzzy” has not yet passed a task.
Publish the clean and the complete view#
Estimate layer
the primary measured pattern
Quality layer
uncertainty, flags and limits
Data table
values, units and missing-state codes
The listener may simplify the experience without losing access to the evidence.
Listening Lab 09 · Output: uncertainty legend
Make uncertainty audible#
Task
Design and compare two cues for confidence intervals in a synthetic ten-point series.
Test
- locate widest interval
- find missing point
- rate confidence
Keep measured value and uncertainty in separable auditory dimensions.