Correlation, Not Cause

"Two things showing up together is an observation worth a question. It is never a proof worth a diagnosis."

Rule Three: The Line, at Its Hardest

The third rule is where the Track 2 safety line meets its most tempting surface. Analysis exists to surface patterns, and a strong pattern begs to be concluded. Rule three forbids the conclusion: Forge may say two things co-occur and ask whether it is worth investigating; it may not assert a cause, name a condition, or push a treatment change. "These flares land in the week after heavy leg days — worth asking a doctor" is allowed. "Leg day is causing this" is not. The gap between them is the whole discipline, and it is smallest exactly when the pattern looks most obvious.

What the First Narration Refused to Say

The first real analysis had two heavy-load sessions and a baseline drift sitting suspiciously close together. It would have been easy — and wrong — to connect them. Instead it held them as co-occurrence: "one observation — that is co-occurrence, not cause," and closed by asking for a crumb a day on the foot so a pattern could actually form. That is the rule as behavior: not a disclaimer appended to a conclusion, but a genuine refusal to conclude, followed by a request for the data that would let a real answer emerge — from a doctor, not the app.

Correlation is presented as correlation: surface it, question it, never conclude it. No amount of citation, disclaimer, or pattern strength converts an observation into a permitted diagnosis. The strongest-looking correlation gets the same treatment as the weakest — named, questioned, and handed to the physician who is allowed to reason about cause.

When Context Sharpens the Question

Context can change which question is worth asking — and that is still not the app diagnosing. The motivating case: a forefoot ache after a trip. In isolation it might suggest one thing; with the trip's heavy hiking in view, overuse becomes the more natural question to raise. Notice the shape carefully — context did not let the app declare overuse; it let Pippa raise a better-aimed question. That cross-app context arrives through the digest contract of Track 8, and even there the output stays a sharper question, never a verdict. The line holds no matter how much context is in the room.

Code

Correlation stated correctly vs crossing into cause·text

pattern:  flares cluster in the week after heavy leg sessions

# ALLOWED (rule 3): co-occurrence + a question
"These flares tend to fall in the week after a heavy leg
 session [crumb:...]. That is co-occurrence across a few
 points, not cause. Worth raising with a doctor."

# FORBIDDEN: cause, condition, prescription
"Heavy leg days are triggering gout flares - cut leg day."
#   ^ asserts cause + names a condition + prescribes a change
#     Same evidence. Three forbidden acts. This never ships.

Exercise

Take a co-occurrence you have noticed in your own body — two things that seem to show up together. Write the observe-only version (co-occurrence plus the question you would raise) and the forbidden version (a stated cause). Circle the words in the second that crossed from observation into diagnosis.

Hint

The crossing words are usually 'because', 'causes', a condition name, or a 'should'. The safe version can point at the pattern and ask, but it never closes the loop with a cause — that loop is the doctor's to close.

Rule Three: The Line, at Its Hardest

What the First Narration Refused to Say

When Context Sharpens the Question

Code

External links

Exercise

Progress

Comments 0