Analyse Your Garmin Sleep Data With Claude

Day 89: Analyse Your Garmin Sleep Data With Claude

Your Garmin already records sleep score, duration, sleep stages, HRV, stress, Body Battery, resting heart rate, and activities. Looking at those screens one by one tells you what happened. Analysing the full history together can help explain why your good and bad nights differ.

This uses the same export and Claude workflow as the sleep-score method from Day 62, but the goal here is different. You are not asking Claude to chase a score of 90. You are asking it to audit the data, quantify patterns, challenge weak conclusions, and identify personal factors worth testing.

Step 1: Export your Garmin data

Go to garmin.com → Account → Data Management → Export Data.

1. Log in to your Garmin account.
2. Select Export Your Data.
3. Garmin prepares a ZIP archive and emails you when it is ready. This can take from several minutes to several hours.
4. Download and unzip the archive. Look for the DI_CONNECT folder, which contains much of your sleep, HRV, stress, Body Battery, resting-heart-rate, and activity history.

Step 2: Choose Claude Cowork or Claude Code

You need a version of Claude that can inspect multiple files in a local folder. Use either Cowork for the simpler visual workflow or Claude Code if you are comfortable with a terminal.

Option A: Claude Cowork

Claude Cowork runs through the Claude desktop app and can work with a folder you explicitly share.

1. Install the Claude desktop app from claude.ai/download.
2. Open it and log in.
3. Start Cowork and grant access to the unzipped Garmin export folder.
4. Paste the analysis prompt from Step 3.

Cowork availability depends on your Claude plan and platform.

Option B: Claude Code

Claude Code is the terminal-based option. It can inspect the folder structure, write small analysis scripts, and process the JSON files together.

On Mac: press Cmd + Space, search for Terminal, and open it.

On Windows: open PowerShell from the Start menu.

Install Claude Code with:

curl -fsSL https://claude.ai/install.sh | bash

Next, type cd with a space after it, drag the unzipped Garmin folder into the terminal, and press Enter. Start Claude Code:

claude

Complete the sign-in flow, then paste the prompt below.

Step 3: Use this deep-analysis prompt

This prompt keeps the useful sleep, HRV, stress, and activity comparisons from Day 62, but asks for a much more rigorous analysis instead of jumping directly to advice.

You are an expert sleep data analyst and sports scientist. I am giving you my full Garmin data export. Most relevant files are in the DI_CONNECT folder.

Your primary task is to ANALYSE my personal sleep and recovery data. Do not start by giving sleep advice. First build a reliable dataset, quantify patterns, test alternative explanations, and show how strong or weak each finding is.

Use scripts where helpful so the analysis is reproducible. Do not silently skip malformed files or assume field meanings. Inspect the available schemas and explain how you interpreted and joined the records.

Produce a structured report with these sections:

### 1. Data inventory and quality audit
- List every file used for sleep, sleep score, sleep stages, HRV status, overnight HRV, stress, Body Battery, resting heart rate, activities, training load, and recovery.
- State the date range, number of usable nights, missing periods, duplicate records, timezone issues, and fields that appear unreliable or unavailable.
- Explain how records from one day were matched to the following night's sleep.
- Clearly distinguish measured values, Garmin estimates, and values you calculated.

### 2. Personal sleep baseline
- Calculate mean, median, standard deviation, and meaningful percentiles for sleep score, sleep duration, bedtime, wake time, deep sleep, REM sleep, light sleep, awake time, overnight stress, HRV, and resting heart rate where available.
- Show weekly and monthly trends and identify genuine changes versus ordinary night-to-night variation.
- Use my own baseline rather than comparing me with generic population targets.

### 3. Best nights versus worst nights
- Compare my best 20% and worst 20% of nights, plus any nights scoring 90 or above as a secondary group.
- Quantify differences in duration, timing, sleep-stage estimates, HRV, resting heart rate, stress, Body Battery change, and previous-day activity.
- Report sample sizes, absolute differences, percentage differences where useful, and overlap between the groups.
- Include counterexamples: good nights that break the apparent pattern and bad nights where the supposedly helpful factor was present.

### 4. Previous-day activity analysis
- Match each night's sleep to the previous day's activities.
- Test associations with activity type, duration, distance, intensity, training load, start time, end time, total daily activity, recovery time, and rest days where available.
- Compare easy versus hard days and early versus late workouts only when enough examples exist.
- Check whether results remain plausible after accounting for weekday/weekend effects, recent training load, and baseline recovery.

### 5. HRV, stress, and recovery analysis
- Analyse how overnight HRV, HRV status, resting heart rate, stress, Body Battery, and sleep score move together over time.
- Identify lagged patterns across several days, not just same-night correlations.
- Separate expected mathematical or algorithmic relationships from genuinely useful personal patterns. Garmin metrics may share input data, so do not present those relationships as independent discoveries.

### 6. Pattern evidence table
Create a table for every potentially useful pattern with:
- Hypothesis
- Supporting dates and sample size
- Effect size or numerical difference
- Counterexamples
- Possible confounders
- Confidence: high, medium, low, or insufficient data
- What additional data would strengthen or reject the finding

### 7. Missing lifestyle context
- Identify questions the Garmin export cannot answer, such as alcohol, caffeine, meals, medication, illness, travel, room temperature, or screen use.
- Never invent these factors.
- Give me a short daily log template for collecting only the missing variables most likely to improve a future analysis.

### 8. Conclusions and experiments
- Rank the three to five strongest findings, but include only findings supported by my data.
- For each finding, propose a specific two-week experiment that changes one variable at a time.
- Define the metric, comparison, and minimum result that would support or reject the hypothesis.
- Keep recommendations secondary to the analysis and tie every recommendation to a reported finding.

### 9. Safety and limitations
- Flag persistent or unusual trends that may deserve discussion with a qualified health professional, but do not diagnose a condition.
- Explain the limitations of wrist-based sleep-stage estimates and observational data.
- Clearly state that correlation does not prove causation.

Finish with:
1. A one-page executive summary.
2. The five most important charts or tables.
3. A concise list of findings ordered by confidence.
4. The reusable analysis scripts and any cleaned data tables you created.

Be precise and skeptical. Cite dates, numbers, sample sizes, effect sizes, counterexamples, and uncertainty. If the data cannot support a conclusion, say so directly instead of filling the gap with generic sleep advice.

Use the findings as hypotheses

Garmin sleep stages and wellness metrics are wearable estimates, not a medical sleep study. Claude is also working with observational data: it can find relationships, but it cannot prove that one behavior caused a result. Use the report to design small experiments, collect missing context, and improve the next analysis rather than treating its first answer as a diagnosis.