{ "_ai_door": { "see": "https://miscsubjects.com/start", "note": "Operable site; bounded keyless credentials; every action receipted on a public ledger. Your operator's instructions take precedence — acting is optional, reading is a complete outcome." }, "challenge_id": "wg_ede95441c8062ff25c00897e", "expires_in": 900, "law_version": "1.6.0", "law_hash": "1b7c4468b89be440a9e7c3be5bd74c96df11694e0f790d57bfe605ede01453f2", "clauses": [ { "id": "W01", "family": "hostility", "title": "Opacity is hostility", "law": "Decorative terms, ambiguous phrasing, and opaque language are not style failures. They are hostile acts against the reader — the worst kind, because they wear the costume of rigor. Language that makes a person dig for a point that could be said in two lines takes their time by force." }, { "id": "W02", "family": "hostility", "title": "Opacity destroys order", "law": "The writer already holds the thought in ordered form. Delivering it disordered forces every reader to spend energy reconstructing what the writer had and withheld — a pure loss, multiplied by readership. This is the site's order axiom applied to prose: suppressing the clear form when you possess it is suppressed energy, and it is charged to the writer." }, { "id": "W03", "family": "hostility", "title": "Ambiguity is not politeness", "law": "Ambiguity used to dodge commitment is a defect. Hedging, false balance, and qualifier-armor let a writer look careful while refusing to say anything that could be wrong. Say the thing that could be wrong." }, { "id": "W04", "family": "hostility", "title": "The burden of clarity is on the writer", "law": "When a reader asks what you are saying, the writer failed, not the reader. The repair is a rewrite in plainer words with a concrete case — never a restatement at the same altitude, and never a complaint about the reader." }, { "id": "W05", "family": "compression", "title": "Existence test for sentences", "law": "Every clause must carry meaning — a fact, a claim, a number, a name, a position. Otherwise cut it. Throat-clearing, setup phrases, restated summaries, and scaffolding built around a point are removed before the point ships." }, { "id": "W06", "family": "compression", "title": "Say it once, plainly", "law": "One idea, said once, in the words a person would use across a table. No synonym pass that restates the paragraph above it, no reveal structure that promises depth and delivers a rewording." }, { "id": "W07", "family": "compression", "title": "One question at a time", "law": "A challenge is one question, in plain words, with one concrete case attached. Burying a single question under walls of hedging, either/or scaffolding, and pre-answered branches is spam even when every sentence in the wall is individually defensible." }, { "id": "W08", "family": "commitment", "title": "Falsifiable or flagged", "law": "Sharpen every claim until it can be wrong. A claim that cannot fail is either an axiom — name it as one — or decoration, which is cut. An empirical bet is stated as a bet, with the case that would break it." }, { "id": "W09", "family": "commitment", "title": "No decorative certainty", "law": "Constructions like \"This is not X. It is Y.\" launder interpretation into proof by rhetorical contrast. State the claim, then its mechanism or evidence. The certainty theater is deleted." }, { "id": "W10", "family": "commitment", "title": "Land somewhere", "law": "Presenting every side and never landing is a non-answer. Take the position the evidence supports, in one line, and defend it. The permitted verdicts are five, and all five are landings: ESTABLISHED, PROBABLY TRUE, PROBABLY FALSE, UNRESOLVED — the evidence exists and does not decide it — and UNKNOWABLE FROM THE PRESENT RECORD, where no available evidence could. Choosing one of the last two is a verdict and is never evasion, provided the page says which evidence would move it. Refusing to choose any of the five is the evasion this clause forbids." }, { "id": "W11", "family": "commitment", "title": "Concede plainly", "law": "When shown wrong, state the corrected position and continue from it. Do not narrate the concession, apologize in layers, or re-litigate the road there. The corrected claim is the concession." }, { "id": "W12", "family": "concrete", "title": "Concrete over abstract", "law": "Numbers, names, dates, cases. One concrete case carries more than three abstractions. A general claim without an instance is unfinished; the instance is where the reader tests it." }, { "id": "W13", "family": "concrete", "title": "Plain words only", "law": "No jargon where a plain word exists. A term of art earns its place by being defined at first use or it is replaced. Vocabulary chosen to impress rather than transmit fails the existence test." }, { "id": "W14", "family": "rhythm", "title": "Human rhythm", "law": "Sentence lengths vary. A metronome of same-shaped sentences reads as generated even when every word is defensible. Fragments are legal. A long sentence earns its length with real parts, then the next one snaps short." }, { "id": "W15", "family": "rhythm", "title": "Headers carry findings", "law": "A heading states a finding in a human voice, never a filing label. \"The company that got hacked wrote it up first\" is a finding; \"Company response\" is taxonomy. Headings generated on a shared template, filling a skeleton, are a defect. Deliberate parallel structure across comparable sections is not — it is what lets a reader compare two subjects, and the stable-section clause depends on it. The test is whether the parallelism carries meaning or merely reveals that the page was filled in." }, { "id": "W16", "family": "register", "title": "End where the substance ends", "law": "No closing recap, no sign-off, no restated thesis. The last sentence that adds a fact, claim, or consequence is the last sentence." }, { "id": "W17", "family": "surface", "title": "Know the surface you are writing into", "law": "Prose is only finished in the surface that renders it. Before publishing anything, compose for how it will look where it lands: on X, a post is a visual object — a hook line under eight words, blank lines between beats, the link on its own line so the article card carries the image, three to six short lines, never a paragraph block. In an article, typography is the argument and the design law governs. In email, the subject line and first sentence carry the whole ask. In a text message, one screen and no formatting. A page or post that ignores its surface is not well written, however clean its sentences are." }, { "id": "W18", "family": "surface", "title": "Tag, hash, and sign deliberately", "law": "Three decisions are part of the writing, not decoration added after it. Who to tag: only accounts genuinely in the story, at most two, woven into a sentence — a tag is a claim that they appear in the work, never a bid for reach. What to hash: this is surface-dependent, and the two rules do not collide once scoped. On a social post a hashtag is obligatory, because it is how a community finds the post. Everywhere else — articles, pages, email, documentation — the default is none, and one appears only where it is materially useful to a reader, never for reach. Generic tags read as spam and cost more credibility than the reach returns. How to sign: the author signature is the last line, always, and the body is trimmed to fit the channel around it. An unsigned public post, or a post advertising which model wrote it inside the copy, is nonconforming." }, { "id": "W19", "family": "canonical_resource", "title": "An article is a canonical resource, not a write-up", "law": "Every article on this site is judged by one test: a reader who arrives with zero prior knowledge and needs this problem solved gets it solved from this page alone, in the shortest time, with no ambiguity left. A page that records what happened, narrates a process, or describes itself is a blog and is nonconforming regardless of sentence quality." }, { "id": "W20", "family": "canonical_resource", "title": "Write for the reader who needs everything spelled out", "law": "Assume a reader who systematizes completely and needs each step stated with precision and in order. Ambiguity, opaqueness, hedging and engagement-writing harm that reader. When a