Color Delayed Estimation (Visual Working Memory)
A set of coloured squares is shown briefly; after a delay one location is probed and the observer reproduces its colour on a continuous colour wheel. The feature is a hue ANGLE and the response is a point on a circle, so the outcome is a continuous circular ERROR — there is NO correct/incorrect. The headline analysis is not the mean error but a von Mises MIXTURE decomposition of the error distribution (Zhang & Luck 2008; Bays, Catalao & Husain 2009): a target-centred von Mises (memory, with a precision), a uniform component (guessing — the probability the item was not in memory), and a von Mises around the NON-target items (swaps / misbinding). This is the catalog's first family with no correctness; the contract drops `correct` and carries `responseErrorRad` and the non-target feature values, the latter so swaps can be modelled rather than misattributed to guessing.
- Defining contrast
guess-rate (mixture-decomposition)- Family
- continuous-report
- Human review
- sourced — unreviewed
- Executable test
- verify:task — an implementation can be driven and judged against this spec
- Schema
- spec version 9
Where this spec stands
This record is sourced. The ladder below is the human-review axis
only — a passing executable test never moves a record up it.
sourcedLiterature-linked, drafted with LLM assistance or by a contributor, and unreviewed. The honest default.corroboratedIndependently cross-checked against the cited sources, but not signed by a named expert.expert-annotatedAll three claims attested by named experts. Derived, never self-declared.contestedA named expert disputes at least one claim. A real outcome, kept visible.
Origin reference
- Wilken, P., & Ma, W. J. (2004). A detection theory account of change detection. Journal of Vision, 4(12), 11. doi:10.1167/4.12.11
Review this specification
Three separable claims: does the contrast identify the paradigm, is it a defensible measure of the named construct, and are the psychometric numbers correctly conditioned? Attest one, contest another — a split outcome is a real result.