Independent research

Research

Do features derived from noninvasive neural recordings time-locked to errors add predictive information beyond behavior?

Research horizon

Two levels of the question.

Long-term question

Could an adaptive policy using recording-derived features improve independent retention or transfer relative to a behavior-only policy?

That intervention question would be tested only if a separate predictive study first established held-out incremental value beyond behavior.

First tractable question

Do those features predict whether a person will later produce the correct response, beyond what behavior already predicts?

Rationale

Why this question

The applied systems work and this research direction share a measurement question: supported performance is observable, but durable independent capability must be tested rather than assumed. Learning Without Gatekeepers presents the institutional argument; the published Real Merit Protocol distinguishes observations, derived measures, inferences, and validation records. A null result remains informative.

First proposed test

Project Zero

Project Zero remains planned. No dataset has been selected, and no protocol, repository, report, or result exists.

Question: Do features derived from noninvasive neural recordings time-locked to errors predict a later correct response beyond behavior?
Behavioral baseline: Required before testing whether recording-derived features add predictive information; the model and evaluation plan are not yet specified.
Incremental-value test: Would adding features derived from the selected recordings improve held-out prediction relative to the behavior-only baseline?
First step: Select a suitable public dataset, identify a reproduction target, and publish a protocol specifying the outcome, behavior-only baseline, splits, leakage controls, and stopping rule.

Epistemic discipline

Evidence chain

Raw measurement
Event alignment
Derived feature
Held-out prediction
Comparison with behavior-only baseline

These are distinct analytical stages, and work may stop at any stage. A measurement does not establish a construct; held-out predictive value does not establish a causal effect. An intervention study, if warranted, would be a separate test.

Revision conditions

What would weaken the idea

Failure to reproduce the selected target effect under prespecified conditions.
No held-out predictive improvement over behavior.
Apparent prediction explained by motor artifacts, task timing, person identity, or leakage.
Failure to generalize across sessions or people.
No advantage for a policy informed by recording-derived features over a behavior-only policy.
Improvement only in supported performance, not later independent retention or transfer.

Scientific and ethical boundaries

What the measurements can—and cannot—mean

Scalp EEG records voltage differences at electrodes arising from aggregate neural activity and non-neural sources. It does not provide direct access to thoughts or uniquely measure a cognitive construct.
Heart rate, PPG, breathing, EDA, pupil response, blink rate, facial movement, and posture are physiological or behavioral context signals. They are not direct brain signals.
A measurement is not automatically a construct, and an association is not automatically useful prediction.
Prediction is not automatically causal control.
Immediate assisted performance is not automatically durable learning.
A model using features derived from noninvasive neural recordings must add value beyond appropriate behavioral and task-context baselines.
Animal mechanisms can motivate a human hypothesis but cannot automatically validate a human EEG interpretation or educational intervention.
Human research requires the applicable ethics determination or approval before recruitment and data collection.
Negative and inconclusive results remain reportable outcomes when the methods are adequate.
No clinical, diagnostic, medical-device, or high-stakes decision claim is proposed here.

Collaboration

Technical critique and collaboration

I welcome substantive critique and collaboration from researchers working in systems neuroscience; cerebellar and cortical prediction research; human EEG and BCI; computational modeling; learning science; research methods and statistics; and neuroethics.

Discuss the research