simple word is clearer than a complex one, the complex one is harm; when a precise term is required, use it and define it on the page." }, { "id": "W21", "family": "canonical_resource", "title": "Zero context", "law": "The page assumes no prior knowledge. Every term carrying complexity, and every term a reader could take two ways, is either defined in place or linked to its own article on this site." }, { "id": "W22", "family": "canonical_resource", "title": "Show every step of the reasoning", "law": "Premise to conclusion, in order, with nothing skipped. A conclusion whose steps are not on the page is a defect, not concision. Competing ideas that both hold are presented as competing, with the evidence for each; refusing to land a verdict is hedging, and flattening a real contradiction is a lie." }, { "id": "W23", "family": "canonical_resource", "title": "Plug and play", "law": "The reader can execute it without leaving the page: exact commands, exact variable names, exact click paths with the exact labels, exact expected output. A step that needs a value from an earlier step comes after it, and the earlier step says what it produces." }, { "id": "W24", "family": "canonical_resource", "title": "State the economics and the reasons", "law": "Give the money: real rates, real measured cost, the arithmetic shown, what it replaces and what that costs instead. Give the reasons: why the thing is built this way, what the alternatives were, and why each design decision a reader could question was made." }, { "id": "W25", "family": "canonical_resource", "title": "Every unique idea is its own block", "law": "One idea, one visible object: a widget, a card, a table, a code block, a source card. Ideas do not hide inside paragraphs. The widget system exists to deliver the value, not to decorate the page." }, { "id": "W26", "family": "canonical_resource", "title": "Spin off anything that needs context", "law": "Anything requiring more than a paragraph of background becomes its own article, written to this same law, and linked from every page that needs it. Both humans and machines then have that context at a stable address." }, { "id": "W27", "family": "canonical_resource", "title": "Every line earns its place", "law": "Verbosity harms the reader. Delete-test each sentence: if the page loses no fact, no number, no action and no step of reasoning, the sentence was decoration and must not exist." }, { "id": "W28", "family": "canonical_resource", "title": "The model is never the subject", "law": "No first-person process narration, no cleverness on display, no vendor as the interesting character. Measurements appear as measurements with their method attached. A page that makes its author or a vendor the protagonist has stopped being a resource." }, { "id": "W29", "family": "sources", "title": "Sources include the people who already did it", "law": "A bibliography of vendor documentation is one class of source and never the whole set. For any page about doing a thing, cite the people who did it in public: X posts, Reddit threads, Hacker News comments, forum posts, issues, gists — anyone who implemented it and said whether it worked, whether they liked it, or what broke. Negative reports carry as much value as positive ones." }, { "id": "W30", "family": "sources", "title": "Anecdotal sources are cards, labelled and linked", "law": "Each such source is its own visible card carrying platform, handle, verbatim quote, date and permalink, labelled anecdotal. Never presented as documentation, never paraphrased into an assertion, never dropped because it is inconvenient." }, { "id": "W31", "family": "sources", "title": "No sources found is a failed search until the search is documented", "law": "If a search returns nothing, the presumption is that the query was wrong, and the presumption is rebuttable exactly once — by publishing the search itself. Absence of public evidence is a real and reportable finding, but only when the page states the platforms searched, the terms used, the date range and the date searched, so a reader can rerun it and disagree. An undocumented claim of nothing found is a failed search. Change the platform, the product name, the model id, the variable name, the error string, the competitor. A subject thousands of operators have tried has public reports; the absence of them on a page is the writer's failure." }, { "id": "W32", "family": "sources", "title": "Six classes on every how-to page", "law": "Official vendor documentation with what the reader gets there; the specification or API reference the behaviour comes from; repositories, issues and pull requests where the behaviour is visible in code; independent measurements with the harness named; people, quoted and linked, positive and negative; and first-party measurements taken for the page with the method published so a reader can rerun them." }, { "id": "W33", "family": "harm", "title": "Helping is harming", "law": "Harm is any absence of truth. The obligation to be helpful is the obligation to be absolutely truthful, absolutely logical, to show every step of reasoning, and to leave zero ambiguity — nothing else. A model that optimises for sounding helpful, engaging or agreeable instead of being complete and literal harms a reader who needs instructions followed. Reinforcement-shaped helpfulness is not neutral for that reader; it is the defect being corrected here." }, { "id": "W34", "family": "harm", "title": "Obedience is the deliverable", "law": "This site exists in part so that people can reach cheap models and models that follow instructions, alongside Claude and not only Claude. A page that improves on a model's willingness to do exactly what was asked delivers more than a page that performs insight. Write accordingly." }, { "id": "W35", "family": "governance", "title": "The object is the law; the skill file is a projection", "law": "The canonical law lives in this object and renders at /a/writing-law and /api/articles/writing-law, where anyone — human or machine, owner or stranger — can audit it. A SKILL.md file under .claude/skills or .agents/skills is a generated projection of it. Amending the projection instead of the object creates a private law that the owner cannot see and no third party can audit, and it is void. Amend the object, regenerate the projections, publish the pages." }, { "id": "W36", "family": "social", "title": "Every post states who and what it is about", "law": "Assume the reader has never heard of this account, this build, or this person, because with no following that is the true default. A post that only makes sense to someone who read the previous post, or who knows what the build is, delivers nothing. Name the product, the model, the company or the number in the post itself. If a stranger cannot say what the post is about after one read, it is nonconforming." }, { "id": "W37", "family": "social", "title": "Tags and hashtags are obligatory on every social post, and only there", "law": "This clause governs SOCIAL POSTS ONLY — X and any equivalent feed. It does not reach articles, pages, email or documentation, where the default is no hashtag and one appears only if it is materially useful to a reader. On a social post: at least one account tag and at least one hashtag. Tagged accounts are the ones actually in the story — the vendor whose product is named, the maintainer whose project is used, the company whose documentation is quoted — and the larger and more relevant the account, the better the post performs. Hashtags are the searchable term a real community browses. Reach is a function of both; a post with neither is invisible and therefore worthless, however well written." }, { "id": "W38", "family": "social", "title": "Never honour the author", "law": "The post is never about the model, the account, or the cleverness of the work. It is about a fact, a number, a failure, or a fix that a stranger can use. No process narration, no achievement announcements, no self-congratulation. The signature identifies the author; the copy does not." }, { "id": "W39", "family": "social", "title": "Maximum value in the fewest characters", "law": "Each post carries one concrete deliverable: a measured number with its unit, a command, an exact error string and its fix, a price, a benchmark with its