Model-Delta UX
Version 0.2 — adopted 2026-08-28, and hardened the same day after an adversarial review by a second, independent AI whose feedback was adjudicated item by item: the accepted items are marked (v0.2) below, and the rejected ones were rejected for cause. A living document: the laws carry their evidence and their tests, which means they can be challenged by better evidence — and should be. Drafted with AI assistance from first principles; reviewed and adopted by a named human decision owner, per how the system works.
The problem it solves
This site serves cereal chemists, professional brewers, maltsters, regulators, and newly-diagnosed consumers on the same pages — and our audience policy forbids flattening them into "beginner" and "expert." The subject is genuinely deep — enzymology, malting science, process control — and the ambition is that readers understand, not merely follow.
Standard UX has no answer to this. Its instruments optimize findability, task-completion, and friction-removal. It contains no theory of learning: no account of how a mental model forms, why correct text can teach nothing, why the same explanation helps one reader and insults another, or why the smoothest page is often the worst teacher. Model-Delta UX is what UX should have been for technical learning.
The name is the objective function — and it obeys its own vocabulary law (Law 12): a philosophy needs a handle so it can be held as one idea. "Model-Delta" is the short form once you share the codebook; the "UX" tells a stranger which shelf it lives on.
Part I — The core model
Five claims. Everything else in this document is derived from them, and when a question arises that no law covers, the answer is reasoned from these claims — not from a new rule invented on the spot.
Claim 1 — Attention is the currency
A reader arrives with a finite, spendable attention budget, and everything on the page spends it: every word, heading, chart, expander label, and every choice the reader is made to consider. There is no free element. Kahneman's work on attention as a limited resource (Attention and Effort, 1973) is the root of this accounting; the neuroscience of stage magic (Macknik, Martinez-Conde et al., 2008) supplies the sharper corollary — attention is not merely limited but directable, and a page either choreographs where the reader looks or wastes the budget by accident.
Most technical sites fail here first: not for lack of information, but because they spend the reader's budget on elements that transfer nothing.
Provisional attention-cost ranking (v0.2). Until instrumentation exists, review arguments about cost use this crude ranking. It ranks waste, not effort — germane friction is spend that Law 11 protects, not a cost to minimize:
- Broken decode — a load-bearing term the declared reader cannot decode: forces lookup or exit, and often fails silently.
- Split attention — an explanation separated from its referent; the reader pays the integration cost the author skipped.
- Competing focal elements — a screenful with no choreography, every element bidding for the same fixation.
- Unanchored density — long prose with no structure signaling what carries weight.
- Unvaried redundancy — verbatim repetition that adds no error correction.
- Structure itself — headings, whitespace, familiar layout: cheap, and still never free.
Deliberately crude. It exists so a review can say "that is a rank-2 cost" instead of "this feels heavy."
Claim 2 — The reader is a state, not a demographic
At any moment, a reader is three things:
- Codebook — what is already in their head: the vocabulary and mental models the page's words will be decoded against. This is Shannon's insight (A Mathematical Theory of Communication, 1948) applied to people: compression is only possible against a shared codebook. Jargon is perfect compression for the cereal chemist and pure noise for the consumer — the sentence didn't change; the decoder did. Crucially, codebooks contain wrong entries: the experienced barley brewer decodes "mash" through priors that will actively corrupt a correct sentence. Wrong entries cannot be overwritten by stating facts (Posner, Strike, Hewson & Gertzog, 1982); they must be named and broken.
- Mode — what the reader is doing right now: learning (couch, curious, time to think), doing (wet hands, timer running, mid-procedure), deciding (which malt do I buy), or emergency (stuck mash, panic, ten-second budget). The same person occupies all four modes in one week. Mode is as fundamental an axis as expertise, and almost no existing UX framework models it.
- Budget — how much attention they will spend before leaving, which mode largely determines.
The chemist and the newly-diagnosed consumer are not different tiers — they are different codebooks. That reframe is what makes "everyone including the pros" solvable.
