Five free browser games that measure how fast you notice a rule has changed. Built on reversal learning — the same paradigm researchers use to probe cognitive flexibility — with spatial memory included as a control.
Pick a door. One of them holds the reward — work out which. Fourteen trials. Nothing will tell you if anything changes.
Free · No account · Nothing leaves your device · Updated 31 July 2026
KooDeck measures cognitive flexibility: how quickly you abandon a strategy once it stops working. You learn that one action earns a reward, the rule silently reverses, and the games count how many trials you need to notice and how often you keep picking the dead option.
That second number — the repeat of a choice that has already stopped paying — is called a perseverative error, and it is the measurement the whole battery is built around. Getting a rule right is easy. Letting go of one is the hard part, and it is the part that reversal learning isolates.
A fifth game measures spatial memory instead, and it is deliberately kept out of the flexibility score. Reporting both separately is the entire design, for reasons in the next section.
A 2026 Texas A&M study in Nature Communications reported that in 5xFAD Alzheimer's mouse models, reversal learning broke down before spatial memory deficits could be detected. The animals kept repeating an action that no longer paid, while still navigating and remembering locations normally.
The brain signature the researchers described was overactivity in the medial prefrontal cortex spreading into the striatum, alongside quieter cholinergic interneurons — the loop that lets you change your mind.
Five games per level, played in order: a reaction-time baseline, three flexibility tasks, and one spatial-memory control. Level 1 takes about nine minutes with two choices throughout. Level 2 takes about twelve and uses three choices everywhere, plus a shot clock and reverse recall.
Tap the port that lights up. Not scored into the index — it exists so a slow tapper isn't mistaken for an inflexible thinker.
Signal Storm: three ports, a one-second window, and no-go traps you have to not tap.
One door always pays. Six correct in a row and the rule silently flips. Three reversals. Measures recovery speed and perseverative errors.
Three Vaults: three vaults, five reversals, and a streak multiplier riding on it.
The good port pays four times in five. A single miss proves nothing, so this separates genuine flexibility from twitchiness.
Drift Storm: three ports at 75/25 and a 2.6-second clock on every choice.
Sort cards by a rule nobody tells you. Novel cards arrive with the same rule, then the rule moves to a different feature entirely.
Rule Storm: three bins, three features, and the correct bin lights up when you miss.
Watch a path of tiles flash, then tap it back. Spatial memory — the measure the research says should hold steady. Never folded into the flexibility score.
Path Storm: a 5×5 grid, and the path comes back backwards.
Run a simulated player — adaptive, average, perseverative or over-switching — through a whole level in about fifteen seconds, at the speed you choose. Useful for seeing the paradigm without playing it.
Two indices that are never averaged together. The Flexibility Index weights four subscores: reversal adaptation 24%, perseveration resistance 24%, evidence handling 34%, set-shift efficiency 18%. The Spatial Memory Index is scored on its own axis, and the report shows the gap between them.
Two decisions in there are worth stating plainly, because they are where naive scoring goes wrong:
A single index still cannot tell stuck apart from twitchy, so the report carries a separate response-style flag instead of pretending one number covers it.
Four simulated player profiles were run through 16 to 24 complete batteries each, per level. The perseverative profile is parameterised to reproduce the pattern from the research — learning rate crushed, perseveration high, spatial span left intact — and the scoring recovers it.
| Simulated profile | Level | Flexibility | Spatial memory | Gap | Style flag |
|---|---|---|---|---|---|
| Adaptive | 1 | 77 | 81 | +4 | measured |
| Adaptive | 2 | 75 | 80 | +5 | measured |
| Average | 1 | 61 | 65 | +4 | measured |
| Average | 2 | 63 | 62 | −1 | measured |
| Over-switching | 1 | 72 | 56 | −16 | over-switching |
| Over-switching | 2 | 67 | 62 | −5 | over-switching |
| Perseverative | 1 | 29 | 74 | +45 | over-holding |
| Perseverative | 2 | 28 | 68 | +40 | over-holding |
The two highlighted rows are the design working: flexibility near 28 while spatial memory sits near 70. Level 2's three-choice format also widens the gap between the adaptive and over-switching profiles from 5 points to 8, because a random switcher can no longer stumble onto the criterion.
These are simulated agents, not people. They demonstrate that the scoring behaves sensibly across known-different players; they say nothing about human performance. Data licensed CC BY 4.0.
It is not a medical test. KooDeck cannot detect, screen for, or rule out Alzheimer's disease, dementia, or any other condition. No score it produces means anything about your health.
The reference values behind the scores are illustrative anchors chosen so the scale behaves sensibly — not validated clinical norms. There is no normative sample, no age stratification, and no evidence of clinical validity.
Scores swing with sleep, caffeine, distraction, screen size and practice. A single run tells you very little; that is true of the real instruments too.
