Background knowledge: the skill nobody lists.

Ask any hiring manager or L&D lead what they want more of, and the answers arrive in verb form: critical thinking, reading comprehension, problem-solving, learning agility. The nouns — what people actually need to know — have fallen out of fashion. They get dismissed as mere facts in an age when facts are a search away. Teach the skills, the reasoning goes, and they will apply themselves to any content.

The fashion has a cost, and it can be measured. Cognitive science has been testing the skills-first idea since the 1970s. The results are consistent, large, and largely ignored: the skills ride on the knowledge. Strip the knowledge away and the skills mostly evaporate; supply the knowledge and the skills often appear without being taught. The literature does not say facts matter instead of thinking. It says facts are largely what thinking is made of — and that has sharp consequences for how organizations build capability.

The baseball study

The cleanest demonstration cost almost nothing to run. Junior-high students were sorted two ways: strong or weak readers by standard test, and high or low baseball knowledge by a separate quiz. All of them read the same account of a half-inning of a baseball game, then rebuilt and summarized what happened. Baseball knowledge dwarfed reading ability. Weak readers who knew baseball clearly beat strong readers who didn’t — rebuilding the game’s events more accurately and summing them up more sensibly (Recht & Leslie, 1988). The “skill” the reading test measured turned out to be worth less, on this text, than knowing what a sacrifice bunt is.

The result was no fluke of sport. The same design — cross ability with knowledge, hold the material constant — has replicated across domains and ages. German researchers studied children who were soccer experts. The experts beat non-expert peers on soccer memory and understanding no matter their measured aptitude — low-aptitude experts beat high-aptitude novices (Schneider, Körkel & Weinert, 1989). The memory literature’s most famous reversal: child chess players recalled chess positions better than chess-naïve adults, upending the assumption that adult minds generally beat children’s (Chi, 1978). On domain material, the child expert’s knowledge beat the adult’s maturity.

Why knowledge does the work

The mechanism runs straight through the bottleneck our cognitive-load review describes. Working memory holds about four new elements at a time. Long-term knowledge decides what counts as an element. To a baseball novice, “the runner tagged up and advanced on the fly ball” is a dozen fragile pieces; to a fan it is one chunk, recognized rather than assembled. Chunking is how expert working memory looks superhuman while being biologically ordinary. The classic chess studies showed masters rebuilding mid-game boards at a glance — not through better memory, but through a vocabulary of tens of thousands of stored patterns, an advantage that vanished when pieces were arranged at random (Chase & Simon, 1973).

Knowledge also supplies what texts and situations leave out. Everyday speech and text are heavily compressed. Writers and speakers omit everything they assume the audience knows. So understanding is always a joint product of the message and the reader’s prior knowledge. The construction–integration account of reading formalizes this: understanding a text means building a model of the situation, and the bricks come mostly from long-term memory, with the text supplying assembly instructions (Kintsch, 1988).

Readers without the bricks can decode every word and still build nothing. That is exactly what the data show at scale. Large-sample work finds a knowledge threshold below which comprehension of a topic collapses. Near that line, background vocabulary predicts comprehension more sharply than general reading skill does (O’Reilly, Wang & Sabatini, 2019).

Even picking up new knowledge depends on knowledge. Learners who know more about a domain extract more from the same exposure. In the classic demonstration, high- and low-knowledge listeners heard the same baseball story. The high-knowledge group encoded more of the goal-relevant structure and remembered more of what mattered (Chiesi, Spilich & Voss, 1979). Knowledge compounds: the rich get richer with every text, meeting, and course. That is why early knowledge gaps widen rather than wash out.

“Critical thinking” has a domain

The skills framing survives partly because transfer feels like it ought to work. Surely a person taught to weigh evidence in one domain can weigh it in another? The expertise literature’s answer is deflating. Expert reasoning is soaked in domain content. The doctor’s diagnostic eye, the engineer’s failure instincts, the analyst’s feel for a suspicious number — all are pattern recognition against a large stored library, and they travel poorly outside it (Chase & Simon, 1973), (Hambrick & Engle, 2002).

General critical-thinking rules of thumb exist: consider alternatives, seek disconfirming evidence. But they are thin tools. They bite only when the thinker knows enough about the claim at hand to come up with the alternatives and to spot the evidence that cuts against it. The rule is one sentence; knowing what a plausible alternative looks like in this domain is the education.

Working-memory capacity — the general resource most tied to fluid reasoning — interacts with knowledge in an instructive way. Domain knowledge helps everyone, and in several designs it helps lower-capacity people most (Hambrick & Engle, 2002). Knowledge works as a capacity prosthetic, pre-building the structures that high-capacity people can partly improvise. For workforce development the upshot is almost subversive. A company cannot much change its people’s general mental horsepower. But it can absolutely change their knowledge — and knowledge is the lever that narrows performance gaps rather than widening them.

Reading the classic studies as one design

Notice that the baseball, soccer, and chess results are the same experiment wearing three uniforms — and the shared design is what makes the conclusion hard to escape. In each case, a general ability measure and a domain knowledge measure are crossed, and the material is held constant. And the domains are ones where knowledge is spread unevenly and almost by accident — nobody assigns children to be baseball fans. Across materials, ages, and countries, the knowledge factor keeps dominating on domain tasks. The interaction keeps favoring the “wrong” group: the less able but knowledgeable beat the more able but ignorant (Recht & Leslie, 1988), (Schneider et al., 1989), (Chi, 1978). A result that survives that much variation in everything except its core variables has earned the status the field gives it.

