Transparent methodology

How Sarenica works

Every core metric has a documented method. This page explains the parts that matter most — how signals are collected, how patterns are derived, and when advanced insights unlock.

Principles

Four principles behind every metric

Everything else follows from these. They are also what separates Sarenica from black-box scoring tools.

Deterministic

Reproducible results using documented algorithms.

Pattern-based

Detects patterns in your own behaviour, not generic benchmarks.

Quality-adjusted

Sensor confidence penalties for more accurate readings.

Individual

Compare you to yourself over time, not population averages.

Fatigue state

Fatigue state detection

The base signal most other metrics are derived from. Classified from four input streams with a confidence score attached to every reading.

Data sources
  • • Blink rate from the camera (blinks per minute)
  • • Microsleep events (eyes closed more than 500 ms while active)
  • • Typing rhythm (KPM, WPM, error-rate variability)
  • • Mouse movement patterns (speed, distance, idle time)
Low fatigue — normal blink rate, consistent typing, active mouse.
Normal fatigue — slightly elevated blink rate, minor rhythm changes.
High fatigue — more microsleeps, typing errors, slower mouse.
Very high fatigue — frequent microsleeps, erratic patterns, prolonged idle.
Confidence scoring

Each reading carries a confidence percentage based on sensor quality. Camera FPS above 15 and face-detection ratio above 80% score normally; below those thresholds a 0–50% penalty is applied, and readings below 50% confidence are flagged unreliable.

Hydrogen method

Burnout risk scoring

A weighted combination of seven factors Hydrogen computes from your own history — never a comparison to other users.

Factor
22%
Current fatigue level

Your fatigue state right now (Low / Normal / High / Very High).

13%
60-minute trend

Direction of fatigue change in the last hour.

18%
Accumulated high-fatigue minutes

Total minutes at High or Very High fatigue today.

18%
Session duration

Continuous work time without breaks (over two hours is a risk.)

9%
Time-of-day effects

Post-lunch dip (1–3 PM) and late evening (after 9 PM).

10%
Multi-day fatigue streak

Consecutive days with elevated fatigue burden.

10%
Signal quality dampening

Reduces score reliability when sensor confidence is low.

Risk levels
Low (0–30): Normal fatigue, no intervention needed.
Moderate (31–60): Elevated fatigue, consider a break.
High (61–80): Significant burnout risk, take a break soon.
Critical (81–100): Immediate break recommended.
Hydrogen method

Best work hours detection

Ranks 3-hour windows by lowest fatigue and highest concentration, weighted toward recent days.

Algorithm
  1. 1. Analyse hourly fatigue averages across 7–14 days of data.
  2. 2. Calculate concentration scores (inverse of fatigue plus productivity).
  3. 3. Identify 3-hour windows with lowest fatigue and highest concentration.
  4. 4. Weight recent days more heavily (exponential decay, 0.95^days_ago).
  5. 5. Rank windows by combined score.
Requirements
  • • Minimum 240 minutes tracked (maturity Level 2)
  • • At least 2 different days of data
  • • Coverage across multiple hours of the day
Output example
🥇 9:00 AM – 12:00 PM (Score 87/100) — lowest fatigue, highest focus.
🥈 2:00 PM – 5:00 PM (Score 72/100) — moderate fatigue, good productivity.
🥉 7:00 PM – 10:00 PM (Score 65/100) — evening energy spike.
Hydrogen method

Personal factor correlations

Pearson correlation with significance testing across your tracked factors. Reports confounders so you can interpret the number honestly.

What is analysed
  • • Sleep hours (from wearable) vs. next-day fatigue
  • • Keyboard activity intensity vs. fatigue accumulation
  • • Mouse activity patterns vs. concentration
  • • Session length vs. post-session fatigue
  • • Time-of-day effects on your specific patterns
Statistical method

Pearson correlation coefficient (r) with significance testing:

  • r > 0.5: strong positive correlation
  • r < −0.5: strong negative correlation
  • p < 0.05: statistically significant
  • • Minimum 7 days of data required for reliability
Confounding factors reported
  • • Weekend vs. weekday differences
  • • Day-of-week effects
  • • Seasonal variations (when enough data exists)
Unlocking

Data maturity levels

Advanced insights only appear when there is enough data to make them statistically valid. Showing best-hours from one hour of tracking would be misleading.

Level 0 — Bootstrap
0 min, 0 days

Just started. Basic tracking only: /now, /today.

Level 1 — Learning
60 min, 1 day

Learning your patterns. Unlocks /why_fatigue.

Level 2 — Emerging
240 min, 2 days

Patterns emerging. Unlocks /best_window.

Level 3 — Established
600 min, 5 days

Solid data. Unlocks /patterns, /report, /compare.

Level 4 — Mature
1,200 min, 14 days

Expert mode. Unlocks /correlations and full research capabilities.

Promise

Our transparency promise

Every Hydrogen command accepts a /methodology flag that returns the coverage, confidence, and technical details used to produce that answer.

  • All algorithms are deterministic and reproducible
  • Quality adjustments are transparent and documented
  • Statistical methods include confidence intervals and significance tests
  • Confounding factors are identified and reported
  • Updates to methodologies are versioned and announced
FAQ

Methodology FAQ

Quick answers to the questions we hear most.

Does Sarenica use black-box AI to calculate fatigue metrics?

No. Core fatigue and pattern metrics are calculated using deterministic, documented methods. The AI layer helps with interpretation, question understanding, and response presentation, not raw metric calculation.

Why are some advanced insights locked until I collect more data?

Sarenica uses a data maturity system so advanced insights are only shown when there is enough data to reduce misleading conclusions. This improves reliability for comparisons and correlations.

How does Sarenica handle low-quality or incomplete signals?

Sarenica applies confidence scoring and quality penalties to sensor-derived signals. Low-confidence readings are flagged and can be excluded from reliable-minute calculations and advanced analysis.

Can I inspect the logic behind a specific Hydrogen agent answer?

Yes. Sarenica provides methodology explanations, coverage details, confidence cues, and technical details on demand so you can understand how a result was produced.