If you or someone close to you has noticed real changes in planning, adapting or memory in daily life, that is a conversation for a doctor — not for a game.
Reversal learning is a task where you first learn that one action earns a reward, and then the rule silently changes so a different action pays instead. Nobody tells you it changed. The measure is how many trials you need to notice and switch, and how often you keep choosing the option that stopped working.
It has been a standard probe of cognitive flexibility in both animal and human research for decades, because it separates learning a rule from letting one go.
No. KooDeck cannot detect, screen for, or rule out Alzheimer's disease, dementia, or any other condition, and no score it produces means anything about your health. It is an educational game built on a research paradigm, using illustrative scoring anchors rather than validated clinical norms. Concerns about memory or thinking belong with a doctor.
This is not legal hedging. There is no normative sample behind these numbers, so a low score is as likely to mean you were tired as anything else.
Not necessarily. A 2026 Texas A&M study in Nature Communications reported that in 5xFAD mouse models, cognitive flexibility measured by reversal learning declined before spatial memory deficits were detectable. That finding is in animal models and has not been shown to transfer to people, so it is a research direction rather than established fact.
In clinical practice, memory complaints remain the most common presenting sign, but Alzheimer's can also begin with changes in language, visual processing, or executive function.
Cognitive flexibility is the executive function that lets you change strategy when circumstances change: taking a detour when a road is closed, or adapting when job responsibilities shift. It is distinct from memory. You can remember a route perfectly and still struggle to abandon it once it stops working.
In a reversal, the relevant feature stays the same and only the correct answer flips: colour still decides, but the other colour now wins. In a set shift, the relevant feature itself changes: colour stops mattering and shape starts. Set shifts are usually more costly, and KooDeck measures the gap between them.
That gap is the classic extradimensional shift cost. Rule Rush produces it by introducing cards you have never seen at each shift, so an intradimensional shift stays cheap while an extradimensional one is expensive.
A perseverative error is choosing an option that has already stopped paying. KooDeck counts the run of consecutive re-picks of the dead option straight after a rule change, closed by your first correct response. Roughly one to three per reversal is typical; long runs are what the reversal paradigm is designed to expose.
Level 1 takes about nine minutes for five games, and Level 2 about twelve. You can play a single game on its own in one to three minutes. There is also a demo mode that runs a simulated player through an entire level in roughly fifteen seconds.
You can pause between trials at any point, and the games can be played in any order once the one before is done.
It is free with no account, no sign-up, and no email. It is a single HTML file that runs entirely in your browser, so it also works offline once loaded. There are no ads, no payments, and no premium tier.
No. Results live in your browser's memory for the length of the session and are never transmitted. Closing the tab erases everything. If you want to keep a run, the report has a download button that saves your trial-by-trial data as a JSON file to your own device.
There is no analytics script, no cookie, and no server to send anything to. The optional Support button loads Stripe's secure checkout only if you click it - nothing loads otherwise.
It is a 0 to 100 composite of four measures: how fast you recover after a rule change, how rarely you repeat a dead option, how well you use noisy feedback, and what a set shift costs you. It is benchmarked against illustrative reference values, not clinical norms, so treat it as a game score.
The report shows all four subscores separately, which is more informative than the composite.
Because their separation is the point. The research that prompted this build found flexibility declining while spatial memory stayed intact, so averaging them into one number would hide the only interesting pattern. Flexibility and spatial memory are scored on independent axes and reported with the gap between them.
Level 1 uses two choices throughout. Level 2 uses three everywhere, which drops chance from 50 to 33 percent, plus a 2.6 second shot clock, no-go traps, a streak multiplier, five reversals and reverse recall. Each level is scored against its own reference values so the numbers stay comparable.
Level 2 also measures better: with three options a random switcher can no longer hit the criterion by luck, so flexibility and impulsivity separate more cleanly.
You will get better at these specific tasks with practice. Whether that transfers to everyday flexibility is genuinely unresolved, and the evidence for far transfer from commercial brain training is weak. KooDeck includes a training mode, but it makes no claim that improving your score changes anything outside the game.
Anyone curious about how reversal learning works, students and educators who want a hands-on demonstration of an executive-function paradigm, and developers who want an open, documented implementation to adapt. It is not built for clinical use and should not be used to assess anyone.
There is a larger-text mode, a sound toggle, a pause, keyboard-navigable controls with visible focus, and reduced-motion support. Colour is never the only cue: in the set-shifting game each colour also carries a texture, so the task stays solvable with any colour-vision deficiency.
All text meets WCAG AA contrast on the dark background, and the layout adapts to landscape and short viewports.
Talk to a doctor, and do not use a game score to reassure or alarm yourself. Real concern usually comes from changes noticed in daily life by you or people close to you. A clinician can assess that properly with validated instruments and the rest of your history.
About nine minutes. Nothing to install, nothing to sign up for, nothing saved.
Play KooDeck