Cover illustration for Tracing persistent late-afternoon mood drops to a neglected lunch baseline

What the AI found

“Your lowest mood scores coincide with days you skipped a midday protein source, not your meeting load or sleep deficit.”

Before

Vague 4pm slump logged as low mood for months

After

One 5-minute Sunday pattern check she acts on

The same system, three states — real screens, not a screenshot

1Starting
Apple Notes
  • Oct 14: 16:15 - flat, irritable. client meeting dragged on.
  • Oct 16: 16:40 - brain fog, heavy. 6h sleep last night.
  • Oct 22: 15:55 - zero focus. maybe too much screen time?
  • Oct 28: 16:20 - exhausted again. need a holiday.
2Working
Claude

Prompt

Here is my 6-week CSV containing daily sleep, calendar meeting hours, 4pm mood ratings (1-10), and lunch notes. Identify what actually correlates with my mood drops below 4. Do not offer medical advice; isolate the mathematical co-occurrences.

Here is my 6-week CSV containing daily sleep, calendar meeting hours, 4pm mood ratings (1-10), and lunch notes. Identify what actually correlates with my mood drops below 4.

AI

Meeting volume (r = -0.12) and sleep duration (r = 0.18) show negligible correlation with your afternoon mood drops. However, days with mood <= 3 (14 instances) share a consistent signature: 13 of those 14 days had estimated lunch protein under 12g (pastries, salads without protein, or skipped meals). On days with 25g+ protein at lunch, your average 4pm mood is 7.4/10 regardless of meeting load.
3Implemented
Apple Health

28g (stable)

Midday Protein Avg

0.8 / wk (was 4.2)

Late Afternoon Slump

5 min

Sunday Review Duration

IndividualAll-Access in use

Tracing persistent late-afternoon mood drops to a neglected lunch baseline

How an architectural project manager decoupled afternoon emotional crashes from work stress using a unified weekly review.

A 36-year-old architectural project manager, Northern Europe

Tools used

The real tools used here — swap any for your own equivalent. Each links to how we’d set it up.

4 min readWellness & AI editorial
1

Before anything was set up

Four months of sporadic notes on her phone read like an inventory of fatigue. She had convinced herself that her afternoon mood crashes were the byproduct of open-plan office acoustics, difficult clients, or broken sleep. She tried blue-light glasses, twenty-minute meditation tracks, and afternoon espressos. None shifted the baseline. The logs accumulated haphazardly, unexamined and contradictory.

Apple Notes
  • Oct 14: 16:15 - flat, irritable. client meeting dragged on.
  • Oct 16: 16:40 - brain fog, heavy. 6h sleep last night.
  • Oct 22: 15:55 - zero focus. maybe too much screen time?
  • Oct 28: 16:20 - exhausted again. need a holiday.
2

All-Access, doing its job

Using the unified setup from All-Access, she consolidated six weeks of fragmented data into a simple five-column table. She paired her subjective 1-to-10 mood ratings with her sleep exports and meeting durations, alongside a raw sentence describing what she ate at midday. She pasted the batch into Claude with instructions to isolate mathematical correlations rather than offer generic lifestyle tips.

Claude

Prompt

Here is my 6-week CSV containing daily sleep, calendar meeting hours, 4pm mood ratings (1-10), and lunch notes. Identify what actually correlates with my mood drops below 4. Do not offer medical advice; isolate the mathematical co-occurrences.

Here is my 6-week CSV containing daily sleep, calendar meeting hours, 4pm mood ratings (1-10), and lunch notes. Identify what actually correlates with my mood drops below 4.

AI

Meeting volume (r = -0.12) and sleep duration (r = 0.18) show negligible correlation with your afternoon mood drops. However, days with mood <= 3 (14 instances) share a consistent signature: 13 of those 14 days had estimated lunch protein under 12g (pastries, salads without protein, or skipped meals). On days with 25g+ protein at lunch, your average 4pm mood is 7.4/10 regardless of meeting load.
3

The finished system, running on its own

The stack now runs silently. Daily entries take forty seconds into Apple Notes. On Sunday evenings, Claude ingests the export, calculates simple correlation coefficients against her calendar tags, and returns a single three-bullet summary. She does not alter her work schedule or chase elaborate rituals; she packs a tin of sardines or boiled eggs for office days. The late-afternoon fog simply ceased appearing.

Apple Health

28g (stable)

Midday Protein Avg

0.8 / wk (was 4.2)

Late Afternoon Slump

5 min

Sunday Review Duration

Down from 4.2 to 0.8 per week

Afternoon mood dips

5 minutes every Sunday

Log review duration

7.6 / 10 (was 3.4)

Self-reported clarity at 4pm

Apple NotesFrictionless capture

Zero-latency offline text entry directly from the lock screen.

ClaudePattern synthesis

Superior reasoning over multi-variable text tables without hallucinated trends.

Apple HealthPassive biometrics

Reliable background collection of sleep duration and resting heart rate.

These are the tools used in this story. Any can be swapped for an equivalent you already trust.

Explore the All-Access archive

This story runs on All-Access. The tools and prompts above are the real build — swap any tool for your own equivalent and follow the same steps.

Suggested for you

Based on what you've been reading — always learning.

See all →