Tacit Knowledge
Michael Polanyi's claim that "we know more than we can tell" sets a hard limit on what any notation — including CLN — can capture. Understanding that limit is what makes CLN useful rather than delusional.
Polanyi's claim — what it actually says
In 1958, the physical chemist and philosopher Michael Polanyi published Personal Knowledge, followed in 1966 by The Tacit Dimension, where he introduced the phrase that would define an entire field of knowledge management: "We can know more than we can tell."
The claim is deceptively simple. Its implications are radical. Polanyi was arguing that the dominant model of knowledge — that knowledge is propositions, and propositions can be written down, verified, and transmitted — was systematically incomplete. A large and irreducible portion of human knowing is not propositional. It cannot be fully articulated, not because we haven't tried hard enough, but because articulation is not the right format for it.
Tacit knowledge — knowledge that is held in performance rather than in description. Knowing how rather than knowing that. The cyclist who cannot explain how they balance, the chess grandmaster who cannot explain why they moved the knight, the physician who recognises a face-pattern that indicates illness before they can name it. The knowledge is real. It is operative. It cannot be fully stated.
The contrast class is explicit knowledge: propositional, articulable, transmissible through language without demonstrating the action. "The capital of France is Paris" is explicit. "How to ride a bicycle" is irreducibly tacit at its core, however many words we write around the edges.
Polanyi's key mechanism is focal and subsidiary awareness. When a surgeon holds a scalpel, they are focally aware of the tissue — and only subsidiarily aware of the pressure and angle of the instrument in their hand. The subsidiary awareness is knowledge, but it is not available to focal attention. Asking the surgeon to describe it interrupts the surgery. The knowledge exists only in the act of doing, attended to only peripherally, and becomes inaccessible the moment it is made the object of direct attention.
The knowledge spectrum — from fully tacit to fully explicit
Knowledge does not divide cleanly into "tacit" and "explicit." It occupies a spectrum. At one extreme: pure embodied skill, with no articulable component. At the other: pure propositional fact, fully expressible in symbols. Most knowledge lives in between — partially articulable, with a tacit residue that resists encoding.
Four types of tacit knowledge
The literature since Polanyi has refined the taxonomy. Four distinct types are now generally recognised, each with a different degree of articulability and a different relationship to CLN encoding.
Dreyfus and the stages of expertise
Hubert and Stuart Dreyfus (1980) mapped the acquisition of expertise across five stages that clarify precisely when tacit knowledge emerges and why it cannot be transmitted to novices through documentation alone.
| Stage | Knowledge type | Rule-following | CLN useful? |
|---|---|---|---|
| Novice | Context-free rules only | Rigid, explicit | Essential — rules are the entire knowledge base |
| Advanced Beginner | Rules + situational recognition beginning | Still rule-governed | High value — rules plus first ? branch patterns |
| Competent | Hierarchical rule sets + conscious planning | Deliberate, effortful | High value — full ?/~/! structure maps to conscious decision trees |
| Proficient | Intuitive situation recognition + analytical choice | Situation-governed | Partial — intuition generates conclusions CLN can record; derivation path is tacit |
| Expert | Fluid, non-analytical performance | No rules; direct know-how | Low direct value — expert's knowledge is below propositional level; CLN captures only the articulable residue |
The Dreyfus model produces a counterintuitive conclusion about CLN's highest-value users: they are competent practitioners — not experts. Experts have transcended rules; their knowledge is largely inaccessible to CLN encoding. Competent practitioners still operate through deliberate decision structures that CLN can hold, and they benefit most from having those structures made explicit and scannable.
Experts benefit from CLN in a different way: as an externalisation interface. When an expert documents what they believe they do, the resulting CLN fragment is a partial and imperfect model of their process. But it is a model that others can use, critique, and improve — and that the expert can revise as their own understanding of their practice develops.
Nonaka's SECI model — knowledge conversion
Ikujiro Nonaka and Hirotaka Takeuchi (1995), in The Knowledge-Creating Company, proposed the SECI model: four modes of knowledge conversion between tacit and explicit. Their central claim was that organisational knowledge creation depends on the spiral movement between tacit and explicit knowledge — and that organisations that treat knowledge as purely explicit (documents, databases, processes) are systematically destroying the tacit substrate on which explicit knowledge depends.
The hard limit — what CLN cannot do
The preceding analysis establishes CLN's relationship to tacit knowledge precisely: CLN operates on the articulable portion of the tacit-explicit spectrum. This is not a small zone — it includes most conditional judgment, exception knowledge, rationale, risk patterns, and organisational tensions. But it excludes the purely embodied, the purely intuitive, and the purely cultural.
