The science behind
every number.

Understand the response.

A neurological reflex is an automatic, involuntary response. A browser reaction task instead asks for a deliberate response to a visual cue. Noetryx draws on cognitive-science task concepts to practice response selection and visual attention; it does not test reflex arcs, measure brain activity or provide a neurological examination. NIH neuroscience background

A window into task performance.

Detecting a cue, choosing a response and withholding a response involve multiple sensory, cognitive and motor processes. Browser behavior does not isolate a brain region or diagnose a neurological condition. Our tasks draw on established experimental ideas; this particular implementation has not been independently validated.

Choice and response-control assessments allow 2,000 ms per cue. Correct responses below 100 ms are classified as anticipatory and excluded from reaction-time summaries. This is a protocol rule, not a biological threshold. At least five correct timed responses are required to save a session. A correct HOLD has no reaction time.

Use the same device, browser, input method and setup label. Switching tabs or losing window focus aborts an active task. We have not calibrated this implementation against laboratory timing hardware. Research on experiment software demonstrates why timing verification matters. [1]

Find one target.
Ignore the distractions.

Each 3 × 3 grid contains the digits 1–9 once. Read a new target number above each grid and select it within 3,000 ms. The target changes on every trial. Across 18 trials, each digit is a target twice and each position holds a target twice. Correct responses under 100 ms are classified as anticipatory. Accuracy uses all 18 trials; the median and interquartile range use correct timed responses only.

Missed deadlines, wrong targets and early responses are reported separately. You can complete the exercise without an account-storage connection; signing in is required to save results. This is practice in visual search and deliberate response, not proof of improved general concentration or a clinical attention assessment.

Adjust the challenge.
Keep assessment stable.

Practice uses an explicit between-session staircase. It is a rules-based controller, separate from the machine-learning model. The next window uses only your latest saved practice session for the same task, setup and input method.

1. Start with a 1,500 ms response window.
2. At 90% accuracy or above, shorten the next window by 100 ms. Response-control practice also requires a false-alarm rate of 10% or below on displayed HOLD cues.
3. Below 75% accuracy, extend the next window by 150 ms. Otherwise, keep it unchanged.
4. Keep the window within 400–2,000 ms. Fixed assessments always remain at 2,000 ms.

These are initial engineering choices, not an optimized treatment or a proven learning schedule. Harder practice is not itself evidence of better cognition.

A small model.
An inspectable decision.

Noetryx implements ridge regression in its own service code. It learns numerical weights from a person's comparable reaction-time history. Ridge regression is an established algorithm; we do not claim to have invented it or trained a neuroscience foundation model.

The model predicts a recorded task result. It does not calculate IQ, read thoughts, determine neurological health, or establish which intervention caused a change. Saved model weights and held-out predictions can be exported from the lab.

A faster task is not
a higher IQ score.

Practice can improve performance on a trained task. Whether benefits extend to untrained tasks, general intelligence or everyday performance is a separate empirical question. A large randomized study reported improvement on trained tasks without evidence of transfer to its untrained tasks. That result does not settle every possible intervention, but it prevents us from assuming broad benefit. [2]

Noetryx does not administer a standardized IQ test and has no evidence that it raises IQ. Its assessment ledger compares the first three and latest three fixed assessments for the same protocol, setup and input method, on separate date windows. It reports speed and accuracy changes together. That comparison is descriptive: practice effects, familiarity, sleep, motivation and other factors may explain a difference.

Expand the evidence
before the promise.

Working memory, sustained attention and cross-task learning are potential research directions. The separate Signal Engine processes supplied EEG files and has a secure live acquisition-bridge receiver. It calculates spectral measurements and experimental calibrated visual feedback. Physical-device integration and benefit still require validation. Behavioral task responses do not measure neural activity. Open the engine or read its signal protocol.

A separate public-data feasibility experiment explores post-match mental-load ratings using physiological summaries. It is not the personal reaction model, and it did not pass its predefined research gate.

Focus Grid protocol 1.1 records the changing target number. Older fixed-7 sessions retain protocol 1.0 and a separate v1 exercise label; new sessions use v2 and are not mixed with them. Current cognitive protocol 1.1 calculates false alarms using HOLD cues that were actually shown. Responses before a cue remain incorrect trials. Older 1.0 records retain their original scoring in the journal; the current practice rule and assessment ledger use 1.1 sessions only.

From observations to a plan.

Immediate feedback identifies a task-specific practice priority. In the Agent Lab, saved records drive a transparent baseline/practice/recheck workflow and a personal-model evaluation. These recommendations are product rules, not a validated training prescription.

Read a week in context.

The current window contains today and the previous six recorded dates; the comparison uses the seven dates before that. Your selected time zone determines today. Only current-protocol fixed sessions matching the latest task setup and input method are used. Manual imports, adaptive practice, future-dated entries and other setups are excluded.

Each window needs at least three sessions on two different dates. This is a product rule to avoid presenting a comparison from one session, not a validated statistical threshold. We compare the median of session response times and the mean session accuracy separately. Response-time change = (current median − previous median) / previous median × 100. Accuracy change is shown in percentage points.

All sessions receive equal weight. The report is descriptive; it contains no significance test or causal conclusion. Repeat exposure, device latency and other factors may explain a change. The time budget and chosen schedule are personal preferences, not a scientifically established training dose.

Understanding result comparisons

Every task result includes illustrative speed and accuracy examples. These examples are authored teaching aids, not population norms, validated thresholds or percentiles. No age-normal classification is provided. Simple reaction, visual search and choice tasks must not be compared as interchangeable tests.

Your personal reference becomes available after three comparable saved sessions and uses the median of up to five recent sessions. Matching includes task version, mode, response window, input method and setup where recorded. Accuracy is compared alongside speed. Smaller timing spread means more consistent responses within that session; it is not a clinical assessment.