What Is the Change Detection Task?
Change detection is the standard laboratory measure of visual working memory capacity. The rules are simple to state: a set of objects appears for a moment — colored shapes, scattered symbols — then disappears behind a brief blank or mask. When the display returns, one thing is different: an object swapped, added, or moved. Your job is to say what changed. The task gets hard fast because the mask deletes the visual transient — the flicker that would normally point your eyes straight at the change — so you have to find it by comparing the new scene against what you actually encoded, not by watching it happen.
The paradigm was codified by Luck and Vogel in 1997, who used it to estimate visual working memory at roughly four integrated objects — not four features, but four bound bundles of color-plus-shape. Alvarez and Cavanagh (2004) refined the estimate: capacity isn't a fixed slot count, it shrinks as objects get more complex. And the related change blindness literature (Rensink, O'Regan & Clark; Simons & Levin) showed observers can miss enormous changes when the transient is masked. As for training claims — the honest picture is the same as for most working-memory tasks: practice gains stay close to the practiced task, and claims of far transfer have mostly failed to replicate.
How to tell a faithful change-detection task
A few details separate a faithful implementation from a lookalike:
- A real mask or concealment interval. If objects simply swap in place with no blank or distractor field in between, the motion transient does the work for you and it's not a memory task at all. The concealment interval is the mechanism, not decoration — and it should lengthen as difficulty rises, since retention decays over the delay.
- Set sizes that respect the capacity wall. The literature anchor is ~4 objects. A serious progression starts below the wall and pushes toward the teens only at elite levels — not the other way around.
- Encoding time that scales with the set. Fixed short study windows collapse into guessing as the array grows; per-object study time should stay roughly flat. Moving objects need more encoding time than static ones, not less (Saiki & Holcombe, 2012).
- Per-item response time. If the response window shrinks while the search set grows, per-object search time collapses and the difficulty is compounding — a bug pattern, not a design.
- Distractor pressure. Similarity and density of the surrounding field, plus shrinking color palettes, are legitimate difficulty axes that mirror how the paradigm taxes object binding rather than raw speed.
Change detection in Avoken
Avoken ships a family of five change-detection variants. The core pair — What's Changed? and its moving-field sibling What's Changed? MV — run the classic swap: 4–11 objects on the field, one replaced during a concealment interval that lengthens from 0.5 to 1.2 seconds, study time of 3.0–4.75 seconds, and a response window of 5–12 seconds tuned to hold roughly constant per-object search time (~1 second each). The MV variant drifts the objects while hidden and slows movement into the range where multiple-object tracking stays feasible; on its first four levels the changed object also swaps color as a scaffolded cue. What's New? and What's New? MV invert the logic — objects are added during the mask (1–3 newcomers among 3–11 existing), and a single wrong tap ends the round. What's Moved? uses the displacement variant: 6–13 objects on a grid where exactly one relocates, with a longer retention interval (0.8–2.2 seconds) since position memory is more robust than feature binding. Across the family, color palettes narrow from 8 toward fewer colors so objects get harder to tell apart at the top — and every level step was audited against the published anchors (Luck & Vogel 1997; Alvarez & Cavanagh 2004; Alvarez & Franconeri 2007 for movement speed).
Avoken is a training app, not a treatment. We describe what the task actually asks, and we don't claim the practice travels. If the paradigm interests you, this is a faithful implementation of it.