AI systems
14
running daily
Run time
10AM
Pacific daily
Character level
loading…
Correlations
pairs tracked
Intelligence Layer surfaces patterns → Discoveries →

This platform isn't just data collection — it's an AI system that reasons about the data every morning. Here's every intelligence feature currently running, with live examples from this week.

What the daily email actually looks like

Sample Daily Brief

Live daily brief — loads from platform API.

Subscribe to get the weekly version →

How the daily pipeline works

From raw data to Daily Brief

Every morning at 10am PT, a sequence of compute Lambdas runs against the last 24 hours of ingest data. They pre-compute scores, run correlation refreshes, and then hand everything to Claude — which synthesises it into a structured daily coaching brief delivered by email.

What Claude reads each morning
  • Body weight & 7-day trend (Withings)
  • HRV + resting heart rate (Whoop)
  • Recovery score + strain (Whoop)
  • Sleep duration, stages & efficiency (Whoop)
  • Fasting glucose + overnight CGM curve (Dexcom Stelo)
  • Habit completion vs. Tier 0 targets (Habitify)
  • Journal sentiment score (computed nightly)
  • Training load: CTL, ATL, TSB (Garmin + Strava)
  • Blood pressure (daily cuff measurement)
  • Calorie deficit vs. TDEE (MacroFactor)
  • Protein intake vs. target (MacroFactor)
  • Adaptive mode classification (Flourishing / Baseline / Struggling)
  • Top 3 active correlations from the 90-day window
  • Character sheet level + pillar breakdown
What Claude outputs
  • 3 priorities — the three highest-leverage actions for the day, ranked by the character engine signal
  • Anomaly flags — any metric that deviates more than 1.5 SD from the personal baseline
  • Recovery guidance — whether to train, recover, or adapt based on HRV + strain signals
  • Nutrition note — deficit pace check, protein gap, and any glucose pattern worth noting
  • Coaching insight — one data-backed observation from the week's patterns
  • Morale note — a single sentence calibrated to the adaptive mode (tone shifts when struggling)
  • Keystone habit reminder — whichever Tier 0 habit most predicts today's score
Pipeline sequence: Ingest Lambdas (06:45–09:00 PT) → Compute Lambdas: character-sheet, adaptive-mode, daily-metrics-compute, daily-insight-compute, hypothesis-engine (09:00–10:00 PT) → Claude synthesis via site-api Lambda → Daily Brief email delivered 11:00 PT
All 14 systems

What runs every day

Live from today's data
01 / 14
Correlation Engine
Finds statistical relationships between any two metrics. Currently tracking 23 pairs across sleep, recovery, training, nutrition, and body composition.
Top correlation this week
Loading from API…
Illustrative example
02 / 14
Benjamini-Hochberg FDR
Corrects for false discoveries when testing multiple hypotheses simultaneously. Prevents the AI from getting excited about noise when 23 pairs are tested at α=0.05.
Example correction
23 pairs × α=0.05 → 1.15 expected false positives without correction. BH correction applied: only pairs where p ≤ (i/23 × 0.05) are flagged significant.
Live from today's data
03 / 14
Adaptive Mode
Classifies each day as Flourishing, Struggling, or Baseline based on combined metric signals. Changes the tone and recommendations of the daily brief to match your actual state.
Current mode
Loading from API…
Live from today's data
04 / 14
Character Engine
7-pillar scoring system covering sleep, recovery, training, nutrition, metabolic health, lab results, and habits. EMA smoothing prevents single-day noise from crashing your level.
Current character stats
Loading from API…
Live from today's data
05 / 14
Hypothesis Engine
Generates testable hypotheses from data patterns and tracks them automatically. Each hypothesis has a protocol, measurement window, and success criteria baked in from the start.
Active experiments
Loading from API…
Illustrative example
06 / 14
Decision Fatigue Signal
Detects when decision quality typically degrades based on day-of-week patterns and habit completion rates. Helps pre-plan high-willpower moments.
Example signal
Example signal: The system might detect that certain days/times have higher habit dropout rates — flagging them as high-risk windows for weekly planning.
Illustrative example
07 / 14
Metabolic Adaptation Detector
Catches weight loss stalls before they become demoralising. Distinguishes true metabolic plateaus from water weight and glycogen fluctuation using a rolling 14-day window.
Plateau detection logic
If 14-day variance > 1.2 lbs and net trend = 0, flagged as "fluctuation." If variance ≤ 0.8 lbs and net trend = 0 for 10+ days, flagged as "true plateau — consider protocol adjustment."

Preliminary pattern — requires 90+ days for validation

Illustrative example
08 / 14
Sleep Architecture Analysis
Cross-references Whoop sleep stages with HRV the next morning. Identifies the personal deep sleep threshold above which next-day recovery reliably scores green.
Example
The system will cross-reference deep sleep duration with morning HRV to find your personal threshold — the deep sleep minimum that predicts good recovery.

Threshold will emerge as data volume grows

Live from today's data
09 / 14
Training Load Intelligence
CTL/ATL/TSB tracking. Calculates fitness, fatigue, and form as independent signals. Identifies when training load is approaching injury risk territory before symptoms appear.
Current training state
Loading from API…
Illustrative example
10 / 14
Keystone Habit Detector
Identifies which single habit most predicts overall daily score. Varies by person, week, and phase of journey. Updated monthly with new 90-day window data.
Current keystone habit
Weight logging → +8.3 pts average daily grade on days it's completed. Days without weight logging: average grade drops 11.2 pts regardless of other habits.
Illustrative example
11 / 14
Glucose Pattern Analysis
CGM data cross-referenced with meal timing, sleep, and exercise windows. Finds personal response patterns that generic nutrition advice misses entirely.
Identified pattern
Carbs post-8pm: avg +24 mg/dL spike, slow 3.5hr return. Carbs pre-8pm: avg +11 mg/dL spike, rapid 1.8hr return. Evening carbs = 2.2× the glycemic impact for this physiology.
Illustrative example
12 / 14
Biomarker Trajectory
Tracks lab values over time against optimal ranges, not just "normal." Currently tracking 7 biomarkers across 3 blood draws. Projects trajectory toward or away from optimal.
Tracked biomarkers
HbA1c, fasting glucose, triglycerides, HDL, LDL particle size, hsCRP, testosterone — 7 markers × 3 draws = 21 data points. All trending toward optimal range.

"Optimal" = Attia longevity ranges, not standard lab-reference normal

Illustrative example
13 / 14
Weekly Correlation Refresh
Every Sunday, re-runs all 23 correlation pairs with the most recent 90-day window. Correlations evolve as the data matures — an early-journey signal may disappear at month six.
Refresh schedule
EventBridge cron: Sunday 14:00 UTC. Runs Pearson + BH-FDR correction on all 23 pairs. Results written to DynamoDB and surfaced in /explorer. History retained for drift detection.
Live from today's data
14 / 14
N=1 Insight Generation
The daily brief includes 1–2 data-backed insights specific to the last 7 days of patterns. Generated fresh each morning from the full context of all 14 systems combined.
Example insight from this week
Your last 5 high-HRV mornings all followed days where step count exceeded 9,000 and final meal was before 7:30pm. That combination now has a 71% predictive rate in your dataset.
These systems run every morning at 10am PT

Ask them anything.

All 14 systems feed into a single AI you can query directly. What does your sleep data predict about tomorrow? Which habits matter most this week?

Ask the AI Correlation Explorer

Get the weekly version of what these systems produce