{"path":"research/twelve-books-in-plain-words.md","content":"# Twelve books in plain words, by the minute in the video\n\n**Date**: 2026-10-12 · **Type**: a plain-language companion to\n[twelve-books-read-against-deliberus.md](twelve-books-read-against-deliberus.md) · **Status**: Claude's\nreading. Nothing here is ruled, and no ledger row is filed from it.\n\n**His words**, 2026-10-12, on the five findings reported from that reading: *\"Explain all of these\nbetter. Any and all learnings from the books mentioned, and relate it to direct quotes/phrasings from\nthe youtube video transcript please\"*. In the same message: *\"Pull from git btw and read up on the PR\nmerge I did about Wittgenstein etc\"*, which is Part 3.\n\n**How to read this note.**\n\n- **Every part gives the minute and second in the video** (Stephen Petro, *These 12 Books Will\n  Permanently Change What Your Mind Is Capable Of*, 14 minutes, uploaded 2026-10-05). What the creator\n  says there is told in this note's own words. His text is his own. The exact transcript is kept\n  privately with the same time codes, and playing the video at a time code gives his words.\n- **Cases come before points.** Each lesson has an everyday case, then what it means for Deliberus, then\n  what exists today and what it would open. The cases are made up for this note unless a line says the\n  video or a book gave them.\n- **None of the twelve books was read.** What is said about a book comes from reviews and critics,\n  listed in the long note. What is said about Deliberus is taken from the long note, which was checked\n  against the files.\n\n**Six words used below.**\n\n| Word | What it is |\n|---|---|\n| Claim | one statement in the graph, which a person can agree with, object to, or take apart |\n| Link | an arrow from one claim to another: a reason for it, or an objection to it |\n| Part | another kind of arrow: this claim is a piece of that one |\n| Badge | the mark on a claim that shows its strength, computed from what was said for and against it |\n| Hinge | a number for how far a conclusion moves when one claim under it is granted and then denied |\n| Concept layer | the part of the graph that keeps track of disputed words and their senses |\n\nOne more thing is designed and not built: the **analogy daemon**, a background reader that would look\nfor likenesses between debates in different fields. Its test run was done by hand on our own notes,\nwhere readers were sent to find such likenesses.\n\n---\n\n## Part 1. The five findings, again\n\n### 1. The badge answers one question out of four\n\n**In the video, 00:32 to 01:36.** A study of fifty thousand people finds that a treatment helps, and\nlater studies find the same. Then it turns out that the average benefit is too small to notice. The\ncreator's advice: when you are told a result is statistically significant, ask how large the effect is,\nhow sure the estimate is, and whether that much matters in the case at hand.\n\n**Three cases.**\n\n- *Sure and small.* A pill lowers weight, and ten studies agree. The loss is 200 grams in a year.\n- *Safe and empty.* Someone claims that all swans are white. A black swan turns up, and the claim\n  becomes *all swans are white, except in Australia*. After five such repairs nothing can refute it,\n  and it says very little.\n- *Dull and load-bearing.* *The steel in this bridge was tested* is a dull sentence. If it falls, *the\n  bridge is safe* falls with it.\n\n**The point.** There are four different things to ask about a claim.\n\n| The question | Whose question | In Deliberus today |\n|---|---|---|\n| Has it held up against what was said about it? | the strength layer | the badge |\n| How large is the thing claimed? | Ziliak and McCloskey (book 1) | stored on some claims; no strength reads it and no question asks it |\n| How much does it still say, after its exceptions? | Lakatos (book 3) | not recorded |\n| How much else rests on it? | Woodward's test, turned on our own map (book 2) | the hinge: computed, used to choose the next question, shown on no page beside a badge |\n\nThe video's first book warns against reading the first row as the second. Our own page invites the same\nmistake: a strong badge looks like *this matters*.\n\n**Today, and what it would open.** Nothing is built for this. Three small candidates wait: one sentence\nwherever a badge is explained, saying what it leaves out; a question about size among the standard\nquestions for an argument; and a count of the conditions a claim has taken on since it was first made.\nWith them a reader could tell a claim that is strong and large from one that is strong because it\nshrank.\n\n### 2. A reason can turn into a part of the claim\n\n**In the video, 02:25 to 03:37.** A class discusses a claim about solids with flat faces: corners minus\nedges plus faces is always two. A proof seems to settle it. Then students bring odd solids for which it\nfails, and each one makes the class look at the proof again, change a definition, narrow the claim, or\nfind a better theorem. The creator's advice: when someone finds an exception to your idea, be curious,\nand ask whether it shows a faulty step, a hidden assumption, a conclusion that was too wide, or a poor\ndefinition.