What the AI found
“Your three lowest-efficiency sleep nights each followed dinner after 9:15 PM, not late caffeine or high evening screen time as your log suspected.”
Before
Four years of wearable exports never opened
After
One 8-minute Sunday sync on her own laptop
The same system, three states — real screens, not a screenshot
| export_health_2022.zip | 412 MB |
| export_health_2023.zip | 580 MB |
| export_health_2024.zip | 640 MB |
| sleep_notes_draft.txt | 0 bytes |
Prompt
Here is my Apple Health sleep export for November (CSV) and my Apple Notes dinner timestamps (txt). Identify any correlation between my mid-night awake minutes and dinner timing, caffeine, or screen time.
Here is my Apple Health sleep export for November (CSV) and my Apple Notes dinner timestamps (txt). Identify any correlation between my mid-night awake minutes and dinner timing, caffeine, or screen time.
AI
Your three lowest-efficiency sleep nights each followed dinner after 9:15 PM, not late caffeine or high evening screen time as your log suspected. On nights dinner completed before 8:00 PM, wakefulness averaged 18 minutes; after 9:15 PM, wakefulness averaged 64 minutes.89% (+7%)
Sleep efficiency (30d)
6 of 7 nights
Dinner cutoff adhered
8 mins / week
Review time
Finding the hidden cause of 3am awakenings
How a design director turned unread health exports into an eight-minute personal Sunday audit.
A 38-year-old design director, Northern Europe
Tools used
The real tools used here — swap any for your own equivalent. Each links to how we’d set it up.
Starting state
Before anything was set up
Her Apple Health archive contained four years of undisturbed export archives. Every morning she flicked past the ring notifications and sleep duration bar charts, feeling faintly guilty about the broken orange bars indicating wakefulness between 02:00 and 04:00. She kept an occasional note in her phone—sometimes blaming a late espresso at 16:30, sometimes blaming work emails read at midnight. Nothing was indexed. She possessed thousands of nightly data points and zero operational knowledge.
| export_health_2022.zip | 412 MB |
| export_health_2023.zip | 580 MB |
| export_health_2024.zip | 640 MB |
| sleep_notes_draft.txt | 0 bytes |
Working state
Membership, doing its job
She set up a three-step Membership workflow: export the rolling 30-day sleep table to iCloud, copy her lightweight Apple Notes meal ledger, and drop both into Claude using a structured evaluation prompt. Rather than attempting a sweeping lifestyle overhaul, she asked the model to test three specific hypotheses: evening screen exposure, afternoon caffeine, and meal cutoffs. In under ten seconds, the model isolated the actual correlation.
Prompt
Here is my Apple Health sleep export for November (CSV) and my Apple Notes dinner timestamps (txt). Identify any correlation between my mid-night awake minutes and dinner timing, caffeine, or screen time.
Here is my Apple Health sleep export for November (CSV) and my Apple Notes dinner timestamps (txt). Identify any correlation between my mid-night awake minutes and dinner timing, caffeine, or screen time.
AI
Your three lowest-efficiency sleep nights each followed dinner after 9:15 PM, not late caffeine or high evening screen time as your log suspected. On nights dinner completed before 8:00 PM, wakefulness averaged 18 minutes; after 9:15 PM, wakefulness averaged 64 minutes.Use case implemented
The finished system, running on its own
The workflow runs entirely on her terms. Each Sunday at 18:00, Apple Shortcuts dumps seven days of sleep stages and a plain-text meal log into a single folder on iCloud Drive. She opens Claude, taps her saved Membership prompt, and reads a four-bullet audit of her week. There is no new wearable, no subscription dashboard to log into, and no cloud repository holding her vitals. She kept her existing apps; she simply added a disciplined conversational interpreter.
89% (+7%)
Sleep efficiency (30d)
6 of 7 nights
Dinner cutoff adhered
8 mins / week
Review time
What an outside observer would notice
8 mins (down from none)
Weekly review duration
75 mins earlier
Dinner shifted forward
-24 mins per night
Nightly awake time
The stack — build it yourself
Native device integration that retains raw biometric records locally without third-party cloud hosting.
Zero-friction interface for recording dinner times and subjective morning energy in two seconds.
Handles raw CSV and unstructured text together with high context fidelity and calm synthesis.
These are the tools used in this story. Any can be swapped for an equivalent you already trust.
Go deeper
Do this yourself
Explore Membership
This story runs on Membership. The tools and prompts above are the real build — swap any tool for your own equivalent and follow the same steps.