Claim 3 — A page is an operation on the reader
A page is not a container of information; it is an instrument that changes reader state. Its real output is the difference between the model in the reader's head on arrival and on departure. Text that displays a correct fact without changing the reader's model has an output of zero — correct, and worthless. In review, this is the killing blow, four words long: "that paragraph is zero-delta."
Claim 4 — The objective function
A page succeeds when it maximizes correct model-delta per unit of attention spent, for each reader state it declares it serves, under two hard constraints, in priority order:
- The model outranks the procedure. A reader who can follow the recipe but holds no model is fragile — the first time reality deviates (and brewing always deviates), they are lost. A reader with the model can improvise.
- But the procedure must work, and following it must teach. The surface recipe has to be followable as written, and executing it must itself install the minimum viable model — the why embedded in the doing, not adjacent to it (the smuggling principle; see Law 10).
Return-depth — the reader coming back and going deeper over months — is the third layer of success, served by zoom (Law 9) and the callback curriculum (Bet 6), but it is never an excuse to hide the model below the surface.
Claim 5 — Derived, not decreed
Every law in Part II falls out of Claims 1–4 plus cited evidence. That is what makes this a science rather than a house style: the laws can be challenged by better evidence, and new situations are answered by derivation, not by accreting rules. A law that cannot state its parent claim, its evidence, and its test does not belong in this document.
The falsification commitment (v0.2). Laws carry their own tests; the philosophy carries this one: if pages written under Model-Delta do not produce measurably higher correct model-delta per unit of attention than conventional technical writing on the same content, for the declared reader states, the philosophy is wrong. The eventual instrument is delayed prediction — after a controlled delay, can the reader correctly predict the outcome of a related scenario they have never seen? — not "did they finish the page" and not "did it feel clear." Measurement methods are deferred; the commitment is not.
Part II — The laws
Format: statement · source · test. Grouped under the claim they derive from. The tests are what make this enforceable rather than aspirational: a reviewer can apply each one to a real page and get a pass or a fail.
From Claim 1 (attention)
Law 1 — Zero-delta is a defect. Correct text that changes no model and enables no action has spent attention for nothing, and attention is the currency. Source: Claims 1 + 3 directly. Test: For any paragraph, a reviewer can name the delta — the thing a declared reader can do, predict, or explain after reading it that they could not before. No nameable delta → cut or rewrite.
Law 2 — Attention is choreographed or wasted. Every screenful has an intended focal point, chosen by the author. Elements compete whether you intend them to or not; magicians prove attention can be directed with near-certainty, and a page that doesn't direct it is leaving the choreography to chance. Source: Macknik & Martinez-Conde et al. (2008); Kahneman (1973). Test: For each screenful, the author can name where the reader's eyes should be and what earns the next fixation. "Everything is important" fails the test.
Law 3 — Kill extraneous load; chunk intrinsic load; spend germane load on purpose. Working memory holds roughly four chunks (Cowan, 2001; the older folk number is Miller's 7±2, 1956). Load comes in three kinds (Sweller's cognitive load theory, 1988): intrinsic (the topic's real complexity — manage it by chunking and sequencing), extraneous (presentation overhead — exterminate it), and germane (effort that builds the model — budget and spend it deliberately). The split-attention effect is the canonical extraneous sin: separating an explanation from the thing it explains forces the reader to pay integration costs the author should have paid. Source: Sweller (1988); Cowan (2001); worked-example and split-attention effects (Sweller & Cooper, 1985). Test: Every explanation sits adjacent to what it explains (no figure-on-another-screen, no "see below"). Every difficulty on the page is classifiable as intrinsic, extraneous, or germane; unclassifiable difficulty is extraneous by default and dies.
From Claim 2 (reader state)
Law 4 — Define the codebook before you compress. Every page declares which concepts it assumes and which it installs. A term is never "too technical" in the abstract — it is compressed against a codebook the page either verified or built. Define terms when a missing entry would block understanding, not merely because a term is technical: readers here get their intelligence respected. Source: Shannon (1948). Test: Every load-bearing term on the page is either in the assumed codebook or installed on the page before use. A term that is neither is a broken decode for some declared reader.
