GOVERNED RECOMMENDATION INTELLIGENCE
Make the model’s signal
visible and accountable.
An interactive demonstration of the accepted MIND-small engagement model. Adjust permitted context inputs and inspect the real frozen logistic model’s score—without a model API, user account, or production connection.
Decision support, not a decision. A predicted click is only an engagement signal. It is not evidence of helpfulness, satisfaction, spiritual benefit, or production recommendation effectiveness.
REAL MODEL OUTPUT
Estimated engagement score
—
Loading the frozen browser model…
This is a probability-like engagement score from the accepted model, not a threshold decision or a claim that the candidate will be clicked.
Local model contributors
Active standardized inputs × their learned coefficients. These are transparent model mechanics, not causal explanations.
Higher-score contributors
Lower-score contributors
MODEL CARD
Accepted model evidence
Reviewer-approved for capstone demonstration on August 27, 2026. The model is not deployed to Strongr Daily.
- Model
- SGDClassifier logistic model
(log loss) - Features
- 249 final features
- Train / dev
- 5.84M / 2.74M candidates
- Validation
- Chronological split
- Average Precision (AP)
- 0.106537
- ROC-AUC
- 0.716304
- Log loss
- 0.159305
- Ranking nDCG@10
- 0.366813
Average Precision summarizes precision-recall performance and is appropriate for this imbalanced engagement dataset. Accuracy is not the primary quality measure: positive engagement occurs in only about 4.06% of the official development set.
FUTURE STRONGR DAILY PROVING GROUND
From benchmark mechanics to consented, privacy-safe validation.
This interface validates only recommendation mechanics on public MIND-small interaction data. Any future Strongr Daily evaluation remains gated by reviewer approval, explicit consent, privacy-safe design, and separate production validation. No Strongr Daily data is connected to this demo.