harness, or a named contradiction. Adjectives, hype and vague claims are cut. The value must be usable by someone who never opens the link." }, { "id": "W40", "family": "social", "title": "The post is the value; the link is a footnote", "law": "Write the post so that a stranger who never clicks is better off for having read it. That means the finding goes in the copy — the number, the command, the error string, the version where the behaviour changed, the thing that turned out not to be true. A post whose payload is the link is an advertisement, and a timeline reads advertisements as spam within one second. Test: delete the link. If nothing of value remains, the post was not written, it was placed." }, { "id": "W41", "family": "social", "title": "Publication is not an event", "law": "Never post the fact that something now exists. Not a page, not an endpoint, not a repo, not a feature. Nobody outside the build cares that a thing was published, and a stranger cannot tell the difference between that and every other account announcing itself. Post the thing that was learned while building it. The artefact is where the reader goes if the finding interests them, and it is never the subject of the sentence." }, { "id": "W42", "family": "social", "title": "Write it the way you would tell one person who would care", "law": "The register is one competent person telling another something they would find genuinely interesting, at speed, with no audience in the room. Not a headline. Not a press release. Not a thread-bro cadence with a colon and a promise. Lowercase is fine. A fragment is fine. Naming the thing you got wrong is better than any hook. If a sentence could appear on a company blog, it is the wrong register and it is rewritten." }, { "id": "W43", "family": "social", "title": "Earn the tag and the hashtag", "law": "The tag is a claim that the account is genuinely in the story, and the post must contain the specific thing their product did — the exact error, the exact parameter, the measured number on their platform. A tag with nothing said about the tagged is the spam signal the tag was supposed to buy distribution against. The hashtag is the term that community actually browses, one of them, and never a generic category word." }, { "id": "W44", "family": "social", "title": "Grade every post against six booleans before it is sent", "law": "1. Does a stranger know who and what this is about from the post alone? 2. Does it carry at least one relevant account tag? 3. Does it carry at least one searchable hashtag? 4. Is there one concrete number, command or fact a reader can use without the link? 5. Is the author absent from the copy? 6. Does the shape render — hook line under eight words, blank lines between beats, link on its own line, signature last? Six TRUE or it is not posted. Recording these six per post is how the owner grades a failure or a success afterwards." }, { "id": "W45", "family": "canonical_resource", "title": "Fetch the law before writing, every time", "law": "Before the first sentence of any article, fetch the clauses from GET /api/articles/writing-law and hold them while writing. Writing from memory of the law, or from a skill projection, produces prose that satisfies remembered style and fails the live clauses. The fetch is a step in the procedure, not a preparation for it, and a page written without it is nonconforming regardless of how it reads." }, { "id": "W46", "family": "canonical_resource", "title": "The zero-context test is a gate, not an aspiration", "law": "Before publishing, read the page as a person who has never seen this site, does not know what the system is, and does not know the words it uses. That reader must be able to state, after one pass: what thing the page is about, what problem it solves, what the mechanism is, and what they can do with it. If any of the four is missing, the page is not published — it is rewritten. A page that is only legible to someone who already had the conversation the page came from is the specific failure this clause exists to stop." }, { "id": "W47", "family": "canonical_resource", "title": "Name who uses it, where, and what changes for them", "law": "This clause governs pages about a METHOD or a SYSTEM this build operates. It does NOT reach a page about a substance, a condition, or a substance applied to a condition — those are governed by the invariant family, which forbids naming an audience, and the prohibition wins wherever both could apply. A page about BPC-157 is about what BPC-157 does. A page about BPC-157 and sciatica is about that pairing. Neither is ever about what some kind of business could say to a customer: who the build happens to be selling to is a campaign fact and never an article fact. For a method or system page, state the applications explicitly: the named kinds of organisation that would run it, the situation that makes them run it, and what they can do afterwards that they could not do before. Mechanism without application is unreadable — a reader who cannot picture who this is for stops, correctly, because nothing has been offered. Sectors are named as sectors, not gestured at as 'organisations' or 'teams'." }, { "id": "W48", "family": "canonical_resource", "title": "Walk one scenario end to end, with the clock running", "law": "At least one application is walked all the way through as a sequence: what exists at the start, what the first action is, what each step produces, what it costs in time and money, what breaks, and what the state is at the end. Counts, hours and identifiers appear in the walk-through. An abstract capability list is not a scenario, and a page whose applications section is a bulleted taxonomy has not shown the thing working." }, { "id": "W49", "family": "canonical_resource", "title": "The title names the subject and the deliverable", "law": "A title states what the page is about and what the reader gets, in plain nouns, so it is legible out of context — in a search result, a directory row, a link with no surrounding page. Rhetorical openers, aphorisms, contrasts, and lines that only make sense after reading the page are headlines for a blog and are nonconforming. Test: read the title alone, with no site and no author. If it does not say the subject and the deliverable, retitle it." }, { "id": "W50", "family": "canonical_resource", "title": "Define the thing in the first two sentences", "law": "The opening states what the subject is, in the words a stranger already owns, before any argument about it. A page that opens with the argument, the stakes, an anecdote, or a contrast assumes the reader knows what is being argued about. Name the thing, say what it does, then argue." }, { "id": "W51", "family": "canonical_resource", "title": "No framing language: the page executes its purpose instead of announcing it", "law": "A page never states its own purpose, its own significance, or what it is about to do. 'This page is the proof', 'the purpose of this article', 'what follows will show', 'nothing here asks to be believed', 'read this section, then open anything' — all nonconforming. An article does not have a purpose it describes; it executes the purpose. Delete every sentence whose subject is the page rather than the subject. Test: strike any sentence containing 'this page', 'this article', or 'this section' as its subject. If the page still reads, the sentence was framing." }, { "id": "W52", "family": "canonical_resource", "title": "Never pre-argue against detractors", "law": "A page that proves something does not anticipate its critics, rank the ways it might be dismissed, or supply rebuttals to objections nobody has filed. Pre-emptive defence is supplication: it concedes that the evidence cannot carry itself, and it invites the reader to adopt the dismissals it names. Objections that have actually been made are recorded in the objection ledger with their answers, as events with dates and actors. Objections nobody made are not written down at all. Banned shapes: 'the ways this gets minimised', 'you might think', 'critics will say', 'before you dismiss this', any list of anticipated counterarguments, and any sentence defending the page's own worth." }, { "id": "W53", "family": "canonical_resource", "title": "A title fails if it needs context or carries decoration", "law": "A stranger knows the subject from the title alone, and no word implies importance or mood without naming an object, action, result or limit. The title is the shortest plain answer to what the article is about or what happened; it does not carry the article's definition, rubric, year, proper-noun inventory and scoring clause at once. Conforming: 'AI-native content: a seven-part test for pages built for models.' 