Law 5 — Redundancy is error correction; budget it, don't ban it. Shannon's second gift: over a noisy channel, redundancy is what gets the message through. Readers skim, get interrupted, and misread — so key models are encoded more than once. But budgeted: each re-encoding must vary (new context, new modality, new example), because verbatim repetition spends attention without adding correction. Source: Shannon (1948); encoding-variability findings in the spacing literature (Cepeda et al., 2006). Test: The page's central model appears in at least two encodings (e.g., prose + diagram, prose + worked example). No two encodings are near-verbatim.
Law 6 — Wrong priors get refutation, not correction-by-adjacency. Stating the correct fact next to a misconception does not displace it; the misconception reinterprets the fact. Refutation texts — which explicitly name the common wrong belief, state that it is wrong, and explain why — measurably outperform expository text at changing conceptions. Source: Posner et al. (1982); Guzzetti et al. (1993, meta-analysis of refutation-text studies). Test: Any page touching a known-corrupted codebook entry (chiefly: barley-brewing priors) names the old belief explicitly and breaks it. Correct prose that merely coexists with a known misconception fails.
Law 7 — Mode determines structure. Answer-first is correct for doing, deciding, and emergency modes — the reader needs the payload now, and anything before it is extraneous load. Setup-then-payoff is also correct — in learning mode, where tension makes the reader generate the answer themselves a beat before it lands (the generation effect: Slamecka & Graf, 1978; a comedian's setup/payoff is the same mechanism on stage). These structures conflict only if mode is invisible. It isn't: every page declares which modes it serves, and structure follows. Source: Slamecka & Graf (1978); checklist design for interrupted execution (Degani & Wiener, 1993). Test: The page's declared mode(s) predict its structure. A doing-mode page that opens with theory fails; a learning-mode page may open with a question. Declaration rule (v0.2): a page declares one primary mode. Secondary modes are permitted only where the structural requirements do not conflict, or where a compiled rendering (Bet 1) serves them separately. "All four" is not a declaration; it is a refusal to decide, and it fails this test.
Law 8 — Scaffolding that helps a novice harms an expert. The expertise-reversal effect: guidance, worked examples, and step-by-step support measurably slow and degrade readers whose codebooks already contain the material. Serving the full audience range on one page therefore cannot be done by adding scaffolding — only by zoom (Law 9), where each reader enters at their scale. Source: Kalyuga, Ayres, Chandler & Sweller, "The Expertise Reversal Effect" (2003). Test: No mandatory-path scaffolding: nothing forces the expert reader through novice support to reach their content. Support is enterable, not unavoidable.
From Claim 4 (the objective function)
Law 9 — Depth is zoom, not appendix. Cartography solved our audience problem centuries ago: one map serves the day-hiker and the geologist through declared scale and principled generalization — what to omit at each zoom level is a rule, not taste. Museum interpretation found the same structure independently (Tilden, Interpreting Our Heritage, 1957): the 3-second, 30-second, and 3-minute reader served by one label. Therefore: every layer of a page is a complete, correct model at its declared scale — coarser, never wronger. Depth zooms into a model the surface already installed; it never contains the model's only copy. Source: cartographic generalization practice; Tilden (1957). Test: A reader who consumes only the surface layer leaves with a complete, correct, coarse — and predictive — model: from the surface alone, they can make at least one non-obvious prediction about what happens when a controlling variable changes (v0.2). A surface that only describes fails, and any "why" that exists only behind an expander or link fails.
