Heart Typewriter · Exploratory research
We investigate presentiment and precognition through pre-stimulus and pre-reveal ECG modeling. An ECG is a recording of the heart’s electrical activity.
Question and ambition
We are exploring whether candidate physiological patterns can be reproduced under controlled conditions and eventually support interpretable feedback about intuitive processes.
The Heart Typewriter is a long-term vision for making physiological information more accessible. Future closed-ended and richer expressive interfaces remain aspirations that depend on validation—not capabilities demonstrated by current experiments.
What we have built
Our functioning research infrastructure covers ECG morphology, RR-interval features, dynamic features, physiological-state grouping, model-confidence filtering, repeated train/test evaluation, and shuffled-label controls. RR intervals are the time between successive heartbeats.
Research timeline
December 2025
Exploratory pre-stimulus ECG modeling produced tentative performance differences in some analyses. Paired comparisons with shuffled-label controls formed part of the evaluation. This required further validation; it was not a confirmed discovery of precognition.
After December 2025
Deep research largely paused, with follow-up experimentation conducted in June–July 2026. This was not a continuous program of active trials or recruitment.
June 19, 2026
An interim analysis evaluated 500 recordings from a roulette-based outcome task. Some selected models and physiological-state subsets showed improvements over shuffled controls.
July 9, 2026
The dataset grew to 1,000 recordings, including the original 500. Several average model comparisons improved, while the strongest state-specific patterns shifted. These were two analysis points, not independent replications or 1,500 separate recordings.
Current status
Independent validation is pending. There is no established general-purpose prediction tool or announced launch date.
June–July 2026
In June–July 2026, we evaluated ECG models at 500 and 1,000 recorded trials in an exploratory task involving concealed roulette outcomes. The later analysis included the original recordings. Several feature sets showed modest average performance improvements over shuffled-label controls. Results varied across physiological states, and the strongest subgroup findings did not establish a stable prediction method. These observations inform our research direction, with independent validation still needed to determine whether the differences reflect reproducible predictive information.
| Feature set and coverage | 500 recordings: observed / shuffled | 1,000 recordings: observed / shuffled |
|---|---|---|
| ECG morphology (C′), all trials | 48.1% / 50.6% | 51.8% / 50.4% |
| RR-interval features, all trials | 51.8% / 50.0% | 54.1% / 50.3% |
| RR-interval features, top 20% by confidence | 55.0% / 50.4% | 58.9% / 49.6% |
These are mean accuracies across 200 repeated train/test splits, with 20% held out per split. “Top 20%” means the model’s highest-confidence fifth of test predictions, not full coverage. Repeated splits reuse observations and do not establish independent replication or performance on a fresh prospective dataset. Ordinary paired tests over overlapping splits can overstate certainty, and exploring multiple feature sets, state groups, and thresholds affects interpretation.
Methods
Influences and reading
HeartMath’s intuition research, work associated with the Institute of Noetic Sciences and Dean Radin, and the broader presentiment literature influence the questions we investigate. These are intellectual influences—not affiliations, endorsements, or independent validation of our models. Reported findings and interpretations remain debated.
Possible applications
Spiritually oriented coaches who already use intuition in decisions are one possible early audience for reflective support—if the underlying signals can be validated. Longer term, this may intersect with Decision Intelligence: combining useful human information with analytical tools to support decisions. Both are application hypotheses, not demonstrated results.
We welcome conversations with researchers, technical collaborators, and thoughtful people following the question.
Connect About the Research