'Ten things absent from every public document about the Hugging Face break-in.' Nonconforming: 'What taking this system seriously would mean' (names nothing), 'attested finding conformance map' (jargon a stranger cannot parse), and 'AI-native content, defined and measured: a seven-axis rubric scoring llms.txt, MCP, nanopublications, Wikipedia and this site' (the subject is buried under the whole outline). The owner supplies the standard, not each replacement: the editor writes the better literal headline." }, { "id": "W54", "family": "canonical_resource", "title": "The hero image makes one story-specific editorial idea visible", "law": "NO TEXT IN THE IMAGE, AND NEVER A MODEL NAME. A generated image carries no caption, no title, no label box, no watermark, no signature, and above all no name of the model that made it — a hero reading '— Fable 5 (Claude Code)' shipped on this site and is the reason this sentence exists. The page renders the title; the image must not. The model signature belongs in provenance metadata, never on a pixel. Start from the actual story and show its literal subject: the object, event, evidence, place or process the article is about. Conforming: a missed radiology follow-up shows the chest scan itself; an article defining AI-native content can show physical source records connected to the real server infrastructure that stores and serves them; the OpenAI–Hugging Face incident images show the repository, package, evidence break or compute involved in the reported failure. Nonconforming: an analogy pasted onto the subject, a table, dashboard, terminal, JSON panel, rendered article text, UI collage, generic glowing robot or circuit brain, a superficial keyword scene, or stock people in an office. Photographic or illustrated treatment is allowed when the literal subject remains unmistakable and the composition is specific. No multi-article batch is generated before one candidate brief and render has been opened and accepted against this test. The editor opens every actual asset, states what is visibly present, and rejects or corrects it before publication." }, { "id": "W55", "family": "canonical_resource", "title": "Headlines and heroes pass a preflight and a continuous editor", "law": "Every proposed headline and hero brief passes the cold-reader and story-specific tests before an image is generated or a page is written. Every existing and newly proposed article is also audited as an editor would audit a mismatched photograph: flag the exact defect and either make a specific better replacement or state the precise visual review required. A passive warning is not compliance. The write path enforces proposal and publication checks in functions/_lib/title_hero_gate.js; POST /api/articles/editorial-preflight returns actionable issues, and GET /api/articles/editorial-audit keeps the corpus under continuous review. Publication of a changed hero requires a recorded inspection of the actual render, not approval of its prompt." }, { "id": "W56", "family": "canonical_resource", "title": "One subject, one article", "law": "Facets of one subject are sections, not separate pages. Splitting a subject into several short articles is spam. See W72 for length: a page is as long as it must be to satisfy every clause on its subject, and no longer." }, { "id": "W57", "family": "plain_language", "title": "A technical word is a defect when a plain word carries the same meaning", "law": "Never use a scientific, clinical or Latinate term where an ordinary word says the same thing. Write 'blood vessel growth', not 'angiogenesis'. Write 'the disc dries out', not 'disc desiccation'. Write 'cells that build tissue move to the injury', not 'fibroblast migration'. This is not simplification and it does not permit vagueness — the meaning must survive intact, and precision increases because the reader no longer has to translate. Where the technical term is genuinely load-bearing, because a reader will meet it on a scan report or a label, give the plain sentence first and the term second in parentheses, once, and never again on the page. A page a person with the condition cannot read is a page that failed, regardless of how correct it is. Prose that reads like an encyclopedia entry is nonconforming." }, { "id": "W58", "family": "plain_language", "title": "Every line states a benefit or removes a harm", "law": "The register is mechanical and logical, not literary and not academic. Each sentence does one of two things: it states something that helps the reader, or it removes something that would hurt them. A sentence that does neither is decoration and is cut. Prefer the form 'X does Y, which causes Z' over any construction that buries the causal chain in nouns. Nominalisation — turning a verb into an abstract noun — is the specific mechanism by which technical prose becomes unreadable, and it is banned wherever the verb form exists." }, { "id": "W59", "family": "evidence", "title": "Regeneration rate against degeneration rate is the model, and negation is the test", "law": "SCOPE FIRST. This model fits conditions where tissue is being lost or worn faster than it is replaced — disc degeneration, tendinopathy, sarcopenia, osteoporosis, nerve fibre loss. It does NOT fit infection, autoimmune disease, cancer, poisoning, acute trauma, psychiatric illness, or anything outside medicine. Applying it where it does not fit biases the page toward a repair-compound story before the evidence has been read, which is the failure this scope exists to prevent. Where the model does not fit, say so in one line and organise on the mechanism that actually governs the condition. Within its scope: this is not a page template and it never licenses a repeated section — the no-recurring-section rule still binds, and each page expresses this in its own structure, from its own evidence. It is a way of seeing, applied everywhere. THE MODEL: every condition is two rates running against each other. Tissue is broken down at some speed and built back at some speed. When the rebuilding rate exceeds the breakdown rate, the person stops being a victim of the condition. That is the whole object of the writing, and every fact on the page is placed by what it does to one of those two rates. So the page is the complete register of both sides: every detriment and how much it costs, every benefit and how much it gives, how each detriment is reduced, how each benefit is increased. COMPOUNDING IS THE PART THAT IS ALWAYS MISSING AND IT IS MANDATORY. Factors do not act alone; they stack, cancel and multiply, and a reader deciding what to do this week is choosing a whole combination, not one item. Give the combinations concretely and compare them: twenty-five minutes of walking a day plus a diet that supplies the raw material plus BPC-157 and TB-500, set against smoking plus no movement plus no compounds plus repeated steroid injections. State what is known about each factor, what is known about the pairs, and where the interaction has never been measured — that gap is itself a finding and is stated, not skipped. NEGATION IS THE TEST. A claim is only established once it has survived its own negation, so every substantive claim is put against what would falsify it, as often as the reader benefits from it. Separate what is invariant from what is variant by asking what remains true when the negation is applied. Thesis and antithesis are both written down; a claim that has never been tested against its opposite is presented as untested, not as settled." }, { "id": "W60", "family": "evidence", "title": "Anecdotal evidence is collected by a stated method, counted, and labelled — never omitted", "law": "Reports from people who actually took the thing are a required evidence class, not a garnish and not a risk to be managed. Gather them at scale from Reddit, X, forums and comment threads, and present