AI-Assisted Repertorization: Rubric Suggestion and Transparent Coverage Ranking

Repertorization — matching a case's symptoms against entries in a classical repertory like Kent's, then ranking candidate remedies by how well they cover the selected rubrics — is one of the most labor-intensive methods in classical homeopathic practice. A repertory is a large, densely cross-referenced index. Finding the rubrics that genuinely correspond to a patient's symptoms, out of thousands of possible entries phrased in period-specific language, is a skill that takes years to develop and real time to execute carefully on every case.

Gentle Pills' suggest_rubrics AI task is built to speed up one specific part of that process — the search — while leaving the analytical judgment that follows entirely to the doctor. The distinction between those two things is the whole design, and it's worth being precise about where the line sits.

What AI actually does here

Given a case's structured symptoms, suggest_rubrics proposes candidate rubrics from the currently loaded repertory corpus that plausibly correspond to those symptoms. The doctor sees a list of suggestions, reviews them against their own understanding of the case, and selects which ones actually apply. Nothing is auto-selected into the analysis — suggestion and selection are separate steps, with the doctor doing the selecting.

The suggestions are scoped strictly to rubrics that exist in the loaded repertory corpus. This is the same grounding principle that governs knowledge retrieval elsewhere in the system: the model is only shown the rubrics actually available to suggest from, and it cannot propose a rubric it was never given. That constraint matters because a repertory suggestion tool that occasionally invents a plausible-sounding rubric that doesn't actually exist in the source text would be worse than useless — it would send a doctor looking for something that isn't there. Scoping suggestions to the real corpus removes that failure mode by construction rather than by instruction.

What stays deterministic and open

Once a doctor has selected rubrics, the remedy ranking that follows is not an AI output at all. It's an open, deterministic calculation: a weight × grade coverage score computed over the doctor's selected rubrics, the same kind of math a repertorization has always used, just computed instead of hand-tallied. Every remedy's rank is traceable back to exactly which selected rubrics it covers and at what grade — nothing about why a remedy ranked where it did is hidden behind a model's internal reasoning.

This matters for a reason that goes beyond convenience. A black-box AI score that says "this remedy is a good fit" without showing its work is not something a doctor can audit, challenge, or explain to a colleague. A transparent coverage calculation is all three. The doctor can see precisely which rubric selections are driving a given remedy up or down the list, and can add, remove, or reweight rubrics and watch the ranking respond in a way that's fully explicable — because the math producing it was never opaque to begin with.

Where the line sits

It's worth stating the division of labor plainly, because it's easy to blur in casual description:

• AI's job: surface candidate rubrics from the loaded repertory that a doctor might not have thought to search for by exact wording, speeding up the search step of repertorization.
• Doctor's job: decide which suggested rubrics genuinely apply to this specific case, select them, and interpret the resulting ranked list using clinical judgment — which remedy actually fits the whole picture, not just the top of a list.
• Deterministic math's job: compute the coverage ranking from the doctor's selections, transparently and reproducibly, with no model involved in that step at all.

Three different mechanisms, each doing one job, none of them overstepping into the others. The AI never ranks remedies. The ranking math never suggests rubrics. The doctor never has to trust an opaque score — every number on the screen traces back to a rubric they chose and a grade they can inspect.

Why search is the right place for AI here

A large classical repertory uses period-specific phrasing, cross-references entries across sections in non-obvious ways, and assumes a level of familiarity with its own indexing conventions that takes years to build. A doctor who knows exactly what they're looking for can usually find it. The harder case is the symptom that's real and clinically significant but phrased differently than the doctor expects the repertory to phrase it — the rubric that exists but isn't where intuition would first look.

That's precisely the kind of search problem AI assistance is suited to: scanning a large, densely structured text for entries that match a symptom's meaning even when the exact wording differs, and surfacing candidates a doctor can then evaluate with their own expertise. It is not suited to — and is not asked to — make the actual clinical call about which of those candidates genuinely belongs in this case's analysis.

The practical effect

For a doctor working a case in Gentle Pills, this means repertorization keeps its classical shape: select rubrics, review ranked remedies, apply clinical judgment to the final choice. What changes is how long the rubric-selection step takes, because AI is doing the first pass of searching a large repertory for candidates, instead of the doctor doing that search manually against every structured symptom in the case. The analytical method underneath — and every number the ranking produces — remains exactly as auditable as it would be done by hand.