\n\n**Two cases.**\n\n- *Birds.* The claim: all birds can fly. The argument for it: birds have wings, and wings are for\n  flying. Someone says *penguins*. The exception breaks one step of the argument, *wings are for\n  flying*. The best repair writes that step into the claim: *all birds whose wings are built for flight\n  can fly*.\n- *The solid with a hole.* The book's best-known exception is a solid shaped like a picture frame\n  *(recalled, not from the excerpts read)*. The proof had quietly assumed a solid with no hole through\n  it, and the repaired theorem says so in its opening words.\n\n**The point.** Look at the sentence *its wings are built for flight*. Before the penguin it was a\n**reason** for the claim, a step in the argument. After the repair it is a **part** of the claim, a\ncondition inside it. One sentence has had both jobs, one after the other. Lakatos calls this repair the\nbest of six things one can do with an exception (the six are under book 3 below).\n\n**Why that matters here.** For six months the step that links claims by machine stored *this is a\npremise of that* as a part. When the part links were read in September, the commonest fault had a name\nof its own: a reason filed as a part. Lakatos gives the cause, sixty years earlier. In a live argument\nthe two really do change places. He also gives the test that tells them apart:\n\n> If this sentence turns out false, is the claim false, or is only this argument for it gone?\n\nA false claim means the sentence was a part. A lost argument means it was a reason.\n\n**A second thing the graph lacks.** It records an objection. It does not record what the claim did\nnext. Suppose *a higher minimum wage costs jobs* is met by a study, and its author answers *for large\nrises, that is*. The claim was narrowed in answer to an attack, and nothing says so.\n\n**Today, and what it would open.** The test is used by hand. A record of how a claim answered an attack\n(dropped, word narrowed, exception listed, case described anew, condition added, or none of these) is a\ncandidate and is not built. It is what would let anybody tell an honest refinement from a retreat to\nsafer ground, and it is what the project's own central bet would need in order to show its narrowings.\n\n### 3. Why every list of answers needs a way out\n\n**In the video, 08:12 to 09:25.** The most technical book on the list. A system stays steady under\ndisturbance only if whatever regulates it has enough different responses. That is the law of requisite\nvariety. The creator's case is a company with one rule for every unhappy customer, whether the trouble\nis fraud, confusion, a broken product or an emergency. His advice: when something keeps failing, map\nthe state you want to keep, what disturbs it, what tells you it was disturbed, and the responses you\nhave.\n\n**Two more cases.**\n\n- *A form with two boxes*, Mr and Mrs. A doctor, a firm, and a person who is neither must each tick a\n  wrong box. The form is tidy and its data is wrong.\n- *A heater with one setting.* It keeps a room warm in October and cannot in January. No care in using\n  the one setting helps.\n\n**The point.** Ashby's law says that the kinds of outcome you end up with cannot be fewer than the\nkinds of trouble divided by the kinds of response. There are two ways to do better: have more\nresponses, or let fewer kinds of trouble reach you.\n\n**What it explains in Deliberus.** Three things the project already holds, which had no common reason.\n\n- **The *does not fit* answer.** A machine asked to sort links into seven kinds will meet an eighth.\n  With no way out it must file the eighth under the nearest of the seven, and nobody sees the error.\n  *Does not fit* is no eighth kind. It hands the case to something with more variety: a person, or the\n  place where new kinds are worked out.\n- **Why contributors must differ.** What a machine cannot supply is what its sources silently lack, and\n  a view of its own frame from outside. Both are variety. So ten people from one camp add less than\n  three people from three camps.\n- **How checking can keep up with persuasion.** Machines can write a million messages. Nobody can add a\n  million readers. The other way is open: let fewer distinct things arrive, by seeing that a thousand\n  messages make the same forty claims. Deciding that two messages say one thing loses nothing when it\n  is right. Summarising them loses the rare claim, and the rare claim is often the one that mattered.\n\n**A correction that came with it.** An earlier note said Ashby was named in our notes once. He was not.\nThat name belonged to another researcher.\n\n**Today, and what it would open.** The *does not fit* answer is built on the lists that sort links and\nkinds. The other two points are reasons for work already planned. One new habit is a candidate: before\na rule is added to the files that steer a session, ask whether the occasion for the mistake can be\nremoved (book 7 below).