The educational fight, briefly

None of this is neutral territory. Reading instruction spent decades teaching comprehension “skills” — finding the main idea, making inferences — as content-free strategies practiced on scattered passages. The knowledge-effects literature argued the opposite: comprehension is mostly knowledge in action, so curriculum should build coherent knowledge step by step (Willingham, 2006). The strategy teaching is not worthless — explicit comprehension strategies deliver a real but limited boost that arrives quickly and then plateaus. But the knowledge base is what keeps paying, text after text, year after year. The practical synthesis: teach the strategies briefly, then spend the reclaimed time building knowledge — that is where the compounding lives (Willingham, 2006).

Corporate learning inherited the same error with less scrutiny. Course catalogues bulge with content-free skills — communication, analytical thinking, decision-making — taught through generic frameworks and practiced on toy cases. Meanwhile, the domain knowledge those skills would actually run on — the product’s design, the regulation’s structure, the market’s history — is left to build up on its own or not at all. The knowledge-effects literature predicts exactly what L&D leaders privately report. The frameworks are recalled fondly and applied rarely, because back at the desk the missing variable was never the framework.

The compounding economics of a knowledge base

Because knowledge speeds the gaining of more knowledge (Chiesi et al., 1979), investments in the base behave like capital rather than spending. They change the return on every later learning hour. The practical lessons are worth spelling out. Onboarding is the highest-leverage window an organization gets. The core schemas installed in the first months set the compounding rate for years — which is why front-loading coherent domain knowledge beats the common habit of drip-feeding disconnected procedures. And a knowledge gap left alone is not a fixed deficit; it is a growing one, because every meeting and document that assumed the missing knowledge deposited less than it should have.

Teams compound too. A group that shares core knowledge talks in compressed form; the chunks are common currency. A group with uneven bases pays a constant translation tax. Every discussion either loses the low-knowledge members or bores the high-knowledge ones into tuning out. Some of what firms experience as communication problems, meeting bloat, and “alignment” issues is measurable knowledge variance wearing a company costume. Leveling the base is a communication fix that no communication workshop can deliver.

Measuring the base, not the exposure

If knowledge is infrastructure, it deserves infrastructure-grade measurement — and course-completion records are not it. Completion certifies exposure; the knowledge-effects literature is entirely about what was retained and structured, which only direct assessment reveals. The threshold findings raise the stakes on precision. Below a knowledge floor, comprehension drops off a cliff rather than a slope (O’Reilly et al., 2019). So the difference between a learner at 60% and 75% of a role’s knowledge inventory may be the difference between functioning and floundering. No manager can see which side of the line a team member stands on without measurement built to ask.

The assessment design follows from the mechanism. The value of knowledge lies in its organization, so testing should probe structure, not just recognition. Can the learner apply the concept to a case, tell it from its neighbors, supply the missing step? Those are the levels our Bloom’s-taxonomy review maps. And because the base decays like all memory, measurement must recur; a knowledge inventory certified once is a photograph of a melting object. The organizations that treat the knowledge layer as an asset audit it like one: defined, measured, maintained, and mapped to the roles whose skills depend on it.

But can’t they just look it up?

The modern objection deserves a modern answer. Yes, facts are a search away. But the search only helps a mind that already knows enough to ask the right question, judge whether the answer is plausible, and fold it into a running line of reasoning. Understanding the retrieved page is itself knowledge-dependent (Kintsch, 1988), (O’Reilly et al., 2019). Spotting that an answer is subtly wrong for your case is pattern-matching against stored examples. And the fluent reasoning that marks the expert happens at the speed of retrieval from long-term memory, not at the speed of a query.

Looking things up is what knowledge-rich people do well. It is a complement to the internal library, and a poor substitute for it. That line has only grown sharper as AI assistants have made outside answers cheaper — and the judgment to evaluate them more valuable.

What the evidence doesn’t show

• It doesn’t show general ability is irrelevant. Reading skill, working memory, and general ability all carry independent weight; the finding is that knowledge is a comparably powerful and far more trainable factor, not the only one (Hambrick & Engle, 2002).
• It doesn’t license trivia curricula. The knowledge that pays is structured and connected — schemas, not isolated facts. A thousand disconnected product factoids chunk nothing; a coherent model of how the product works chunks everything that touches it.
• Strategies aren’t worthless. Brief explicit instruction in comprehension and reasoning strategies produces real gains; the error is spending years on what plateaus in weeks (Willingham, 2006).
• Thresholds are topic-specific. The comprehension threshold is about knowledge relevant to the material at hand (O’Reilly et al., 2019); nobody needs encyclopedic coverage of everything, which is precisely why mapping which knowledge a role actually requires is the design step that matters.

Where the evidence stops

1. 1It doesn’t show general ability is irrelevant
2. 2It doesn’t license trivia curricula
3. 3Strategies aren’t worthless
4. 4Thresholds are topic-specific

© 2026 FUTURE PROOF™

What this means for practice

The prescriptions follow from the mechanism, and the first is a translation exercise. Rewrite the capability conversation in nouns. For every skill a role demands, ask the working question: what would this person need to know for that skill to function here? The exercise converts vague competencies into teachable, testable knowledge inventories. The product’s failure modes behind “troubleshooting”; the regulation’s actual structure behind “compliance judgment”; the customer economics behind “commercial acumen.” That inventory, not the skill label, is the curriculum.

Then build the base the way the memory literature says durable knowledge is built: retrieval, spacing, and upkeep, aimed at the mapped inventory. Use diagnostics to locate each learner’s specific gaps rather than marching everyone through everything. Sequence it — knowledge compounds, so the core schemas come first and every later course gets cheaper. And measure the base directly instead of guessing it from course completions — the threshold findings warn that the gap between almost-enough and enough knowledge is the gap between struggling and understanding. Skills training has its place — on top. Underneath, there is no substitute for knowing things, and there never was.