theme rss
visitor@fireatwill:~/projects$ ↑ history · enter to run
~/projects/0x0108
fireatwill ¬ 0x0108 ¬ projects2026.11

Health tool

author Sofus Bech status in daily use stack html · js · claude artifact · obsidian source private

A personal reference for nutrition, supplements, hydration and labs where every daily target has a name attached — Rhonda Patrick, Andy Galpin, Robert Lustig, or the official tables when none of them gives a number. It checks meals and a supplement stack against those targets, flags lab values worth a second look, and keeps its entire state in one markdown note in my Obsidian vault. It may become an app. For now it is one HTML file.

healthnutritionobsidianclaude

Health advice arrives as podcasts. Three hours on omega-3, four on protein and hydration, two on sugar — all persuasive, all in different units, and none of it in front of me when I'm holding a can of sardines wondering whether it counts. I wanted the numbers out of the audio and into one place that knows my body weight and can do the arithmetic.

The problem

The advice isn't the hard part; reconciling it is. One source gives protein per kilo, another per pound, the official tables a flat minimum. Some numbers are targets to reach, some are ceilings to stay under, and some have a toxicity limit sitting above them. A spreadsheet can hold all that, but it can't answer "does today's food cover it?" — and a calorie-counting app answers a different question, with somebody else's targets.

Every number on the page has to be able to answer one question: says who?

How it works

Each of the 23 tracked nutrients stores what every source says about it, with the reasoning attached, and the target you see is resolved from those — never typed in as a bare number. The order is fixed and boring on purpose:

resolveTarget()
# who sets a target — first match wins
1. a number I typed in myself
2. my pick, where the experts differ by more than 15%
3. the experts, in a fixed order
4. diet-specific practice
5. the official tables      # NNR 2023 · EFSA · IOM

# then: per-kg values are multiplied by body weight, condition
# profiles adjust the result (add / at least / at most), and upper
# limits from toxicity data stay on as rails whoever won above

Against those targets, every nutrient gets a gauge: supplements, the foods I eat on purpose each week averaged per day, and whatever meals I log — with the upper-limit zone hatched in so "more" never reads as "better". Meals are typed the way I think (3 eggs, 150 g laks) and matched against a table of 85 whole foods with Danish aliases. Hydration gets its own tab: a fluid and sodium target that moves with training minutes and sauna, on Galpin's numbers.

The part I'm happiest with is storage. There is no database — the note is the database. The tool reads and writes a single file in my Obsidian vault through GitHub, and that file is a normal, readable note with the machine state fenced inside it:

Health/Health.md
# Health

<!-- health-app:start -->
## Daily targets         # regenerated on every save,
## Supplements           # readable without the tool
## Labs and measurements
## Flags
## Where the experts disagree

## Data
```json health-data      # the actual state
{ "profile": { … }, "supplements": [ … ], "labs": [ … ] }
```
<!-- health-app:end -->

## My own notes          # outside the fence — never touched

Only the region between the two markers is rewritten; anything I write by hand around it survives. Because the vault is a git repo, every save is a commit, so I get history and undo without building either. And if the tool vanished tomorrow, I'd still have a well-organized note.

Claude does the reading I'd otherwise skip. Paste a blood panel, or type a sentence about something I measured or started taking, and it comes back as a checklist of structured entries to tick before anything is written. Photograph a supplement label and it reads the nutrition table. Scan an ingredient list and every E-number is explained, with the processing level (NOVA 1–4). Questions are answered with my targets, labs and sources as context. The rule throughout: the model proposes, I review, then it's saved — never the other way round.

Lab values run through a small set of rules taken from the same sources, in Danish lab units. They produce flags like "check this" or "talk to a doctor" — pointers, not diagnoses.

What I'd do differently

Roadmap

Maybe an app. Three things would have to change first: storage (the note-as-database trick doesn't travel — an app needs its own store and, for health data, a privacy stance as strict as Letting Go's: on-device first), the food database, and Apple Health as the source for weight, heart rate and training minutes instead of typing them in. The part I'd keep exactly as it is: every number says who it came from.


Not medical advice — it's a reference that does arithmetic, built by someone who is not a doctor. It draws on the public work of the people named and is not affiliated with or endorsed by them. Thinking about sleep instead? Run cat 0x0107 above — or get in touch.