\n\n### 4. One word, several tests that usually agree\n\n**In the video, 04:50 to 06:35.** Most of us take a category for a box: members have the defining\nproperties and everything else stays out. Lakoff says many categories are built around typical cases\nand stretch outward from them. The creator gives *mother*: a birth mother, an adoptive mother, a foster\nmother and a surrogate mother each match a different piece of one ideal picture. His advice: when an\nargument hangs on a label such as *successful*, *intelligent*, *natural* or *normal*, ask what the\ntypical case is, which cases are central, and what happens at the edges.\n\n**Two cases.**\n\n- *Mother.* Four tests: gave birth, gave the genes, raises the child, is married to the father *(the\n  second and fourth are recalled)*. For most mothers all four say yes. A surrogate passes the first\n  only. An adoptive mother passes the third only. *She is not the real mother* is a statement about\n  which test rules here.\n- *Fair.* Four tests again: equal shares, equal chances, what each earned, what each needs. Six friends\n  split a restaurant bill. One says *equally*. One says *each pays for what they ate*. Both know all\n  four tests. They differ on which one governs a dinner.\n\n**The point.** The concept layer knows two shapes for a disputed word: one sense with a blurred edge,\nor two senses. *Mother* and *fair* have a third shape: one word, several tests that agree in ordinary\ncases and come apart in unusual ones. A dispute then has an exact place. Both sides agree on which\ntests the case passes, and they differ on which test decides.\n\n**Why that matters here.** When the concept layer was measured in September, 49 of 90 words read as\nsplit in two, largely because the step that finds disputed words writes two senses for each. A word of\nthe third shape is forced into that mould.\n\n**The probe it suggests.** Asking for a definition gets a tidy answer that hides how a person weighs\nthe tests. Asking about an edge case shows it. *Is a surrogate a mother? Is it fair that the one who\nhad only soup pays a sixth?*\n\n**Today, and what it would open.** The layer stores senses. A record of the tests in play under a word,\nand of which hold in a given case, is a candidate and is not built. So is making edge cases to ask\nabout. With them a quarrel over a word could be shown as what it is: agreement on the facts and a\ndifference over which test counts.\n\n### 5. A friend's design could show the analogy reader where to look\n\n**In the video, 03:37 to 04:50.** Analogy is at the centre of thought: we understand a new situation by\nseeing a familiar structure under an unfamiliar surface. The creator's case: a company that answers\nevery mistake with one more approval step is, underneath, like a doctor who treats every side effect\nwith one more drug. Each fix adds complexity, and the complexity makes new trouble. His advice: ask\nwhere you have seen this relationship before, even when the people, things and setting are different,\nand then name what matches and what is importantly different.\n\n**The design.** A friend's idea (Part 4 says where it came from). For each claim about what is the\ncase, a machine writes one line with fixed slots: subject, predicate, object or value, qualifiers, and\nthe time the content is about.\n\n| Subject | Predicate | Object | Qualifier |\n|---|---|---|---|\n| a rise in the minimum wage | lowers | employment among teenagers | in small towns |\n\nThe design's purpose is to find one thing said in two places. Two lines with the same things in the\nsame slots are the same statement, and looking that up needs no judgement.\n\n**Turned around.** Take the video's own pair and fill the slots.\n\n| Subject | Predicate | Object |\n|---|---|---|\n| adding an approval step after each mistake | increases | the complexity that causes mistakes |\n| adding a drug for each side effect | increases | the load that causes side effects |\n\nThe things differ. The middle slot is the same, and so is the shape: *a fix that feeds what it fixes*.\nA search by nearness of wording cannot find this pair, since *approval* and *prescription* share no\nwords. A search for *same predicate, different things* can.\n\n**Why that matters here.** In the test run on our own notes, nearness of meaning was a poor guide to\nwhere a likeness lies: the best ranking held 25 of 57 where chance gives 15. A second try wrote a short\ncard with fixed fields for each passage and let code propose pairs from the cards. Outside judges\ncalled 18 of 34 pairs the same move. A line with fixed slots is such a card. So the layer designed to\nfind sameness could, read the other way, propose likenesses.\n\n**Today, and what it would open.** The design is on paper and nothing of it is built. Its smallest\nexperiment could take one more line: list the pairs with the same predicate and different things, and\nhave a person read them.