Miasmatic Pattern Analysis with AI: Applying Kent's Classical Framework at Scale

Miasm theory is one of the more interpretively demanding parts of classical homeopathic case analysis. Identifying which underlying miasmatic layer or layers — per Kent's framework and related classical sources — a case's pattern of symptoms suggests isn't a matter of matching a single symptom against a single category. It requires holding a case's entire structured picture in mind at once and reasoning about how its parts fit together as a pattern, which is a fundamentally different task from the kind of single-field extraction or classification that most of Gentle Pills' AI tasks handle.

That difference is why miasmatic analysis gets routed differently than everything else in the system.

Why this task runs on our strongest model tier

Gentle Pills routes every AI task to one of three Azure OpenAI deployment tiers, chosen by what a wrong answer actually costs, not by how impressive the task sounds. Simple, deterministic-checkable extraction runs on the fastest, cheapest tier, because a mistake gets caught the moment a doctor reviews it before anything saves. Grounded summarization and mapping tasks run on a mid tier. The isolate_miasmatic_layers task — the AI task behind Miasmatic Deconstruction — runs on our top tier, the strongest reasoning model we deploy.

That's a direct consequence of what the task actually requires, not a marketing decision to put "the best AI" on a feature that sounds advanced. Extracting a symptom's location or modality from a sentence is a bounded, checkable task — there's a right answer, and a doctor can verify it at a glance. Synthesizing a case's full symptom picture into candidate miasmatic patterns is not bounded in the same way. It requires weighing how multiple symptoms across different categories relate to each other as a whole, which is exactly the kind of open synthesis task where a stronger model's reasoning capacity actually earns its higher cost and latency. Put a smaller, faster model on this task and you'd get answers that sound plausible but don't actually hold up under the kind of cross-referencing a doctor familiar with miasm theory would expect.

What the output actually is — and isn't

This needs to be stated plainly, because it's the part most likely to be misunderstood if left vague: the output of isolate_miasmatic_layers is a set of candidate miasmatic layers, each with the model's stated rationale, presented for the doctor to evaluate. It is not a diagnosis. Nothing in Gentle Pills treats it as one, and nothing in the product flow lets it become one without a doctor's own judgment sitting on top of it.

The distinction matters because miasm theory, as applied in classical homeopathic practice, is itself a matter of trained clinical interpretation — it's not a lookup table where a given symptom set maps deterministically to a given miasm. Two experienced homeopaths can reasonably read the same case differently. An AI system has no standing to resolve that kind of interpretive judgment on a doctor's behalf, and Gentle Pills doesn't pretend otherwise. What the tool does is perform the first, more mechanical pass — cross-referencing a case's structured symptom data against the classical framework's patterns — so the doctor starts their own evaluation from a set of reasoned candidates instead of a blank page.

Put differently: the AI's job here is pattern recognition assistance across a classical interpretive framework that a doctor would otherwise have to perform entirely by manual cross-reference, symptom by symptom, against everything they remember or can look up about Kent's system. It is not a substitute for the clinical judgment that miasm theory actually requires to apply well.

How this connects to the rest of the system

This task doesn't operate in isolation from Gentle Pills' broader approach to grounding AI output. The model reasons from the case's actual structured symptom data — the same data a doctor has already reviewed and, in many cases, confirmed through earlier AI-assisted extraction — rather than from a generic, invented pattern-match against homeopathic theory in the abstract. It isn't free-associating about miasms; it's working from the specific picture in front of it, the same way a doctor would be expected to.

That consistency matters more than it might seem. A system that grounds its symptom extraction in a doctor's actual case notes, grounds its rubric suggestions in the actually-loaded repertory, and grounds its knowledge citations in actually-retrieved passages — but then lets its highest-stakes reasoning task float free of that discipline — would be a contradiction in the architecture. Miasmatic Deconstruction is built to the same standard as everything else: reason from what's actually there, say what the reasoning is, and leave the interpretive decision with the person qualified to make it.

Why this still belongs to the doctor

Kent's miasmatic framework is a 19th-century interpretive system, not a diagnostic instrument with a known accuracy rate. Treating an AI's output here as anything more than a structured starting point for the doctor's own analysis would misrepresent both what the model is doing and what miasm theory itself claims to be. Gentle Pills is built for qualified homeopathic doctors specifically because tools like this only make sense in the hands of someone who can evaluate the candidates critically, weigh them against the rest of the case, and decide — using their own training and judgment — what the pattern actually means for this patient.