How tracking works in Iris — from "I had a headache" to structured data you can analyze

You say "had a terrible headache after lunch, maybe a 6/10." Iris responds with something helpful. But behind the scenes, something else happened: a specialist agent turned your sentence into structured data — headache, severity 6/10, time approximately noon, possible trigger food-related. That data point is now stored, categorized, and available for analysis alongside every other entry you've ever logged.

This is the distinction that makes Iris different from a health journal. You're not writing diary entries. You're building a queryable dataset. And the Data Analyst can run correlations across that dataset in ways no human could by reading journal entries.

Three ways to log data

Conversational logging is the most common. You mention something in conversation — a symptom, what you ate, how you slept, your stress level — and Iris recognizes it as health data. It delegates to the Tracking Agent, which creates a structured entry. You see an entry card attached to the response showing what was logged. If something's wrong — Iris misheard the severity, got the timing wrong — you can tap to edit.

Multi-entry logging works too. "Slept 7 hours, had oatmeal for breakfast, and went for a 30-minute run" creates three separate entries in one go.

Tracking sprints provide structure when there's a specific question worth answering. A sprint is a focused stretch of consistent logging — usually a week or two — that you and Iris agree to run together because there's a hypothesis on the table ("is dairy actually a trigger?", "what's driving the afternoon crashes?"). During a sprint, Iris reaches out at the times you've agreed on and asks the few things the investigation needs — how you slept, whether the symptom showed up, what you ate around the suspected window. You answer in conversation; the Tracking Agent does the structuring. Sprints aren't the default mode of using Iris. Most days you don't track at all. Sprints are the mode you shift into when there's real signal worth catching.

Direct entry creation through the app's entry interface, if you prefer to log without a conversation.

What happens to your data

Every entry gets categorized — one of 23 defined categories like Sleep, Exercise, Mood, Meal, or Symptom. Each category has its own data schema: a sleep entry extracts duration, quality, and wake-ups; a meal entry extracts ingredients, timing, and estimated macronutrients; a symptom entry extracts severity, location, and possible triggers.

The extraction pipeline validates the structured data against the schema. If something doesn't parse correctly, it retries. The schemas enforce consistency — every headache entry has the same structure, every meal entry has the same structure. This consistency is what makes cross-referencing possible later.

Notes versus entries

These are different things, and understanding the distinction helps.

Notes are memory — markdown files about you. Your conditions, medications, investigation findings, life context. They're managed by the Memory Manager and persist indefinitely. Think of notes as what Iris knows about you.

Entries are data — structured health observations with timestamps. Your headache at 2pm, your sleep last night, your stress level today. They're created by the Tracking Agent and queryable by the Data Analyst. Think of entries as what happened to you.

When you say "I've been taking magnesium for migraines," that's a note — a fact about you that should be remembered. When you say "took magnesium this morning," that's an entry — a data point with a timestamp. Iris handles the distinction automatically, but knowing it exists helps you understand what the system is doing with your information.

How your data gets analyzed

The Data Analyst is a specialist agent that runs SQL queries on your entries. When you ask "what correlates with my headaches?" or when the analysis prompt fires from a guide article CTA, the Data Analyst examines your data across multiple dimensions: which factors correlate with your symptoms, whether there are time lags (stress today → pain tomorrow), and whether factors interact (poor sleep alone is manageable, but poor sleep plus high stress reliably predicts a flare).

The Data Analyst runs synchronously — Iris waits for results before responding. This means when Iris tells you "your pain averages 2 points higher after nights with less than 6 hours of sleep," that's a computed finding from your actual data, not a generic suggestion.

External data

If you connect a Garmin wearable, entries are created automatically from your device data — daily activity summaries (steps, active minutes), sleep data (duration, stages, quality scores), and stress/HRV data. This enriches your dataset without additional manual logging.

Why consistency matters when you're in a sprint

When you're in a tracking sprint, two weeks of brief, consistent data — logged at roughly the same times each day — gives the Data Analyst enough to identify real patterns. Three to four weeks is better. The analysis needs consistent data points across multiple variables to separate signal from noise.

What doesn't help, even inside a sprint, is logging obsessively for two days, skipping three days, logging a wall of text on Saturday, then nothing for a week. Sporadic detailed entries are less useful than brief consistent ones. If a day is too bad to track, skip it — that gap is itself a data point. Sprints are designed to be low-effort: a few specific things, in conversation with Iris, on the days you've agreed to log.

Outside a sprint, none of this applies. Tracking isn't the baseline relationship — it's a tool you pick up when there's a question worth answering and put down when there isn't.

References

Measuring diet intake and gastrointestinal symptoms — PubMed Central, 2020. Precise timing data revealing correlations missed by subjective recall.
Ecological momentary assessment in health research — Pain Medicine, 2019. Real-time health tracking improving pattern detection.
Patient-generated health data and clinical outcomes — Health Affairs, 2016. Longitudinal patient data improving health investigation.