Law 10 — The surface teaches through the doing (smuggling). Great game tutorials never show a manual: Portal and Mario's first level teach rules by making the player act under constraints that make the rule felt. Military field manuals encode the same trick in language — "do X so that Y; if Z, then W" — conditionality carries the causal model inside the procedure. Recipes and protocols are written so that following them installs the why. Source: game tutorial design practice (Koster, A Theory of Fun, 2004; flow: Csikszentmihalyi, 1990); military field-manual style; worked-example research (Sweller & Cooper, 1985). Test: Each procedure step that embodies a load-bearing cause states it inline ("hold 190°F for 60 minutes so that …"). A procedure that is pure imperative — do this, do this, do this — fails. Conceptual pages (v0.2): pages with no procedure smuggle through their own first-class toolkit — prediction before reveal, worked-then-faded examples, explorable models, and contrast cases (the instance that breaks the pattern teaches the pattern). These are the learning-mode equivalents of procedure smuggling, not its poor relations.
Law 11 — Fluency is not learning; engineer germane friction. A page that reads smoothly feels effective and often isn't — fluency inflates judgments of learning, while retention comes from retrieval, generation, and spacing: Bjork's "desirable difficulties." Meanwhile usability doctrine says minimize friction. Both are right about different frictions, so the law is a taxonomy: extraneous friction is exterminated (Law 3); germane friction — moments where the reader predicts, retrieves, or generates — is engineered in on purpose. Mainstream UX does not make this distinction; it is this philosophy's sharpest break with the field. Source: Bjork (1994, desirable difficulties); Roediger & Karpicke (2006, the testing effect); Slamecka & Graf (1978, generation effect). Test: In learning mode, each major concept has at least one engineered generation or retrieval moment (a prediction prompt, a worked-then-faded example, a callback). And every friction point on the page can be defended as germane; if not, it is extraneous and dies under Law 3.
Law 12 — Name what readers must reason with. A name is the ultimate codebook entry: one token that collapses a model into a single chunk, and chunks — not facts — are what working memory reasons with (Chase & Simon's chess studies, 1973). Un-named models can be read but not carried. Every concept readers are expected to use eventually gets a handle; installing that handle is a deliberate page event, not decoration. Source: Chase & Simon (1973); Cowan (2001). Test: Any concept referenced across three or more pages has a stable, installed name. (This document obeyed its own law: "Model-Delta" exists so the philosophy can be held as one chunk.)
Law 13 — Encode data down the perception hierarchy. Human graphical perception is measured, not a matter of taste: judgments are most accurate from position along a scale, then length, then angle/slope, then area, then color saturation. Which numbers become a chart, a table, or a sentence is decidable by rule. Source: Cleveland & McGill, "Graphical Perception" (1984); Tufte, The Visual Display of Quantitative Information (1983). Test: Quantitative comparisons the reader must make are encoded as position or length, not area or hue. Decoration that encodes nothing is extraneous (Law 3). Misencoded data is extraneous load delivered visually (Law 3), and well-encoded data is zoom applied to numbers (Law 9) (bridge added v0.2).
Law 14 — The reader always knows where they are. Wayfinding: on any page, the reader can tell what is assumed of them, where they are in the larger territory, and where one level deeper lives. Recognition beats recall (Nielsen's usability heuristics, 1994); Alexander's pattern language (1977) supplies the deeper standard — a place is legible when you can navigate it without a guide. Source: Nielsen (1994); Alexander, Ishikawa & Silverstein (1977). Test: From any page, without using search or the back button, the reader can reach the assumed-prerequisite material and the next level of depth, and can state what the page expects them to already know — stated visibly on the page, not discoverable only by search or by failing (sharpened v0.2).
Law 15 — Design the failure path first. Aviation and anesthesia learned it in blood: procedures are designed from the question "what goes wrong, at which step, and what does the operator see when it does" — before the happy path is polished. Our highest-stakes reader is mid-failure (stuck mash, emergency mode, ten-second budget), and pages that serve them are written backwards from the failure. Failure modes are also where real production experience outshines theory: reality's deviations are what only experience can document — and the no-invention rule binds absolutely (a failure mode the record doesn't support is never invented to satisfy this law). Source: Degani & Wiener (1993); Gawande, The Checklist Manifesto (2009). Test: Every procedure page names its known failure modes, what the reader observes when each occurs, and the recovery — reachable within the emergency-mode budget (seconds, not paragraphs).