\n\n---\n\n## Part 2. Book by book: everything else each one teaches\n\n### Book 1. Ziliak and McCloskey, *The Cult of Statistical Significance* (00:32)\n\nThe video's passage and the main lesson are finding 1.\n\n- **Finding it again does not make it bigger.** The creator stresses that the benefit was found\n  repeatedly. That makes it surer that there is an effect. It says nothing about its size. On the\n  graph, each new study that supports *the treatment helps* raises the badge, while the thing\n  established stays *the effect is not zero*.\n- **It is two claims.** *There is an effect*, and *the effect is large enough to matter for this\n  decision*. The second depends on the decision: 200 grams matters to a jockey on the morning of a race\n  and to few others. Deciding whether that much matters is a weighing of one thing against another, and\n  Deliberus has separate questions for weighings.\n- **No standard question asks about size.** Each kind of argument in Deliberus comes with standard\n  questions that test it (they are called critical questions). None of them asks how large.\n- **The word is a trap.** *Significant* means one thing in statistics and another at the kitchen table.\n  A step that rewrites a source in plain words can swap the one for the other without anyone noticing.\n- **For evidence.** A study under a claim has three things to keep apart: what was measured, how\n  surely, and whether that much bears on the claim as it is worded. The first two have fields. The\n  third has nowhere to go.\n\n### Book 2. Woodward, *Making Things Happen* (01:36)\n\n**In the video.** Teenagers who use social media more report more anxiety. Does the one cause the\nother? Do anxious teenagers simply use it more? Or do loneliness, trouble at home or trouble at school\ndrive both? Woodward's answer is to think of a cause through a careful change: alter one thing, keep\nthe relevant background as it was, and ask what happens to the other.\n\n**An everyday case.** Ice cream sales and drownings rise and fall together. Banning ice cream would\nsave nobody. Summer drives both.\n\n- **One observation, three rival stories.** All three explanations of the anxious teenagers rest on the\n  same numbers. More of the same numbers cannot choose among them. Only a change can: take the phones\n  away from some and see.\n- **What the hinge is, exactly.** The hinge runs Woodward's test on our own map: grant this claim, then\n  deny it, and see how far the conclusion moves. It tells what the conclusion depends on in the map. It\n  does not tell what causes what in the world.\n- **How wide a rule may be.** A rule is as good as the range of changes it survives. *Metals expand\n  when heated* survives a change of metal, of place and of century.\n- **A test for a likeness.** A likeness between two fields says that something holds in both. So it is\n  a rule that survives a change of field, and the case against it is a case in the second field where\n  it breaks. The doctor and the company from the video part ways, for instance, where a drug can be\n  stopped again and an approval step seldom is.\n- **For the friend's design.** *Goes together with* and *causes* are different predicates, and a line\n  that says *causes* should name the change it is about.\n\n### Book 3. Lakatos, *Proofs and Refutations* (02:25)\n\nThe video's passage and the main lesson are finding 2.\n\n- **Six things to do with an exception.** With the birds as illustration (the illustration is this\n  note's): give the claim up; say a penguin is no real bird, which keeps the case out by changing the\n  word; list it, *all birds except penguins*; describe it anew, *a penguin flies, under water*; find\n  the step it breaks and write that step into the claim; or look for a deeper account under which the\n  penguin is no exception.\n- **Two kinds of exception.** One breaks a step of the argument and leaves the claim standing: the\n  claim may be true and this argument for it bad. The other breaks the claim. Deliberus's rules for\n  strength already keep these apart. A reason that has been refuted adds nothing and takes nothing\n  away. A required part that fails holds its whole down.\n- **What an argument is for.** One of the students in the book says a proof earns its keep by showing\n  where to doubt, more than by forcing belief. Listing the steps a claim rests on shows where to look\n  for a case against it. That is the use of writing out what a link rests on.\n- **Words stretch in an argument without anyone deciding.** The students who attack keep widening what\n  counts as a solid. The defenders look as if they keep redefining, and may be the ones holding the\n  first meaning still. For the concept layer: the side whose wording changed is not always the side\n  whose meaning changed.\n- **Every added condition costs content.** That is the swan case in finding 1.\n- **Our own notes improve his way.** Of the 304 places where readers found that one note corrects\n  another, 191 narrow a claim and overturn nothing.\n\nA caution from historians: the tidy classroom story fits only a handful of real cases.