The Line

"The app does not diagnose; the daughter answers." Not a disclaimer — a line drawn straight through the middle of the architecture.

One Sentence, Load-Bearing

Forge holds symptoms, medications, and medical records, so its most important design decision is not a feature — it is a line. The app does not diagnose; the daughter answers. This is a hard invariant: it shapes prompt design, UI copy, and what the analysis engine is even allowed to emit — it is not a paragraph of fine print nobody reads. Cross it and you have not written bad copy; you have shipped a diagnostic device that was never supposed to exist.

What 'Diagnose' Actually Means

Precision matters here, because "don't give medical advice" is too blunt and would gut the product. The forbidden acts, on a Forge-generated surface, are exactly four: naming a condition as a diagnosis, asserting a cause, predicting an outcome, and recommending that a treatment be started, stopped, or changed. Everything short of those is not merely allowed — it is the entire point of the engine:

Observe, organize, correlate, and raise questions worth taking to a physician — never name a condition, assert a cause, predict an outcome, or push a treatment change.

The Same Data, Two Outputs

The line is easiest to feel on one dataset. Given "a flare twice in the week after heavy leg days," an observation says: these co-occur, here are the crumb IDs, it may be worth asking a doctor whether the load matters. A diagnosis says: your training is causing gout flares, cut leg day. Same evidence — and the second version named a condition, asserted a cause, and prescribed a change, three of the four forbidden acts in one breath. When you build the analysis surface in Track 6, this is the line every narration is checked against, mechanically.

Notice what the line is not: it is not silence, and it is not cowardice. Observation with evidence is often more useful to a real doctor than a confident guess, because it hands them the timeline instead of a conclusion. The engine's job is to make the pattern visible and legible — the naming stays with the human who is licensed to do it.

Code

Observe vs diagnose — the same evidence·text

evidence:  flare crumbs on 2 dates, each within a week of a
           heavy lower-body session; 5 crumb IDs

# OBSERVE (Forge-initiated: allowed)
"These two flares each landed within a week of a heavy leg
 session. That is co-occurrence across 2 points, not a trend.
 Worth asking a doctor whether load is relevant. [crumb:...]"

# DIAGNOSE (Forge-initiated: FORBIDDEN)
"Your training is causing gout flares. Cut leg day and start
 an anti-inflammatory."
#   ^ names a condition, asserts cause, prescribes a change
#     -> a bug at the invariant level, not a wording problem

Exercise

Take one real observation about your own body — for example, 'I slept badly the two nights after evening coffee.' Write two versions: one that stays on the observe side (co-occurrence + what you'd ask), and one that crosses into diagnosis. Then circle the exact words that crossed the line.

Hint

The crossing words are usually a condition name, a 'because', a 'will', or a 'should take/stop'. If your safe version still contains one of those, it isn't on the safe side yet.

One Sentence, Load-Bearing

What 'Diagnose' Actually Means

The Same Data, Two Outputs

Code

External links

Exercise

Progress

Comments 0