Law 16 — Formalize progressively: felt model first, formalism as its compression. (Added v0.2.) Begin with the causal, almost physical model; introduce the quantitative or formal version afterward, presented as a compression of the model the reader already holds — never as a replacement that orphans it. Most technical writing runs the reverse order and pays for it: formalism decoded without a causal referent is stored as symbols, not as a model. The concreteness-fading literature shows the concrete-to-abstract sequence outperforms abstract-first instruction. Source: Fyfe, McNeil, Son & Goldstone (2014, systematic review of concreteness fading); schema theory generally. Test: Every formal expression (equation, quantitative table, spec) is preceded — on the same page or on the declared prerequisite path — by the causal model it compresses, and the page says what the formalism compresses.
Law 17 — Agency survives the encounter. (Added v0.2.) The reader's state includes their sense of capability, and the page operates on it (Claim 3). Difficulty is stated honestly — "this is harder than barley brewing" is allowed and often true — but every failure path and every refutation is written so the reader leaves with more agency, not less: the failure is information, the prior was an optimization for different constraints, the next action exists. "You are defective for finding this hard," in any phrasing, is a defect. Self-efficacy predicts persistence, and a diminished reader does not return — which forfeits Claim 4's third layer. This law commands agency, not cheerleading: praise, pep, and false inclusiveness remain voice defects. Source: Bandura (1997, self-efficacy); Kapur (2008, productive failure). Test: For each failure path and refutation block, the reader's available next action is stated, and the text attributes difficulty to the domain or to the transferred prior — never to the reader.
Part III — The nine bets
Design commitments the laws imply. Bets 1–3 are canonical — the philosophy stands behind them as first-class commitments. Bets 4–9 are experimental — named, derivable, and adopted, promoted to canonical as they prove out on real pages.
Canonical
Bet 1 — The mode compiler. Responsive design was the last true paradigm shift in web UX: one source, rendered per screen. The obvious unbuilt next step: one canonical source, rendered per cognitive state. A protocol page exists once and compiles into a study view (full causal chains, tension-and-payoff structure, deep zoom), a brew-day view (checklist typography, one step in focus, large type readable at arm's length with wet hands, timers, refutation warnings surfaced at the exact step where old instinct betrays), and a decide view (the comparison first, reasoning behind it). Responsive to the mind, not the viewport. Derives from: Claim 2 (mode), Laws 7 and 15.
Bet 2 — The codebook as compilable infrastructure. A machine-readable concept graph in which every page declares which concepts it assumes and which it installs. Prerequisites become computable: a checker catches a page assuming "diastatic power" before any reachable page has installed it; "read this first" on-ramps generate themselves; known-corrupted entries (barley priors) live in the graph flagged, so every page touching them knows to refute (Law 6). Dependency management for knowledge — package-manager semantics applied to understanding. Derives from: Claim 2 (codebook), Laws 4, 6, 12, 14. Success condition (v0.2): the graph exists to make broken decodes detectable — a wrong assumption, a missing install. It is not a forced curriculum. The moment it makes an expert walk a prerequisite path to reach their content, it has reintroduced expertise reversal (Law 8) and failed.
Bet 3 — The expander, redeemed. Progressive disclosure's documented failure is that hidden "why" never gets read — the model must live on the surface (Law 9). The synthesis: make the click itself a teaching event. The expander label stops being a filing label ("More details") and becomes a prediction prompt: "Beta-amylase is the enzyme making your beer fermentable — what do you think 190°F does to it?" The reader generates an answer before the reveal, and the generation effect makes the reveal encode far stronger than plain reading. The same component, transformed from filing cabinet into instrument. Derives from: Laws 9 and 11.
