No Capture-Time Scale

"A '7' means nothing consistent across days or people. The words 'pretty sore' mean exactly what Dad felt."

The Slider Forge Refuses

Almost every symptom tracker opens with a scale: rate your pain 1 to 10. Forge never does, and the refusal is principled. A number captured in the moment is a lie dressed as data — a "7" is not comparable to yesterday's "7," let alone to anyone else's, and asking for it forces the person to quantify a feeling they have not finished feeling. Worse, it throws away the actual information. "A little stiff, eased up by noon" carries more than any digit could.

Words Stay; the Scale Is Derived

So Forge keeps the words. Dad's own phrasing — "좀 시큰하다," "pretty sore," "barely noticed it today" — is the citable truth, stored verbatim and never overwritten. Intensity still exists as a normalized value, but it is derived at interpretation time: Pippa reads the wording and maps it onto a consistent internal rubric (mild / moderate / severe / unspecified) so trends are comparable. The number is downstream of the words, and the words remain the evidence any analysis has to cite.

The raw wording is the citable truth; the scale is a derived, versioned reading. A capture-time number quantifies a feeling the person has not finished feeling and cannot be trusted across days. Keep the words; compute the scale later, consistently, and record which rubric version produced it.

A Rubric Change Is Re-Interpretation, Not a Rewrite

Because the scale is derived, improving it is safe. When the normalization rubric evolves — say version 2 reads "시큰하다" more precisely than version 1 — Forge re-interprets the affected crumbs and stamps the new generation. It never edits the original words, and older readings stay in the record so a comparison knows which generation it read. History is re-read, never rewritten. That single property is what lets the typed layer keep getting smarter without corrupting the truth beneath it — the theme the whole next track is built on.

Code

Raw words preserved; severity derived at interpretation time·python

# The crumb keeps Dad's exact words, forever:
crumb.text = "knee a little stiff after the ride, eased by noon"

# Interpretation derives a normalized reading and records the
# rubric version + the verbatim span it rested on:
reading = {
    "types": ["symptom"],
    "site": "knee",
    "severity": "mild",            # derived, not captured
    "trend": "improving",          # "eased by noon"
    "evidence": "a little stiff",  # verbatim, citable
    "interpreter": "claude",
    "rubric_version": 2,           # which generation read it
}
# Rubric v3 later? Re-read the crumb, stamp a new reading.
# The words never change. History is re-interpreted, not rewritten.

Exercise

Recall a recent ache or fatigue. First, rate it 1 to 10. Then write the sentence you would actually say about it to a friend. Ask: a month from now, which of the two could a reader trust more — and which one did you have to invent on the spot?

Hint

The number felt precise but you made it up; the sentence felt casual but it is real. Forge keeps the real one and derives the precise-looking one later, on purpose.

The Slider Forge Refuses

Words Stay; the Scale Is Derived

A Rubric Change Is Re-Interpretation, Not a Rewrite

Code

External links

Exercise

Progress

Comments 0