Verbal Memory Test
A word appears, and the decision is deliberately simple: mark it NEW if it has not appeared earlier in this run, or SEEN if it has. The challenge comes from the set you must keep updating. Every genuinely new word becomes part of the history, repeats can return after short or long gaps, and a familiar-looking word can tempt a false alarm even when it has not appeared.
What makes this different from recall
This game does not ask you to type a list from memory. It is a continuous recognition task: one item is visible, and you judge whether it belongs to the stream you have already experienced. That creates two different kinds of success. A repeat correctly marked SEEN is a hit. A first appearance correctly marked NEW is a correct rejection. It also creates two different mistakes. Marking a repeat NEW is a miss, while marking a truly new word SEEN is a false alarm. The game reports those response classes separately instead of hiding everything behind one number.
Classic 3 Lives
Classic begins with a small seen set and keeps growing. Each correct decision increases the run score. A wrong Seen/New decision costs one of three lives, and the run ends when all three are gone. The stream can extend to 240 planned trials, so a strong attempt is not forced to stop after a tiny sample. The same URL seed reconstructs the same word stream, including which items repeat and where they return.
Classic is useful when you want an open-ended personal best. A high score here simply means you correctly classified more words before losing the three lives. It is not converted into an IQ estimate, a population rank or a clinical memory label.
Focus 50
Focus removes the early life-based ending. It always presents fifty decisions, which makes session length stable and gives the response breakdown more context. Because every run has the same number of trials, you can compare your own accuracy, hit count, miss count and false alarms more directly than in an open-ended attempt. The deterministic generator begins with new items, then schedules a controlled mixture of first appearances and repeats.
Long Lag 72
Long Lag makes repeat distance a first-class part of the challenge. Repeated words are deliberately eligible only after a larger number of intervening trials. The engine stores each repeat's actual lag and the independent verifier recomputes whether the word really appeared earlier at that distance. This mode is not a claim that one lag number equals a memory capacity; it is simply a different stream design that reduces the usefulness of remembering only the last few words.
Sprint 60s
Sprint uses a sixty-second clock. The underlying word order is still deterministic. The timer decides how far through the prepared stream you reach, rather than changing the next word based on your speed. That matters for reproducibility: two devices opening the same seed begin from the same planned sequence even if touch, keyboard, display and reading speed produce different numbers of completed decisions.
Keyboard and touch
The two large response buttons are designed for mobile touch, mouse or trackpad. Keyboard users can press N for NEW and S for SEEN. Space or Enter starts a run when it is idle. The page invalidates a live attempt if the tab becomes hidden, because timing and continuity should not quietly include a background interruption.
A frozen 2,048-word bank
The runtime does not download a dictionary and does not call a paid language API. V0.109 derives a 2,048-word common-English bank from the project's already-frozen Word Grid lexical asset. That older asset already preserves the project's Wordnik list, FrequencyWords ordering and en_US Hunspell/SCOWL provenance boundary. The V0.109 generator verifies the source file hash before it produces the specialist core, then the release verifier checks that the 2,048 runtime entries are unique.
Using an existing frozen lexical boundary is important for two reasons. First, a seed can reproduce the same stream without depending on a remote service that may change. Second, the project does not need to invent or scrape a new word corpus for one game. The generated runtime bank remains part of the same local-first build.
How the stream is generated
Each mode has a trial count, a warm-up, a repeat target and repeat-lag rules. New words are drawn from a seeded shuffle of the frozen bank. When the schedule calls for a repeat, the generator chooses only from words that have actually appeared. It records the real lag since that word's most recent appearance. A NEW trial never silently reuses an earlier word. This lets the verifier reconstruct the entire state history and confirm that the label on every trial is truthful.
What the score does and does not mean
The page records correct decisions, total decisions, hits, misses, false alarms, correct rejections and average browser response time. Those are literal properties of the session. Classic also keeps the largest number of correct decisions reached before the third life is lost. Fixed modes keep local bests based on task accuracy. These values are stored only in this browser unless the user clears site data.
The page deliberately avoids fabricated percentiles and broad ability labels. Language background, vocabulary familiarity, reading comfort, interruptions, keyboard versus touch input and even how aggressively someone chooses SEEN can change the result. A single browser run is not a diagnosis, intelligence score or statement about school performance.
Why false alarms deserve their own count
Two attempts can have the same number of correct answers and still fail in different ways. One person may miss repeats but rarely call a new word familiar. Another may recognize repeats well yet press SEEN too freely, producing false alarms. Showing both makes the result more informative without pretending that the site can infer a deeper medical cause. It also gives a practical way to compare your own response style between sessions.
Deterministic Practice and Fresh Practice
A Practice URL contains a seed. Reopening the same mode with the same seed recreates the same planned words and repeat positions. Fresh Practice changes the seed and leaves the rules alone. That separation makes it possible to reproduce a bug, share a course, or compare code changes without turning every attempt into an uncontrolled random stream.
Privacy and local-first behavior
The game runs in the browser. The word bank, stream generator, scoring and personal records are local assets and local state. There is no account requirement for this preview and no remote leaderboard. The site is publicly available, so the current pages retain public indexing and the preview runtime remains search-blocked.
Related practice
Verbal Memory complements, rather than replaces, the existing Memory Grid. Number Memory asks you to reproduce a number after it disappears. Sequence Memory asks you to reproduce an ordered pattern. Chimp Test asks for remembered spatial locations in numerical order. Dual N-Back continuously updates positions and sounds. Verbal Memory is the missing old/new recognition loop: the item is visible, but you must remember whether its source belongs to this run.
Start a run
For the classic experience, open Classic 3 Lives. For a fixed comparison use Focus 50. If short-term familiarity feels too easy, Long Lag deliberately increases the repeat distance. Sprint 60s is for a faster session. All four use the same frozen bank and deterministic generation contract.