RESEARCH / EXPERIMENT 001
Real data.
Every result visible.
Building our own intelligence starts with a question we can test. This is our first public-data feasibility benchmark, with its sources, method and limitations open to inspection.
COMPLETED · EXPLORATORYDid not meet the internal screening rule
This experiment does not support a model-advantage claim. The result is retained so the next research decision can be based on evidence.
THE QUESTION
Can physiology summaries
estimate reported mental load?
We used heart-rate, skin-conductance, muscle-activity and facial-temperature summaries to estimate a participant’s post-match mental-load rating. A fixed ridge-regression model was compared with a simple training-set median.
This is a retrospective estimate of a self-report. It does not demonstrate real-time brain sensing, improved gaming performance or effective coaching. No EEG data were used in this experiment.
PRIMARY EVALUATION / HELD-OUT PLAYERS
The actual results.
Lower mean absolute error is better. Errors are in units of the dataset’s mental-load rating.
1.181
Average of held-out participant errors
1.283
Training-fold median rating
8.0%
A negative reduction means the model was worse
Keep the test honest.
Each primary fold holds out one participant. Imputation and scaling are fitted only on the training records. A separate leave-one-match-out evaluation produced model MAE 1.260 versus baseline MAE 1.291.
The exploratory 95% paired group-bootstrap interval for model-minus-baseline error is [-0.195, 0.001]. The internal gate required at least a 10% MAE reduction and an entirely negative interval. This is a research screening rule, not clinical validation.
Inspect the participant-level results
| ANONYMOUS PARTICIPANT | RECORDS | MODEL MAE | BASELINE MAE |
|---|---|---|---|
| amateurs/player_0 | 11 | 1.081 | 1.364 |
| amateurs/player_1 | 11 | 1.035 | 1.273 |
| amateurs/player_2 | 11 | 1.604 | 1.909 |
| amateurs/player_3 | 11 | 1.112 | 1.273 |
| amateurs/player_4 | 11 | 1.421 | 1.545 |
| pros/player_0 | 10 | 0.976 | 1.100 |
| pros/player_1 | 11 | 1.320 | 1.091 |
| pros/player_2 | 11 | 1.171 | 1.182 |
| pros/player_3 | 11 | 0.944 | 0.818 |
| pros/player_4 | 11 | 1.142 | 1.273 |
PROVENANCE
Credit the source.
Show the method.
Third-party dataset: Smerdov, Zhou, Lukowicz and Somov, Collection and Validation of Psychophysiological Data from Professional and Amateur Players: a Multimodal eSports Dataset (2020).
Source data are published under CC BY 4.0. Noetryx created the processing code, feature summaries and experiment. The original dataset is not a Noetryx-owned or exclusive dataset.
View source dataset Read source paperInspect the evidence.
Full benchmark results JSONPrespecified protocol MDDerived research features CSVFitted research model JSONSource versions & checksums JSONSource data license TXTSource commit: 4cd8d63a30ea97614d023fa707089645108545be
WHAT THIS DOES NOT ESTABLISH
A first experiment.
A clear next standard.
Independent validationNot completed
Both tests reuse a small public cohort. Held-out players can share match conditions, and held-out matches can contain previously seen players.
Prospective benefitNot completed
A controlled, prespecified study must test whether an intervention improves a meaningful outcome. Estimating a self-report is a different question.
Our own consented cohortNot collected
Noetryx needs appropriately permissioned recordings with synchronized gameplay, device details and independently defined outcomes.
Physiological productsResearch & hardware required
EEG sensing, neurofeedback and neuro-cognitive scores are not provided by this experiment.
Build the next piece
from real measurements.
Your private workspace starts empty. Record a measured session and build your own history.