Reflexable game benchmarks
These benchmarks describe games completed on Reflexable. They count attempts, not unique people, and are not population norms.
Current game samples
Find your place ↗Current scoring versions only. Snapshots refresh approximately hourly; each row shows its observation date in UTC. Earlier versions remain stored separately.
| Game / version | Eligible attempts | Excluded runs | Mean / median | Collection window |
|---|---|---|---|---|
| ReactVersion 1 · ms | 283 | 614 | 449.7 / 290 msProvisional sample | – Observed 05:29 PM UTC |
| RecallVersion 2 · levels | 157 | 0 | 4.9 / 5 levelsProvisional sample | – Observed 05:29 PM UTC |
| PrecisionVersion 2 · hits | 477 | 0 | 10.5 / 8 hitsProvisional sample | – Observed 05:29 PM UTC |
| Brain BlitzVersion 2 · questions | 142 | 0 | 2.6 / 1 questionsProvisional sample | – Observed 05:29 PM UTC |
| Photo FinishVersion 2 · levels | 234 | 0 | 1.6 / 0 levelsProvisional sample | – Observed 05:29 PM UTC |
| Missing LinkVersion 2 · levels | 83 | 0 | Collecting dataSummaries after 100 eligible attempts | – Observed 05:29 PM UTC |
| Track the PackVersion 5 · races | 96 | 0 | Collecting dataSummaries after 100 eligible attempts | – Observed 05:29 PM UTC |
| Replay BoothVersion 2 · rallies | 61 | 0 | Collecting dataSummaries after 100 eligible attempts | – Observed 05:29 PM UTC |
We hold mean and median summaries until 100 eligible attempts. That is a display policy, not a statistical guarantee. Even a larger sample can overrepresent frequent players. The interactive charts show smaller samples with an early-data notice.
What is collected?
When a supported game finishes and its submission succeeds, the score service records a random attempt identifier, the game, scoring version, score, eligibility, mode, and server receipt time. A repeated submission of the same attempt identifier is counted once. Starting a new completed run contributes another attempt.
The collection window begins with the first accepted submission in that game version, including any excluded React runs, and ends at the snapshot time. It is an observation window, not a claim that someone played continuously throughout it. Offline retries can arrive later than the game was played.
Which attempts count?
Completed low scores and mistakes count. Abandoned games do not. React requires five clean responses for its timing distribution; runs with false starts are retained separately and appear in the excluded column. Other games count valid completed runs, including a run ending before the first level is cleared.
The server recalculates scores from submitted game details and checks the format, version, and game rules. Signed attempt tickets, duplicate protection, and request limits reduce invalid submissions. These checks do not prove that a human played fairly. Results are not certified records.
Why scoring versions stay separate
A change in difficulty, lives, or scoring can change a distribution even if the players are the same. Each materially different ruleset gets a new comparison pool. Earlier point-based runs are preserved; they are not converted into survival levels or added to the new charts.
React compares five-round averages in milliseconds, where lower is better. Precision compares targets hit; Brain Blitz compares questions solved. Recall and Missing Link compare completed levels, Photo Finish cleared races, Track the Pack fully cleared races, and Replay Booth correct rallies. Higher is better in these seven games. These units cannot be summed into a meaningful overall intelligence score.
How the charts and percentages are calculated
The mean weights each exact score by the number of attempts at that score. The median is the middle recorded score, or the average of the two middle scores when the count is even. Every eligible attempt has equal weight. A player's tenth attempt counts just as much as their first.
The comparison separates attempts you beat, those tied with you, and those ahead of you. It uses exact scores, not the width of a chart bar. The displayed range adapts to the distribution; a final bar marked “+” includes that score and above. It is not a maximum possible score.
For example, if a result beats 60 of 100 attempts and ties 10, the display reports 60% beaten and 10% tied. It does not turn that into “better than 70% of people.” These are illustrative counts, not a Reflexable observation.
What these data cannot tell us
All devices and input methods share one pool. Display and input latency can influence a reaction-time score. We do not record age or a stable player identifier with attempts, so we cannot publish age norms, unique-player totals, device-specific benchmarks, or a study of individual improvement over time from this dataset.
Players choose whether and how often to play. That self-selection, repeated attempts, changing devices, and unverified browser input limit interpretation. An arcade distribution is not a representative sample of humanity. Read what a good reaction time means and how browser measurements work for more context.
Your personal history and the shared totals
Your own results are saved in this browser. Clearing site data can remove your personal history and chart marker. It does not remove attempts already recorded in the shared database, which are retained without an automatic expiry to build the all-time distributions.
No account, name, email, stable device identifier, or raw IP address is attached to a stored score. Temporary keyed address hashes are used separately for rate limiting. Website hosting and Google Analytics have their own data practices, explained in the privacy policy.
Our data, external research, and examples
The sample table above comes from Reflexable’s score database. References to published research or another website’s statistics are labeled and linked in our guides. Worked examples are labeled as illustrations. We do not substitute another site’s average for missing Reflexable data.
These summaries are not a downloadable research dataset. A future data release would need a defined collection window, frozen ruleset, documented exclusions, and a clear reuse license before we present it as one.