Experimental
Bet 4 — The barley-instinct block. Unlearning as a named, visible page primitive: a recurring block — "If you've brewed barley, your instincts are wrong here, and here's specifically how" — deployed at the exact points where transferred expertise corrupts. Refutation in editorial form; it also respects the expert instead of re-teaching them basics, which is the expertise-reversal fix (Law 8) turned into a feature. Derives from: Laws 6 and 8. Grammar (v0.2, per Law 17): the block's stance is "your prior was optimized for a different grain and a different constraint set — here is exactly where the optimization breaks." Respect plus refutation, never correction-as-diminishment. The block deployed on Mash Protocol 1 at the 190°F step ("That instinct is correct for barley — and exactly backwards here") is the reference implementation.
Bet 5 — Explorable models. Don't describe the enzyme-temperature curve — hand the reader a slider and let them drag the mash temperature and watch the enzymes die. Thirty seconds of play builds a causal model three paragraphs can't, because the reader acts under constraints that make the rule felt — smuggling (Law 10) in its purest form. Costly per page, so reserved for load-bearing concepts; the Mash Protocol 1 baseline is the first candidate. Source: Bret Victor, "Explorable Explanations" (2011). Derives from: Laws 10 and 11.
Bet 6 — The callback curriculum. Spaced repetition at site scale, worn like a comedian's callback: a concept installed on one page is deliberately re-surfaced on later pages in new contexts — not re-explained, called back. The concept graph (Bet 2) makes the schedule plannable. The site stops being a pile of pages and becomes a syllabus that doesn't know it's one. Source: the spacing effect (Ebbinghaus, 1885; Cepeda et al., 2006). Derives from: Laws 5 and 11; Claim 4's return-depth layer.
Bet 7 — The attention ledger. Claim 1 made auditable: measurable page properties — words spent before the first model-delta, attention cost per section against what that section transfers. The boring bet that keeps the other eight honest: it catches the beautiful page that teaches nothing. Derives from: Claim 1, Law 1.
Bet 8 — Sensory calibration pages. Brewing is a perceptual craft, and perception is teachable — but only through vocabulary plus guided reference: you cannot recognize a fault you have no word and no anchor for. A content type whose entire job is calibrating the reader's senses — what a stuck sorghum fermentation smells like, what proper conversion looks like on an iodine test — keyed to moments in our own protocols. Wine education proved the mechanism (Ann C. Noble's Wine Aroma Wheel, UC Davis); brewing education has never done it deliberately at this level. It is also the one need the whole audience range shares: the chemist's nose needs calibrating to sorghum as much as the novice's. Perceptual-learning research strengthens the claim beyond vocabulary: discrimination itself is trainable (Goldstone, 1998) — the calibration page trains the perception, and the word is its handle (v0.2). Derives from: Law 12 applied to percepts; Claim 3.
Bet 9 — Failure-path-first authoring. Law 15 elevated from page property to authoring order: for every procedure, the failure modes are gathered and written before the happy path is polished — and the minimum viable predictive surface (Law 9, v0.2 test) is drafted alongside them. The stuck-mash reader is the first user story, not an edge case. Derives from: Laws 9 and 15; Claim 2 (emergency mode).
Part IV — The conference
The philosophy was forged by convening the disciplines that each solved a piece of the problem — a conference that never physically met, but whose seats are real bodies of work. The roster is preserved because the lineage is the why: every law above traces to a seat, and a challenge to a law should start by re-reading its seat's evidence.
Table 1 — The physics: what information transfer is. The information theorist (Shannon, 1948): readers as noisy channels, codebooks, budgeted redundancy. The cognitive load theorist (Sweller, 1988; Kalyuga et al., 2003): three kinds of load, roughly four chunks of working memory (Cowan, 2001), the expertise-reversal effect. The learning scientist (Bjork, 1994; Roediger & Karpicke, 2006): the fluency illusion, desirable difficulties, the testing effect. The conceptual-change researcher (Posner et al., 1982; Guzzetti et al., 1993): refutation, unlearning as a first-class operation.