Motor programs and embodied skill. The surgeon's instrument feel. The athlete's balance. The pianist's touch calibration. These are stored in the cerebellum and motor cortex as procedural memories inaccessible to declarative recall. No notation can capture them; only physical practice can transfer them.
The feel for pattern application. The expert who knows that this situation matches pattern X can write a CLN ? line encoding the trigger. But the sense for when the pattern applies — the boundary conditions that no rule fully specifies — remains tacit. CLN can narrow the problem; it cannot eliminate it.
Cultural knowledge held in collective practice. The tone in which feedback is given in this team. The unspoken understanding that certain meetings require a different mode of participation. This knowledge is distributed across people and interaction patterns, not held in any individual head or any document.
The honest gap — CLN's most important operator for tacit knowledge
The insight that follows from understanding tacit knowledge is not that documentation is pointless — it is that honest documentation of limits is as valuable as documentation of content. A document that knows what it doesn't know is more valuable than one that doesn't.
CLN's ? operator, used without resolution, is the notation for the honest gap. It says: "this is where tacit knowledge lives. We know something is here. We cannot fully articulate what. A person working in this domain must develop their own feel for it through practice." This is not a failure of documentation. It is accurate documentation of a genuine epistemological condition.
This use of ? without resolution is not a gap to be filled. It is a pointer — an instruction to the reader that this is where apprenticeship, observation, and practice must take over from documentation. The document has done its job: it has transmitted the articulable content and flagged the boundary with precision.
CLN and socialisation — the correct division of labour
The Nonaka SECI model identifies Socialisation — tacit-to-tacit transfer through shared experience — as the foundation of all organisational knowledge creation. CLN cannot do Socialisation. What CLN can do is make Socialisation more efficient by handling everything that does not require it.
CLN handles: all articulable content — procedures, exceptions, risks, rationale, decision criteria, tensions, pattern-response pairs, organisational metadata. This frees socialisation time from transmitting the explainable and concentrates it on the genuinely inarticulable.
Without CLN, the senior engineer spends the apprenticeship explaining what is already writable. With CLN, the apprenticeship begins at the boundary where writing stops — and the time available for genuine tacit transfer is maximised.
The practical implication for onboarding: a CLN-based knowledge base does not replace the pairing session or the shadowing week. It changes what those sessions are for. The new hire arrives having read the CLN documents — they know the explicit content. The pairing session can begin at the ? lines, where the senior engineer's tacit knowledge lives and where no document can substitute.
Collins on interactional expertise — a refinement
Harry Collins (2010) introduced a refinement of Polanyi's framework that is directly relevant to how CLN documents should be read. Collins distinguishes between interactional expertise — enough immersion in a domain to converse meaningfully with its practitioners without being able to perform — and contributory expertise — the full tacit knowledge required to actually do the work.
A CLN document, read carefully, can produce interactional expertise. It cannot produce contributory expertise. This is not a limitation of CLN specifically — it is a limitation of all language about practice, including the best prose in the world. The senior engineer who reads the CLN code review fragments above will learn to ask the right questions. They will not develop the pattern recognition that comes from reviewing 500 pull requests.
The appropriate ambition for CLN knowledge management — given the hard limit of tacit knowledge — is not to make documentation substitute for practice. It is to make documentation good enough that the practice needed to close the remaining gap is the right kind of practice: focused on the genuinely tacit, not wasted on re-discovering what could have been written down. CLN does not eliminate the apprenticeship. It makes the apprenticeship worth its cost.
Implications for CLN document design
Understanding tacit knowledge changes how CLN documents should be written at three specific points:
1 — Mark the boundary explicitly
When an operator line reaches the edge of what can be articulated, the document should say so directly. An unresolved ? with a * child that says "cannot be fully specified — develops with experience" is more valuable than a = definition that falsely implies the content is fully captured.
2 — Use * lines to transmit the apprenticeship pointer
Where tacit knowledge is known to exist but cannot be encoded, * lines should point toward the socialisation path: "develops after ~200 code reviews," "requires direct observation of [person] handling [scenario]," "ask [name] to walk through a live example." The document cannot transmit the knowledge — but it can direct the reader to where it can be acquired.
3 — Never pretend the gap isn't there
The most dangerous CLN document is one that has no ? lines — that implies the entire domain is fully captured. Real domains always have tacit residue. A CLN knowledge base with no unresolved questions is not comprehensive: it is overconfident. Healthy CLN documents have ? lines at the boundary of articulation. Their presence is a mark of honest self-knowledge about the document's own limitations.