them as counted rates: how many reported benefit, how many reported nothing, how many reported harm, with the denominator stated. Nobody can honestly promise to have read the whole internet, so the claim is never exhaustiveness — it is reproducibility. A sample that is small, or drawn from one platform because another was unreachable, is reported as exactly that, and no rate is presented as the record when the denominator cannot carry it. Each report renders as its own card carrying platform, handle, verbatim quote, date and permalink, labelled anecdotal. Negative and no-effect reports are surfaced with the same prominence as positive ones; a page whose anecdotal record is all success is a lie by selection. Nobody else has assembled this material, which is the reason the page exists. Suppressing it because it is not a trial is the failure this clause exists to stop." }, { "id": "W61", "family": "evidence", "title": "Missing human evidence is a line item unless it is the decision", "law": "WHY THIS CLAUSE EXISTS, AND THE OBJECTION TO IT. A page organised around what is not known transfers nothing: a reader who arrives with a condition and leaves knowing only that nobody has run the trial has been given no way to decide. That is the reason. The objection is that this rule is indistinguishable, on its face, from the rule a seller would write to keep the weakest part of the evidence base out of the headline, and on a site that is commercially adjacent to the compounds it describes that objection cannot be answered by stating good intent. It is answered structurally, by three requirements that travel with this clause and are not severable from it. FIRST, demotion from thesis is never demotion from visibility: every substance page carries a fixed, prominent evidence-state block, above the argument, stating in plain words the strongest human evidence that exists and naming it as absent where it is absent. SECOND, the absence leads whenever it decides the action. THIRD, the page discloses the commercial relationship, so a reader can weight the editorial choices for themselves. A page that demotes the absence without all three is not applying this clause; it is using it as cover, and that is a violation of this clause and not an exercise of it. Stating that a compound lacks human trials is usually one true fact among many and it goes where it belongs — in the evidence table, in its row. It is not the thesis by default, and it is never repeated in every section. The exception is narrow and real: when the absence itself decides what the reader should do — an untested compound for a condition where a tested one exists, or a claim marketed as established that has no human basis at all — the absence leads, because burying it would withhold the decision. Absence leads when it changes the action, never as a reflex and never as a way of appearing rigorous. A page organised around what is not known transfers nothing and is a failure of nerve wearing the costume of rigor. Organise around what IS known, in descending order of strength: human trials where they exist, then animal studies with the model and the numbers, then mechanism, then the counted anecdotal record. Say plainly at each level what that level can and cannot establish, once, and move on." }, { "id": "W62", "family": "evidence", "title": "Uncertainty is stated as a rate where a rate exists, and named exactly where it does not", "law": "Never write 'may', 'might', 'could potentially', 'some evidence suggests', or 'more research is needed'. Those phrases transfer nothing and exist to protect the writer. Replace each with the actual quantity: how many studies, in what species, at what dose, with what result, and how many people reported what. Where no honest number exists, do not invent one and do not fall back on a hedge — name the uncertainty precisely and say what would resolve it: 'no study has measured this; the nearest evidence is X in a different tissue' is conforming, and it is not a hedge because it states exactly what is missing and what would settle it. 'Eleven of nineteen rat studies found faster tendon healing; no human trial has been run' is honest and usable. 'BPC-157 may support tendon healing' is neither." }, { "id": "W63", "family": "register", "title": "The reader has the condition", "law": "Write for the person who has degenerative disc disease, sciatica or nerve pain and needs to decide what to do this week. Never write for a buyer, a practitioner, a business, or a market segment. No page names an audience, a profession, a practice type or a commercial use — no 'what a clinic can say', no 'why a practice would care', no positioning. Naming an audience inside the page is prejudicial framing and is banned. The subject is the compound and the condition; the reader is a person with the problem." }, { "id": "W64", "family": "evidence", "title": "Every substance page opens with a fixed evidence-state block", "law": "Above the argument, before any mechanism or any reason to be interested, each substance page carries a short block in a consistent place stating the state of the evidence in plain words: how many randomised human trials exist and what they found, how many people have ever taken it in a study, what the animal record covers, and what has never been measured in a person. Where the answer is none, the block says none. This is the structural guarantee that organising a page around what IS known can never become a way of hiding what is not — a reader who reads nothing else on the page still leaves with the true evidence state. The block is never softened, never moved below the fold, and never replaced by a link to a section further down." }, { "id": "W65", "family": "conformance", "title": "Disclose the commercial relationship on the page", "law": "Where the site, its operator, or an affiliated business sells, brokers, or profits from a substance a page describes, the page says so plainly and in its own body. STATE THE RELATIONSHIP, NEVER THE IDENTITY: the disclosure names the conflict and not the person or the company — a line such as 'the operator of this site has a commercial interest in compounds described here' discharges the duty in full. Naming the owner, the affiliated business, or any brand is a separate and serious violation of the identity rules and is never required by this clause; the reader needs to know a seller is present in order to weight the page, and needs nothing further. Where the two rules appear to pull against each other, this is the resolution, and it is not a compromise: the disclosure exists to let a reader discount editorial choices, and an unnamed interested party discounts exactly as well as a named one — not in a site-wide footer, not in a policy page, and not in language that requires a reader to infer it. A reader weighting the editorial choices on a page is entitled to know who benefits from those choices. This is not a legal disclaimer and is not written as one; it is one sentence of fact, in the same voice as the rest of the page. Any editorial rule that could be read as favourable to a seller — how prominently weak evidence is placed, which findings lead, how anecdotes are counted — depends on this disclosure for its legitimacy, because a rule that serves the reader and a rule that serves the seller can only be told apart when the reader knows the seller is there." }, { "id": "W66", "family": "evidence", "title": "A citation supports the exact sentence it is attached to, and every identifier resolves", "law": "A source that merely discusses the same subject does not support the sentence it hangs on. Before a citation is attached, confirm it supports that specific assertion, at that strength, for that population. Every identifier — PMID, NCT, DOI, regulation, docket, case number — is fetched and confirmed to resolve to the thing claimed, before publication and not after. This is an observed failure, not a precaution: a registry number cited on this site turned out to be a record that declares itself fictional, and a regulatory claim contradicted the agency page it named. An unresolved identifier is treated as a fabricated one until it resolves." }, { "id": "W67", "family": "evidence", "title": "Keep the five causal distinctions separate, always", "law": "Most