Patanjali opens the Yoga sutras with a deceptively dry classification: mental activity is fivefold. Valid cognition (pramāṇa), mistaken cognition that tracks a real object wrongly (viparyaya), cognition generated by language with no referent (vikalpa), sleep (nidrā), and memory (smṛti). Each is a distinct kind of event. Each has its own structure, its own conditions, and its own way of going wrong.
The classification matters because the mind is otherwise treated as a single processor. “I thought X” gets used for valid perception, for fabricated representation, for half-remembered images, for the residue of dreams. The single verb hides five different cognitive routes, each calling for different attention and different treatment.
The discipline the sutra introduces is the discipline of recognition. Before asking whether a thought is right or wrong, useful or unhelpful, the prior question is: which kind of vṛtti is this? Once classified, the appropriate move follows. Valid cognition gets refined. Mistaken cognition gets corrected against the actual referent. Referentless cognition gets either grounded or marked as design fiction. Sleep is allowed. Memory is checked against present perception. Treating all five as one operation produces uniform responses to non-uniform events, which is the standard mode of cognitive failure.
The system is also reflexive. The yogic project — the cessation of the vṛttis as defined in sutra 1.2 (yogaś citta-vṛtti-nirodhaḥ) — depends on first being able to see what one is doing. The classification is the diagnostic that makes the cultivation possible.
Where English Falls Short
Western cognitive science distinguishes perception, inference, imagination, memory, and sleep, but as separate research areas pursued by different disciplines with different theoretical apparatus. The Yoga frame puts them in a single map and assigns them functional parity — they are all kinds of vṛtti, all forms of mental activity that the practitioner needs to recognise. The unified treatment is the contribution.
The English term “cognition” tends to mean valid cognition by default, with imagination and memory and dreaming treated as exceptions or special cases. The Yoga classification treats all five as primary. There is no privileged mode; there is the practice of discrimination.
The AI literature has begun to need this distinction and is groping toward it. “Hallucination,” “confabulation,” “memory,” “reasoning” are deployed without a unifying frame. The pañca-vṛtti is the unifying frame — a map at the right level of grain, a working taxonomy that any specific failure or success can be located inside.
Where it Shows Up
LLM output classification. Before debugging, locate the failure inside the five-mode map. Is the model doing vikalpa (generating a representation with no referent)? Viparyaya (tracking a real referent wrongly)? Smṛti (retrieving from training data with degradation)? Each calls for a different intervention. Lumping them as “the model got it wrong” produces uniform fixes for non-uniform failures.
User research interpretation. A user reports a memory of a past experience. The classification frame asks which vṛtti generated the report. Smṛti (genuine memory)? Vikalpa (a representation the user constructed afterward to explain their behaviour)? Viparyaya (a real event tracked wrongly)? The data quality of user-reported recall varies enormously depending on which is in play.
Design fiction and speculative work. Vikalpa used deliberately and marked as such. The classification permits the work — referentless representation is a legitimate cognitive event — and disciplines it, by requiring acknowledgement of which mode the artefact lives in.
Personal review of one’s own thought stream. “Was that observation, was that inference, was that imagination, was that memory, was that residue of dreaming?” The five-fold check is the basic move of the practice, and it transfers to any reflective context where the quality of one’s thinking matters.
AI system architecture. A system that conflates retrieval, generation, perception, and reasoning under one mechanism produces failures that cannot be localised. Separating the modes architecturally — different components for memory-like operations, generative operations, perception-like operations — makes failures debuggable.
Diagnostic Question
“Which of the five modes is producing this thought — and is the mode appropriate to the work I am asking the thought to do?”
IKS Roots
The Sanskrit term is vṛtti (वृत्ति), literally “turning” or “modification,” used technically in Yoga for any modification of the citta (mind-stuff). Sutra 1.5 declares vṛttis to be of five kinds, kliṣṭa (afflicted) or akliṣṭa (unafflicted). Sutras 1.6–1.11 enumerate and define them. Pramāṇa (valid cognition, sutra 1.7) is broken into three sub-types: pratyakṣa (perception), anumāna (inference), and āgama (testimony) — a smaller list than Nyaya’s six, since Yoga is mapping cognitive routes for the yogic project; a complete epistemology is the Nyaya undertaking. Viparyaya (sutra 1.8) is mistaken knowledge that does not rest on the form of its object. Vikalpa (sutra 1.9) follows from word-knowledge and is empty of object. Nidrā (sutra 1.10) is the vṛtti whose object is absence. Smṛti (sutra 1.11) is non-loss of an experienced object. The unifying frame, citta-vṛtti-nirodha at sutra 1.2, is the cessation of these modifications — the vṛtti taxonomy is the operational map of what is to be stilled.
See also Y2 (vikalpa — cognition without object), Y3 (viparyaya — mistaken cognition with a real referent), Y4 (avidyā — the master mis-taking that conditions the kliṣṭa vṛttis), and N1 (the wider pramāṇa frame, of which Yoga’s three-sub-type version is a subset).
Further Reading
Edwin Bryant, The Yoga Sutras of Patanjali, for accessible translation and commentary across the vṛtti sutras. Vyasa’s Yoga-bhāṣya on sutras 1.5–1.11 for the foundational commentarial treatment. Vacaspati Misra’s Tattva-vaiśāradī extends it. Christopher Chapple’s Yoga and the Luminous for a modern technical entry.