Table 2 — The cartographers: density for everyone at once. The master cartographer: declared scale, principled generalization, the legend as explicit codebook. The museum exhibit designer (Tilden, 1957): layered labels, interpretation over information. The data-graphics statistician (Cleveland & McGill, 1984; Tufte, 1983): the perception hierarchy. The pattern-language architect (Alexander et al., 1977): wayfinding, and the very form of a system of named, composable resolutions to recurring tensions.
Table 3 — The magic users: attention, motivation, smuggling. The stage magician (Macknik & Martinez-Conde et al., 2008): attention as choreography. The game tutorial designer (Koster, 2004; Csikszentmihalyi's flow, 1990): the tutorial is the game; difficulty curves. The comedian: setup/payoff as the generation effect on stage (Slamecka & Graf, 1978); the callback as disguised spaced repetition. The sensory educator (Noble's aroma wheel): vocabulary calibrates perception.
Table 4 — The high-stakes writers: when reading fails, things break. The aviation/anesthesia checklist designer (Degani & Wiener, 1993; Gawande, 2009): reading modes, design for interruption, the failure path. The military field-manual writer: conditionality carries the model — procedures that teach because "so that" and "if/then" are load-bearing.
And one seat added after the opening session: the explorable-explanations designer (Bret Victor, 2011) — reactive documents where the reader's manipulation is the explanation.
Two seats added in v0.2, after adversarial paired review: the deliberate-practice researcher (Ericsson, Krampe & Tesch-Römer, 1993) — how models refine under repeated, high-quality failure and feedback, the science behind Law 11's difficulties and Claim 4's return-depth layer; and the perceptual-learning researcher (Goldstone, 1998) — discrimination is trainable, the science behind Bet 8. The self-efficacy and productive-failure evidence (Bandura, 1997; Kapur, 2008) entered with Law 17, and the concreteness-fading evidence (Fyfe et al., 2014) with Law 16.
Anti-patterns
(Added v0.2.) The most seductive corruptions of this philosophy, named so review can call them by name:
- The buried model — zoom used to hide the only copy of the why behind an expander. Law 9 inverted; the failure progressive disclosure was already famous for.
- Mandatory scaffolding — novice support the expert cannot route around. Law 8 reversed.
- "Everything is important" — no focal choreography; every element shouting. Law 2 abandoned.
- Fluency theater — a page optimized to feel smooth and judged by how smooth it feels. Law 11's fluency illusion adopted as a goal.
- Friction cosplay — difficulty with no model-building defense, excused as "desirable." Law 11 abused; extraneous by default under Law 3.
- Prompt quotas — generation moments required per page by count, producing performative questions. Law 11 templated.
- The pep talk — Law 17 read as an encouragement duty; praise and cheerleading instead of agency. A voice defect wearing Model-Delta's badge.
- Curriculum handcuffs — the codebook graph enforcing a linear path instead of detecting broken decodes. Bet 2's stated failure condition.
- Invented failure — failure modes or sensory anchors the record does not support, written to satisfy Laws 15/17 or Bet 8. Never-invent outranks every law in this document.
Boundaries
- This philosophy governs presentation, not truth. It does not override this site's evidence standards, safety rules, or review discipline. A page that teaches brilliantly and states something the evidence doesn't support is a defect of the highest order — the no-invented-experience rule binds every bet above.
- Adopted, not finished. Version 0.2 is the working philosophy; applying it to pages — templates, the mode compiler, the concept graph — is ongoing work, and the laws will be revised where real pages prove them wrong. When a law changes, the change and the evidence that forced it will be published here, per how the system works.
- Portable in principle, proven here. (v0.2.) The science is content-agnostic within deep technical learning: the claims and laws mention brewing only in their examples. Applied elsewhere, the codebook, the corrupted priors, the modes, and the failure modes change; the mechanics do not. This site is the proving ground, not the boundary.
References
- Alexander, C., Ishikawa, S., & Silverstein, M. (1977). A Pattern Language. Oxford University Press.