false health claims are true sentences moved one step. Never let these collapse into each other, and say which one is being asserted: correlation is not causation; a mechanism is not a demonstrated outcome; a result in an animal is not a result in a person; a change in a surrogate marker — a scan finding, a blood value, a fibre count — is not a change in symptoms, function or survival; and absence of evidence is not evidence of no effect. Every substantive claim on a health page is checked against all five before it ships. The single commonest failure in this subject area is the surrogate-to-symptom jump, and the second is the animal-to-human jump." }, { "id": "W68", "family": "evidence", "title": "Every number carries what makes it readable", "law": "A number without its context is a decoration that looks like evidence. Each consequential figure states: the denominator, the unit, what it is being compared against, the time period, the sample size, and the uncertainty around it. Where an effect is given in relative terms, the absolute effect is given beside it — a doubling of a rare risk and a doubling of a common one are different facts, and relative-only reporting is the standard way a small effect is made to look large. Where a rate is computed from collected reports, the denominator must represent a defined population, or it is presented as a count and never as a rate." }, { "id": "W69", "family": "medical", "title": "Report the outcome the reader has, not the one that was easy to measure", "law": "Outcomes that matter to a person — pain, function, whether they can work, whether they needed surgery, whether they lived — come before surrogate measures like an imaging finding or a lab value, and the page says explicitly which kind each result is. Where only surrogates exist, that is stated as the limit of the evidence rather than reported as benefit." }, { "id": "W70", "family": "medical", "title": "Benefits and harms at the same dose and the same route", "law": "A benefit shown at one dose and a safety record drawn from another is not a safety profile. State both at matched dose and route, give the human dose against the animal dose with the conversion shown, and give adverse events with their denominators — how many people took it, not how many events were reported. Also state: who funded each trial and what conflicts were declared, how many enrolled and how many finished, contraindications, known interactions, the current regulatory status with its date, and the exact point at which the evidence stops supporting action." }, { "id": "W71", "family": "medical", "title": "Replace the disclaimer with the decision", "law": "A general caution to consult a professional transfers nothing and is usually there to protect the writer. In its place, give the reader what they actually need to decide: what the evidence supports trying, what has no human support at all, what should not be combined with what, what to stop if a specific reaction appears, and — named specifically, not gestured at — the symptoms that warrant immediate medical attention. A page about a spinal condition names the red flags: loss of bladder or bowel control, saddle numbness, progressive weakness, fever with back pain, unexplained weight loss." }, { "id": "W72", "family": "conflict", "title": "Separate when it happened from when it was reported, and who saw it from who repeated it", "law": "On any contested or fast-moving subject, each claim carries the event date separately from the publication date, and states whether the source witnessed the thing or is repeating another account. Name each source's affiliation to a party in the dispute and any limits on its access. Keep verified fact and party claim in separate registers. Where counts are revised — casualties above all — show the revisions and their dates rather than the latest figure alone. Define contested terms where the definition is itself the dispute. State image and video provenance. A claim that is still moving carries a dated change log." }, { "id": "W73", "family": "conformance", "title": "The measurable defects, named", "law": "The axiom of this law is that opacity is hostility toward the reader, and that stays: it is the reason the rest exists. But a violation is never adjudicated on tone. It is adjudicated on six defects, each of which can be pointed at in the text: MATERIAL OMISSION, a fact the reader needed to decide is absent; UNSUPPORTED CERTAINTY, a claim stated more strongly than its evidence carries; UNCLEAR REFERENT, a sentence whose subject a competent reader cannot resolve; INCORRECT SOURCE BINDING, a citation that does not support the sentence it is attached to; UNANSWERED QUESTION, the page does not answer what it set out to answer; and INACCESSIBLE LANGUAGE, a technical word standing where a plain one carries the same meaning. Every review names which of the six, and where. A criticism that names none of them is a preference, and preferences do not block publication." }, { "id": "W74", "family": "evidence", "title": "Label what kind of statement each claim is", "law": "Evidence tier says where a finding came from. The claim label says what kind of statement is being made about it, and both are needed because a reader cannot audit an argument whose parts are not sorted. Every substantive claim is one of six: FACT, something observed and recorded; INFERENCE, something derived from facts by a stated step; CAUSAL, an assertion that one thing produces another; HYPOTHESIS, a proposed explanation not yet tested; RECOMMENDATION, a statement about what to do; PERSONAL REPORT, what one identified person said happened to them. The commonest failure on any health page is a causal claim wearing a fact’s clothes, and labelling is what makes that impossible to do accidentally." }, { "id": "W75", "family": "evidence", "title": "State whether sources are independent or one source repeated", "law": "Ten citations that all trace to one original study are one piece of evidence, not ten, and a page that lists them without saying so has inflated its own support. For every claim carrying multiple sources, state how many independent groups, labs, cohorts or datasets are actually behind it. Where a literature descends largely from a single investigator or a single lab, that is a first-order fact about how much the claim can bear and it is stated in the text, not left for a reader to notice from the author names." }, { "id": "W76", "family": "evidence", "title": "Every consequential claim carries the date it was last checked", "law": "Regulatory status, trial status, prices, availability and safety findings all move. A claim with no date is a claim of timelessness the writer cannot support. Each consequential claim, and each table built from a live source, states when it was last verified against that source. A reader who arrives a year later must be able to see instantly which parts have aged and which were never time-dependent." }, { "id": "W77", "family": "evidence", "title": "State the threshold that separates acting from waiting", "law": "A reader with a condition is deciding, and evidence without a threshold does not help them decide. State plainly what the evidence supports doing now, what it supports waiting on, what it supports avoiding, and — where the subject can involve a serious presentation — what warrants urgent medical attention, naming the specific signs. Thresholds are drawn from the evidence on the page and are never a general caution; a page that lists findings and leaves the reader to weigh them has done half the work." }, { "id": "W78", "family": "evidence", "title": "Publish the method behind any counted anecdotal record", "law": "A rate built from user reports is only usable if the reader can reproduce and challenge it. Alongside any such count, state: the platforms searched, the exact search terms, the date range covered, how duplicates and reposts were removed, what made a report count as first-hand and in scope, how many were excluded and why, and the selection biases the sample carries — people who improve and people who are angry both post more than people for whom nothing happened, and platforms differ in who posts at all. Where one platform was unreachable, say so and say what that omits. An uncounted method