If you've read how FlowState works, you know the basics: an earpiece reads your brain, an app detects when your focus drops, and the earpiece plays audio pulses to nudge it back. The system stops automatically when your brain recovers.
This post is about why we built it that way. Every design decision traces back to specific research findings, and the pattern they point to is consistent: the features that make interventions work in studies are the features that consumer products leave out.
The three things that matter
The research points to three design principles that separate effective cognitive interventions from noise. FlowState is built around all three.
Closed-loop beats open-loop. A closed-loop system measures your brain, acts on what it finds, and adjusts based on the result. An open-loop system runs on a schedule regardless of what your brain is doing. The difference matters because your brain isn't in the same state all day. Playing focus audio when you're already focused is wasted effort. Playing it too late means you've already lost the thread. Timing the intervention to actual neural events is a fundamentally different approach than running on a clock.
Personalization beats one-size-fits-all. People's brains differ. Your baseline theta power, your optimal entrainment frequency, the times of day when your focus is strongest: these vary substantially from person to person. An intervention tuned to your brain produces effects that a generic setting does not. The most striking demonstration of this comes from a 2019 study in Nature Neuroscience by Reinhart and Nguyen, where stimulation was tuned to each participant's individual brain rhythm. The personalization was part of why it worked.
Measurement validates the intervention. If you don't measure the brain, you can't know whether the intervention did anything. You're relying on subjective feeling, which is unreliable for something as subtle as a shift in neural oscillation frequency. Measurement closes the loop and gives you actual data: did the rhythms change? How quickly? For how long?
What the key studies found
Two published studies underpin the design most directly.
Reinhart and Nguyen (2019) worked with older adults whose working memory had declined. They measured each person's brain rhythms, identified the specific theta frequency and the synchronization pattern between brain regions that had weakened, and then used electrical stimulation tuned to each person's rhythm to restore it. After 25 minutes of personalized stimulation, older adults performed at the same level as younger adults on a working memory task. The effect persisted for at least 50 minutes after the stimulation ended.
The study was sham-controlled, which means some participants received fake stimulation and didn't improve. The personalization was central: the stimulation was calibrated to each individual's brain, not applied at a generic frequency.
Woods and colleagues (2024) took a different approach. They embedded amplitude modulations into music at specific rates and measured both brain coupling and attention performance. Modulations in the beta range (around 16 Hz) increased neural synchronization to the audio stimulus and improved sustained attention on a validated cognitive task. This was also sham-controlled. It provides the most rigorous evidence that audio (not electrical current) can drive meaningful neural coupling and behavioral improvement.
These studies complement each other. Reinhart showed that personalized, brain-responsive stimulation produces real cognitive effects. Woods showed that audio modulation specifically can drive neural coupling and improve attention. Neither study did both at once. The architecture that combines measurement, personalization, and audio delivery in a single closed-loop system is what FlowState is testing.
For more on the underlying neuroscience, we've written detailed posts on theta rhythms and memory, beta rhythms and attention, and what flow state actually is.
Why consumer products haven't done this
The reason is straightforward: it's hard. Building a system that reads EEG in real time from a comfortable wearable, processes the signal on a phone with low enough latency to be useful, and delivers a calibrated audio response requires solving problems across hardware, signal processing, firmware, and software simultaneously.
Most products take a shortcut. Meditation apps skip measurement entirely. Consumer EEG headbands measure but don't intervene. Audio entrainment products play fixed-rate audio without knowing your brain state. Each approach captures part of the picture but misses the combination that the research says matters.
The cost of those shortcuts shows up in the evidence base. The consumer neurofeedback and audio entrainment literature is full of inconsistent results. A 2020 paper in Nature Neuroscience by Donoghue and colleagues showed that a large portion of reported EEG band power changes actually reflect shifts in background noise rather than genuine changes in brain oscillations. If you're comparing your raw EEG power to a population average, the number is nearly uninterpretable for any individual. Comparing against your own baseline under consistent conditions is what makes the measurement meaningful.
That's why FlowState calibrates to you. Your baseline, your frequency, your thresholds. The numbers mean something because they're relative to your own brain, not someone else's.
What our pilot shows (and what it doesn't)
Our pilot study across 15 participants recorded 139 consecutive intervention episodes. All 139 self-terminated when the system detected neural recovery. The mean intervention duration was roughly five seconds.
This confirms the detection-and-response loop: the system reliably identifies focus drops, delivers an intervention, and correctly detects when the brain has recovered. That's the engineering claim, and the data supports it.
What the pilot does not show is whether the audio intervention itself caused the recovery, or whether the brain would have recovered on its own in the same timeframe. Proving that requires a sham-controlled study, where some sessions deliver real audio and others deliver a placebo, and you compare outcomes. That study is planned. Until it's complete, the honest framing is: the system detects and responds correctly. Whether the response is what causes the improvement is the next question to answer.
The bet we're making
The bet is specific. Audio modulation can drive neural coupling (Woods showed this). Personalized, brain-responsive intervention produces real cognitive effects (Reinhart showed this). Combining both in a closed-loop consumer device should produce better outcomes than any open-loop, generic-frequency product on the market.
We think the ceiling for this category is higher than existing products demonstrate. The tools that will raise it are the ones actually measuring what they claim to be affecting.
References
Donoghue, T., Haller, M., Peterson, E. J., et al. (2020). Parameterizing neural power spectra into periodic and aperiodic components. Nature Neuroscience, 23, 1655–1665.
Reinhart, R. M. G., & Nguyen, J. A. (2019). Working memory revived in older adults by synchronizing rhythmic brain circuits. Nature Neuroscience, 22(5), 820–827.
Woods, K. J. P., Sampaio, G., James, T., et al. (2024). Rapid modulation in music supports attention in listeners with attentional difficulties. Communications Biology, 7, 1376.