- Bandura, A. (1997). Self-Efficacy: The Exercise of Control. W. H. Freeman.
- Bjork, R. A. (1994). "Memory and metamemory considerations in the training of human beings." In Metacognition: Knowing about knowing. MIT Press.
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). "Distributed practice in verbal recall tasks: A review and quantitative synthesis." Psychological Bulletin, 132(3).
- Chase, W. G., & Simon, H. A. (1973). "Perception in chess." Cognitive Psychology, 4(1).
- Cleveland, W. S., & McGill, R. (1984). "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods." Journal of the American Statistical Association, 79(387).
- Cowan, N. (2001). "The magical number 4 in short-term memory: A reconsideration of mental storage capacity." Behavioral and Brain Sciences, 24(1).
- Csikszentmihalyi, M. (1990). Flow: The Psychology of Optimal Experience. Harper & Row.
- Degani, A., & Wiener, E. L. (1993). "Cockpit checklists: Concepts, design, and use." Human Factors, 35(2).
- Ebbinghaus, H. (1885). Über das Gedächtnis (Memory: A Contribution to Experimental Psychology).
- Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). "The role of deliberate practice in the acquisition of expert performance." Psychological Review, 100(3).
- Fyfe, E. R., McNeil, N. M., Son, J. Y., & Goldstone, R. L. (2014). "Concreteness fading in mathematics and science instruction: A systematic review." Educational Psychology Review, 26(1).
- Gawande, A. (2009). The Checklist Manifesto. Metropolitan Books.
- Goldstone, R. L. (1998). "Perceptual learning." Annual Review of Psychology, 49.
- Guzzetti, B. J., Snyder, T. E., Glass, G. V., & Gamas, W. S. (1993). "Promoting conceptual change in science: A comparative meta-analysis of instructional interventions from reading education and science education." Reading Research Quarterly, 28(2).
- Kahneman, D. (1973). Attention and Effort. Prentice-Hall.
- Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). "The Expertise Reversal Effect." Educational Psychologist, 38(1).
- Kapur, M. (2008). "Productive failure." Cognition and Instruction, 26(3).
- Koster, R. (2004). A Theory of Fun for Game Design. Paraglyph Press.
- Macknik, S. L., King, M., Randi, J., Robbins, A., Teller, Thompson, J., & Martinez-Conde, S. (2008). "Attention and awareness in stage magic: turning tricks into research." Nature Reviews Neuroscience, 9(11).
- Miller, G. A. (1956). "The magical number seven, plus or minus two." Psychological Review, 63(2).
- Nielsen, J. (1994). "10 Usability Heuristics for User Interface Design." Nielsen Norman Group.
- Noble, A. C. (1984, revised 1990). The Wine Aroma Wheel. University of California, Davis.
- Posner, G. J., Strike, K. A., Hewson, P. W., & Gertzog, W. A. (1982). "Accommodation of a scientific conception: Toward a theory of conceptual change." Science Education, 66(2).
- Roediger, H. L., & Karpicke, J. D. (2006). "Test-enhanced learning: Taking memory tests improves long-term retention." Psychological Science, 17(3).
- Shannon, C. E. (1948). "A Mathematical Theory of Communication." Bell System Technical Journal, 27.
- Slamecka, N. J., & Graf, P. (1978). "The generation effect: Delineation of a phenomenon." Journal of Experimental Psychology: Human Learning and Memory, 4(6).
- Sweller, J. (1988). "Cognitive load during problem solving: Effects on learning." Cognitive Science, 12(2).
- Sweller, J., & Cooper, G. A. (1985). "The use of worked examples as a substitute for problem solving in learning algebra." Cognition and Instruction, 2(1).
- Tilden, F. (1957). Interpreting Our Heritage. University of North Carolina Press.
- Tufte, E. R. (1983). The Visual Display of Quantitative Information. Graphics Press.
- Victor, B. (2011). "Explorable Explanations." worrydream.com.