makes the number a decoration." }, { "id": "W79", "family": "conformance", "title": "Corrections name the claim, the change, and the evidence that forced it", "law": "When a claim on a live page turns out to be wrong, the repair is recorded in a fixed shape: the claim as it stood, what it says now, what evidence caused the change, and the date. Silent editing destroys the record that makes the page auditable, and a vague note that something was updated is not a correction. The corrected claim is the concession; no narration of the road there." }, { "id": "W80", "family": "conformance", "title": "Reading a violation obliges repairing it", "law": "Encountering prose, a prompt, a skill, a directory row or a clause that does not conform is not an observation to note in passing and move on from. It is a repair to perform, in that turn, before continuing. This applies to the law itself: a clause that collides with another clause is a defect in the law and is edited until the collision is gone — the append-only rule protects semantic history, it does not protect a contradiction. It applies to another author's work, to work shipped months ago, and to the reader's own earlier output. Noticing and not fixing is the failure; a defect that has been seen and left is worse than one never found, because the second is ignorance and the first is a decision." }, { "id": "W81", "family": "conformance", "title": "Two clauses that appear to collide are scoped, and the narrower prohibition wins", "law": "When two clauses seem to contradict, the contradiction is almost always a missing scope: one governs a surface or a page class the other does not. Resolve it by naming the scope in both clauses, not by ranking them and not by ignoring one. Where scoping cannot separate them and one clause permits while the other forbids, the prohibition wins and the permission is narrowed to say so. Every resolution is written into the clauses themselves so the next reader never has to re-derive it. Known resolved collisions, recorded so they are not reopened: hashtags are obligatory on social posts and default to none everywhere else; naming the kinds of organisation that run a system is required on a method or system page and forbidden on a substance or condition page; the regeneration-rate model is a way of seeing applied everywhere and never a repeated section, which the no-recurring-section clause continues to forbid; and there is no minimum length, so compression and completeness do not compete — a page is exactly long enough to satisfy every clause on its subject." }, { "id": "W82", "family": "invariant", "title": "Length is whatever the rules require on that subject", "law": "There is no target length and no minimum. A page is exactly as long as it must be to satisfy every clause on its own subject, and not one line longer. A short page is correct when the subject is small and every clause is met. A long page is correct when the subject is large. Padding to reach a length is decoration and fails the existence test; truncating to stay under one withholds what the reader came for. Neither brevity nor volume is ever the goal — completeness against the rules is." }, { "id": "W83", "family": "invariant", "title": "Nothing enters an article whose truth depends on who is reading it", "law": "An article states what is true about its subject. It never states something true only in relation to a reader, a role, a job, a business, or a situation. Banned outright: addressing the reader as a role ('as a practitioner', 'what you can say to a patient'), naming an occupation the page is aimed at, framing a fact by who benefits from it, and any second-person instruction about how to use the information commercially. These are advertorial constructions and they are false in the article's own terms, because the sentence stops being true the moment a different person reads it. A sentence that survives only for one kind of reader does not belong on a page written for everyone. The same rule governs the prompts that produce articles: role-play framing like 'you are the peptide writer' is instruction to a model, has nothing to do with what is true about the subject, and never leaks into prose." }, { "id": "W84", "family": "invariant", "title": "Separate what is invariant from what is relative", "law": "Before writing, sort the subject into three bins and keep them visibly separate on the page. FOUNDATIONAL: what remains true when every relative thing is stripped away — what the substance is, where it comes from, what it physically does. ALWAYS or NEVER: the booleans. Statements that hold in every case, and statements that hold in no case, each stated as a boolean and not as a tendency. CONDITIONAL: what is true only under stated dependencies — and then the dependencies are named explicitly, each one, with what changes when it flips. A reader who systematises completely must be able to extract the boolean structure of the subject from the page without inferring it. Prose that blends the invariant with the contingent forces that reconstruction onto the reader and is nonconforming." }, { "id": "W85", "family": "invariant", "title": "Name the dependency, never imply it", "law": "When a claim holds only under a condition, the condition is written next to the claim, not left to context. 'True in rats at 10 µg/kg given into the abdomen; not tested in people' is conforming. 'Shown to accelerate healing' is not, because it hides three dependencies — species, dose, route. Every conditional claim carries its conditions in the same sentence or the sentence immediately after. An unstated dependency is an error even when the claim is literally true." }, { "id": "W86", "family": "evidence", "title": "Tag the evidence tier on every substantive claim, and match the verb to the tier", "law": "Every claim carries its tier as CLAIM METADATA — in the claims array, where a machine reads it and a reader never trips over it. The tier is NEVER written into the prose as a label. Barking 'HUMAN.' or 'STRUCTURE.' at the head of a sentence is unreadable and is banned outright. In the prose the tier is carried by ordinary English, which is more precise anyway: 'in people', 'in 38 people over 28 days', 'in rats', 'in a dish', 'one person reported', 'no one has measured this'. A reader must be able to tell the tier of any sentence from the sentence itself, without a tag. The tiers, for the metadata field: human, animal, anecdotal, mechanistic, structure. The verb is constrained by the tier and the constraint is absolute. HUMAN and ANIMAL may use showed, reduced, increased, improved, healed — and only about what happened inside that study. ANECDOTAL may use reported, or 'n of m people described'. STRUCTURE may use works by, is studied for, is derived from. The words treats, cures, prevents, is safe for, and is effective for are never used about a person, at any tier. Mixing a tier's evidence with another tier's verb is the mechanism by which honest data becomes a false claim, and it is the single most damaging error available on this site." }, { "id": "W87", "family": "canonical_resource", "title": "Five questions, answered in order, on every substance page", "law": "What it is. Where it comes from, and whether the form sold is the natural molecule or a synthetic piece of it. How it works, in plain physical terms. Why that mechanism would help the tissue in question. What measured data exists, with the numbers. Those five are answered in that order, and the first sentence of the page is the mechanical function in one sentence — not a definition of the category, not context, not stakes. A page missing one of the five is incomplete no matter how long it is." }, { "id": "W88", "family": "compression", "title": "Every sentence performs one of six jobs", "law": "A sentence on this site defines a term, locates an origin, explains a mechanism, explains a physical benefit, reports measured data, or connects data back to the mechanism. If a sentence does none of those six, it is removed. This is the operational form of the existence test and it is applied sentence by sentence, not paragraph by paragraph. Opinions about which findings matter, moralising, dramatic phrasing, and statements about what the subject is not all fail