Your Mood Is Not in the Watch: Does Your Export Actually Hold the Subjective Half of Your Health Data?

Half of what matters about your health is not a number. Your mood, your pain, whether a day felt heavy: the Health app invites you to log all of it. Then the export turns it into rows of counts. Here is exactly what survives that trip, checked against a real export schema.

By MetricBridge · Published 16 November 2026 · ~8 min read

AI-assisted, human reviewed. How we use AI

Quick answer. Partly, and only in volume. An Apple Health export stores subjective entries as counts and durations: Mindful Minutes as minutes per day, Mood Changes as an event count per day, and 38 symptom rows as occurrence counts. It does not store what you felt. An agent can tell you that you logged something and how often; it cannot tell you the feeling, and any tool that claims to read it is guessing.

The export is a numeric table, and feelings do not flatten into one number

An export used by AI agents is not the Health app. It is a daily aggregate table, and every row has to collapse to a single number per day. That constraint is what makes the file small enough to hand to an agent in the first place, and it is also why the subjective half gets thinned out on the way through.

Look at how the rows are actually shaped. Each metric is defined by four things: a query name the agent calls, a unit, a daily aggregate, and the underlying HealthKit identifier. Three real rows from the 194-metric catalogue show the pattern:

Mindful Minutes   query: mindful_minutes   unit: min    aggregate: minutes per day   HealthKit: MindfulSession
Mood Changes      query: mood_changes      unit: count  aggregate: sum per day       HealthKit: MoodChanges
Sexual Activity   query: sexual_activity   unit: count  aggregate: sum per day       HealthKit: SexualActivity

Read the aggregate column. Every one of them ends in a per-day number. A mood entry becomes "2" for a Tuesday. A therapy session becomes a duration. The row records that something happened; the schema has no field for what it was like.

What the export actually keeps on the subjective side

It is not empty. A real export carries a meaningful slice of the human half, just in a measurable form:

  • Mindfulness. Mindful Minutes in minutes per day. Good for trend lines: did your practice hold through a hard month?
  • Mood Changes. An event count per day, under the Symptoms group. It answers "how often did you log a change", not "what changed".
  • Symptoms. 38 separate rows, each an occurrence count: headaches, fatigue, and the rest of the symptom list, one row apiece.
  • Cycle and sexual activity. Logged events and cycle rows, all as counts or durations.

Notice the shape of what is missing. There is no valence, no intensity, no free text, and no note. The export is a ledger of events, not a diary.

What an agent can honestly say, and what it cannot

This distinction is worth more than any single metric, because it decides whether you trust an answer. Against your own baseline, an agent can do real work with the subjective rows:

  • Compare this month's mindful minutes to your usual month and flag a deviation from your baseline.
  • Line up symptom counts against sleep or resting heart rate and report an association, with the caveat that association is not cause.
  • Show when mood-change logging went quiet, which often tracks with a busy stretch rather than with how you felt.

What none of that supports is a sentence like "your mood improved". The data cannot carry it. A model that produces it is pattern-matching on a count and dressing it up as a feeling, which is exactly the kind of confident guess a private, inspectable tool should refuse to make. When you ask a health agent about stress or mood, the honest output is the measured part plus a plain statement of what the file does not hold.

Check it on your own file

You do not have to take our word for any of this. Point a local agent at your export and ask it to read the mood and mindfulness rows for the last two weeks. You are looking for output shaped like this, one line per day:

{"date":"2026-11-12","mood_changes":0,"mindful_minutes":0,"resting_heart_rate":61}
{"date":"2026-11-13","mood_changes":2,"mindful_minutes":10,"resting_heart_rate":59}
{"date":"2026-11-14","mood_changes":1,"mindful_minutes":10,"resting_heart_rate":58}

Those numbers are from a seeded series, not from a real person, but the schema is the shipped one. If your own output shows counts and minutes and nothing that reads like a feeling, the export is behaving correctly. The server reads that file read-only, 21 tools with 19 of them read-only, so nothing your agent does can rewrite it. Ask it to call get_mcp_status if you want confirmation of which folder it mounted.

The quiet privacy upside

There is a silver lining to a numeric export. A file of daily counts is a lot less revealing than the Health app's own mood journal. If a device is borrowed, or a backup is inspected, a mood entry can be read by a person; a "2" on a Tuesday cannot be. That is one reason a local, read-only export is a reasonable thing to hand an on-device agent: the agent gets the pattern, and the story stays in the app.

It is not a free pass. Symptom and cycle rows are still sensitive, and a file you move off the phone is personal data no matter how plain it looks. The point is narrower and more useful: know which half of your data you are actually handing over, and ask for the half that exists.

FAQ

Does my Apple Health export contain my mood?

It contains the volume of what you logged, not the content. Mindful Minutes stores a duration, and Mood Changes stores an event count per day, so an agent can see that you logged something and how often, but not the feeling behind the entry.

Why does the export reduce a mood entry to a count?

The export is a numeric daily table. Rows are keyed by query name, unit and daily aggregate, so every metric has to collapse to a number per day. A free-text or rated feeling has no single number to collapse to, which is why it stays out of the table.

Can an AI agent still say something useful about my stress or mood?

Yes, on the parts that are measured. It can compare mindful minutes or symptom logs against your own baseline and flag a deviation. It cannot say how you felt, and it should not guess; treat any emotion claim it makes as unsupported by the data.

Is it more private that my feelings are not in the export?

Partly. A numeric export is less revealing than the Health app's own journal, which is why it is a reasonable thing to hand a local agent. But symptom and cycle rows are still sensitive, and any file you move off the device should be treated as personal data.

Read the half that is actually there

MetricBridge writes all 194 metrics as local JSON, and the open-source server lets your agent read the measured part. The feeling stays in the app.

Where it goes: the app reads Apple Health read-only and writes JSON to a folder you pick. Nothing reaches us or a cloud AI. What you get: all 194 metrics, the MCP tools and on-device Ask. Price: $24.99 one time after a 7-day free trial, not a subscription.

MetricBridge is a general-wellness app, not a regulated medical device. Nothing here is medical advice, and the readiness estimate is measured against your own baseline. Questions? Support.