The Indian schools converged on a question Western epistemology has tended to leave loose: when someone says they know something, by which cognitive route did they come to know it? Pramāṇa-theory is the answer — the systematic mapping of the distinct routes that produce valid cognition. Different schools accept different numbers, but the operational gain is the same in each: instead of treating “evidence” as a uniform category, you can locate any knowledge claim within a specific cognitive structure and ask whether the conditions for that structure are met.
The six routes that the Mimamsakas accept, and which between them cover the bulk of the territory:
Perception (pratyakṣa) — direct cognition through the senses, when the object is present and the perceptual conditions are right.
Inference (anumāna) — cognition grounded in pervasion (see N2): “wherever smoke is, fire is; smoke is here; therefore fire is here.”
Comparison (upamāna) — cognition by similarity: knowing what a gavaya (wild ox) is by being told “it is like a cow,” and then seeing one. Different from inference because it does not rest on pervasion.
Testimony (śabda) — cognition from a reliable verbal source. See N8.
Postulation (arthāpatti) — cognition forced by the requirement of coherence: an observation that cannot stand on its own implies an unobserved fact. See M1.
Non-perception (anupalabdhi) — cognition of absence in a context where the absent thing would have been perceived if present. See N7.
Each pramāṇa has its own validity conditions, its own characteristic failure modes, and its own typical operational use. A knowledge claim that looks like one pramāṇa but is actually another — or one whose conditions are not actually met — is a failure waiting to be named.
Where English Falls Short
Western epistemology has worked out particular pramāṇas — perception, inference, testimony — at varying levels of detail, but it has not produced a unified taxonomy of cognitive routes the way the Indian schools did. “Evidence” or “justification” in English-language epistemology is typically treated as a uniform category, with the work being to ask “how strong is the evidence.” The Indian question is different: what kind of cognitive route produced this knowledge claim, and are its conditions met? The Indian frame is more discriminating at the input layer.
The practical consequence is that English-language design and AI work conflates routes that the Indian frame keeps apart. A model that hallucinates a citation is doing one kind of failed cognition; a model that misperceives an image is doing another; a model that postulates a hidden state to explain an observation is doing a third. The taxonomy lets you see these as different problems with different fixes.
Where it Shows Up
AI evaluation. Most evaluation frameworks treat model output as a uniform category and ask “is it right?” The pramāṇa frame asks first: which cognitive route is the model claiming to use? Perception (interpreting an image)? Inference (reasoning)? Testimony (reporting what a source said)? Postulation (filling in an unstated fact)? Each route has different validity conditions, and “is it right?” decomposes into different questions for each.
Research methodology. When summarising findings, ask: by which pramāṇa is each claim known? A claim resting on direct measurement, a claim resting on inference from correlation, and a claim resting on testimony from interviewees are not interchangeable, even when they appear in the same paragraph.
Decision-making under uncertainty. When teams disagree on a question, the disagreement is often at the pramāṇa level; the data level is downstream. One side is reasoning by inference, the other by testimony, the third by postulation. Naming which route is in play unsticks the conversation.
Auditing knowledge claims in product strategy. “Users want X” — by which route is this known? Survey testimony, observed behaviour (perception), postulation from coherence with other facts, inference from comparable products? The answer changes how much weight the claim can bear.
Designing trust systems. Any system that propagates claims (citation networks, AI-generated summaries, knowledge graphs) implicitly takes a position on which pramāṇas its content is resting on. Making this explicit lets you design appropriate validation for each route.
Diagnostic Question
“By whichpramāṇais this claim being made — and have the conditions for that route to be valid actually been met?”
IKS Roots
The Sanskrit term is pramāṇa (प्रमाण), literally “the means by which pramā — valid cognition — is produced.” The schools differ on how many pramāṇas they accept. Charvaka (the materialist school) accepts only pratyakṣa. Vaisheshika accepts pratyakṣa and anumāna. Nyaya accepts four: pratyakṣa, anumāna, upamāna, śabda. Mimamsa (Bhatta) accepts all six listed above. Advaita Vedanta follows Bhatta Mimamsa. The disputes between schools about which pramāṇas are independent (and which can be reduced to others) are themselves a major part of Indian epistemology — Nyaya’s argument that arthāpatti reduces to anumāna, and Mimamsa’s response, is the canonical example.
See N2 (vyāpti — the relation that grounds inference), N7 (anupalabdhi — non-perception), N8 (śabda — testimony), and M1 (arthāpatti — postulation) for the individual pramāṇas the library treats in depth.
Further Reading
Bimal Krishna Matilal, Perception: An Essay on Classical Indian Theories of Knowledge (Oxford), for the canonical scholarly treatment. Stephen Phillips, Epistemology in Classical India, for an accessible entry. The Karl Potter Encyclopedia of Indian Philosophies volumes on Nyaya and Mimamsa for technical depth on individual schools.