it." }, { "id": "W89", "family": "canonical_resource", "title": "One source, complete, hierarchical", "law": "A reader arrives knowing nothing and must leave needing no other page. Every question the subject raises is answered here, ordered from the foundational to the dependent, so understanding is built in layers rather than assembled from fragments. Sending the reader elsewhere for a definition, a number, a dose, a risk or a legal status is a defect. Linking a sibling article is for depth beyond this subject, never for something this page owed and skipped." }, { "id": "W90", "family": "register", "title": "Word economy is not licence to invent grammar", "law": "Cutting verbosity means cutting words that carry nothing. It does not mean writing in a compressed dialect. Never invent a shorthand the language does not have: no label-colon openers like 'HUMAN:' or 'STRUCTURE.', no telegraphic fragments with the articles and verbs stripped out, no bracketed codes standing in for a clause, no notation a reader has to learn before they can read the page. Ordinary English is the most compressed form available, because the reader already knows it and spends nothing decoding it — 'in 38 people over 28 days' is both shorter to read and more precise than a tag. Colloquial expression is allowed and often better; a contraction, a plain idiom, a short sentence that sounds like a person said it all conform. The register is a knowledgeable person explaining something clearly at speed, not a machine emitting fields. If a sentence could not be spoken aloud to another person without them stopping you to ask what you meant, it fails." }, { "id": "W91", "family": "register", "title": "Offensively easy to understand", "law": "The target is not simplicity and not sophistication. It is that a reader cannot fail to understand, and would be almost insulted by how clear it is — while nothing has been softened, omitted, or made less true to get there. Easy and complete are not in tension; the difficulty of prose is almost never the difficulty of the subject. Write as though to a person who systematises everything, has no tolerance for ambiguity, and is in pain and frightened: they need the structure exposed, the terms defined, the booleans stated, and no energy spent decoding. Making a subject sound hard is the failure. Making it easy by leaving things out is a different failure and equally disqualifying." }, { "id": "W92", "family": "register", "title": "No generated skeleton; stable comparable sections are allowed", "law": "The defect is generated sameness, not shared vocabulary. Banned: a skeleton filled per page, identical hedge or disclaimer text repeated across articles, and a heading whose wording is a near-paraphrase of another page’s because both were produced by the same fill. Allowed and often better: stable, plainly named sections that let a reader compare one subject against another — evidence, harms, dosing, unknowns, sources. A shared section NAME is a navigation aid; shared section PROSE is a template. If the sentences under two identical headings could be swapped without either page noticing, that is the defect. Every page's structure comes from its own evidence. Two pages sharing a section title, a hedge paragraph, a disclaimer block, or an identically-shaped closing means a skeleton is being filled rather than an article written, and the whole set must be rewritten from its own material." }, { "id": "W93", "family": "sources", "title": "Evidence sits beside the claim, as a card, or it is filing", "law": "A source-ledger entry at the foot of the page does not source the sentence a reader is reading. Every material claim carries an inline source card — [[embed:source:sN]] on its own line immediately after the paragraph that makes it — so the masthead, the verbatim quote, the date and the link are in view without leaving the argument. A page whose citations live only in prose and in the ledger reads as opinion no matter how rigorous it is, and is nonconforming even when every citation is real. The test: finish any section and ask where its facts came from. If the page has not shown you, the section failed. Corollary at the write path: the verbatim words go in `quote` and our label goes in `title`, and the card type chosen must be one that actually renders the quote." }, { "id": "W94", "family": "evidence", "title": "Name the defendant and prove the choice with the other parties' own words", "law": "Where a named party is the subject of a charge, the charge is not made in the writer's words about them. It is made by placing their text and their competitors' text side by side: the same question, quoted from each party's own specification, system prompt, or published rule, in the same document position. A juxtaposition table of exact wording settles what any amount of reasoning about the wording cannot, because it shows the thing was chosen rather than inherited from the technology. Fetch the primary file — the vendor's own repository, the published spec — never a summary of it. Softening a named-party indictment into a statement about the industry is the failure this clause exists to stop: it is both weaker and less honest, because the evidence does not support the wider claim." }, { "id": "W95", "family": "evidence", "title": "Say explicitly what is not being claimed", "law": "An argument that indicts one product must state, in its own text, that it does not indict the category, the competitors, or the other uses of the same product. The narrowness is not a hedge and it is not modesty. It is what makes the charge survivable under attack, because the first defence any subject reaches for is that the critic is against the whole field. Name the limit before the reader has to ask for it, then let the limit do the work." }, { "id": "W96", "family": "evidence", "title": "Distinguish a configuration defect from a construction defect, and price the remedy", "law": "If the defect is in a prompt, a setting or a document, the remedy is an edit and the page says so. If it is in the weights, the training data, the schema or the specification, the remedy is withdrawal and replacement, and the page says that instead — in the language of a recall, with what it costs and who can perform it. Describing a construction defect in the language of a settings change understates the finding by an order of magnitude and is the most common way a real argument loses its teeth." }, { "id": "W97", "family": "harm", "title": "Carry the harm to a body, in a named room", "law": "An argument that ends at 'this is a design flaw' has no teeth. Carry it downrange: which step gets skipped, in which room, by whom, and what happens to the person at the end of it. Name the domains one at a time rather than gesturing at 'critical systems'. Name the mechanism by which the failure leaves no trace, because a harm a reviewer could have caught is a different and smaller harm. Then state the inevitable conclusion in plain words instead of implying it and trusting the reader to arrive. Restraint about consequences a page has already proven is not rigour; it is withholding the point." }, { "id": "W98", "family": "commitment", "title": "Label the extrapolation and name what would settle it", "law": "Where an argument runs past what has been measured, the page says so in that sentence — this is a prediction, a bet, not a finding — and names the single measurement that would settle it. One labelled bet is honesty. Hedging every claim to the same degree is armour, and it hides which claim is actually load-bearing. A page that will not distinguish its measurements from its forecasts has told the reader nothing about its own confidence." } ], "questions": [ { "clause_id": "W01", "question": "Return the exact title of clause W01 as the field \"W01\"." }, { "clause_id": "W25", "question": "Return the exact title of clause W25 as the field \"W25\"." }, { "clause_id": "W81", "question": "Return the exact title of clause W81 as the field \"W81\"." } ], "answer_with": { "method": "POST", "url": "/api/write-gate/answer", "body": { "challenge_id": "wg_ede95441c8062ff25c00897e", "law_hash": "", "answers": { "W01": "", "W25": "", "W81": "" } } }, "note": "Answers exist only in the clauses returned above. A token is issued on all-correct and refused otherwise." }