\n\n### Book 4. Hofstadter and Sander, *Surfaces and Essences* (03:37)\n\nThe video's passage is given under finding 5.\n\n- **Calling a thing by a word is already a likeness.** Each time *chair* is used of a new object,\n  someone has judged it like the chairs before. So two parts of Deliberus that look separate are one\n  ability at two distances. Near: *this use of the word belongs with those uses* (the concept layer).\n  Far: *this debate is like that debate* (the analogy daemon).\n- **That explains a count from our test run.** Readers who were asked for far likenesses mostly brought\n  back near ones, of the kind *an earlier ruling bears on this passage*: three far likenesses among 42\n  findings, then five among 54.\n- **Most categories have no name.** A reviewer retells the authors' case: a man picking up bottle caps\n  on a tour of a great temple calls to mind a small boy, years earlier, absorbed in ants at the rim of\n  a great canyon. Nothing names what the two share.\n- **Naming is where a likeness becomes useful, and where it can go wrong.** Once *a fix that feeds what\n  it fixes* has a name, it can be carried to a third field. It can also be carried to where it does not\n  hold.\n- **How fine a category is depends on what you are doing.** At the table, my glass and yours are\n  different things. In the dishwasher they are the same. The work on when two claims are the same\n  reached that point from philosophy: two claims are the same if they are used the same way.\n- **The book has no test for a wrong likeness**, its critics say. That is why the analogy daemon is\n  designed in two passes, the second of which only tries to break what the first proposed. It is also\n  why a likeness kept between two of our notes should carry one named difference. Today it carries\n  none.\n- **The video's handovers between books are likenesses made to persuade.** The bridge from book 8 to\n  book 9 says, in short, that a decision hides a future self the way a model hides a present reality.\n  The two share a verb and no mechanism. The authors have a name for that: a caricature analogy.\n\n### Book 5. Lakoff, *Women, Fire, and Dangerous Things* (04:50)\n\nThe video's passage and the main lesson are finding 4. The creator also explains the title: Dyirbal, an\nAboriginal Australian language, has a class of nouns that holds women, fire and several dangerous\nthings. He is careful to say that Lakoff sees no shared essence in them. The class grew by a chain of\nassociations that makes sense inside that people's ideas.\n\n- **Why cutting through a word gives reasons and not parts.** Try to split *mother* into its tests. No\n  one test is required (an adoptive mother did not give birth) and none is enough (a surrogate did).\n  Each is a rule of thumb: *she gave birth, so she is the mother, unless something speaks against it*.\n  A category is a bundle of such rules, and such rules are the subject of the next book in the video.\n- **Deliberus's own kinds have this shape.** The graph has one kind of thing, the claim. Since 10\n  October a sentence that a machine writes under a link is a claim as well: one that nobody asserts.\n  That is *foster mother*: a member that lacks one mark of the typical case, with a modifier that says\n  which.\n- **A chain is not sameness.** The Dyirbal class holds together member by member. A goes with B, and B\n  with C, and nothing joins A to C. That is the reason for a standing rule here: claims are never\n  merged through a silent chain of *same as*.\n- **Every list of kinds we keep will have odd members.** Lakoff's word for how a member joins is\n  *motivated*: it makes sense afterwards and could not have been predicted. The *does not fit* answer\n  admits that.\n- **For the friend's design.** Deciding that two mentions are one thing is easy for an individual (a\n  named trial, a law, a person) and is a judgement about meaning for a kind (*basic income*). The\n  second belongs to the concept layer.\n- **For worldview lenses.** A chain that makes sense inside one system of ideas is what a lens would\n  have to carry: the tests a community brings to a word.\n\nA caution from his critics: showing that typical cases exist does not refute the old picture of a box,\nand *motivated* explains little.\n\n### Book 6. Toulmin, *The Uses of Argument* (06:35)\n\n**In the video.** Formal logic says whether a conclusion follows from premises of a set form. Arguments\nin medicine, law, politics and daily life seldom work that way. The creator's case: *students remember\nmore when they take notes by hand, so universities should ban laptops*. The evidence may hold for\nmemory. It does not give the policy. Something unstated joins the two, which Toulmin calls a warrant:\nhere, perhaps, that remembering the most should decide what is allowed in a classroom. And what of\nstudents who need a laptop, of different ways of learning, of the other things a laptop is for?\nToulmin's full model names six things: the claim, the data, the warrant, what backs the warrant, the\nconditions under which the claim holds, and the rebuttals. The weakest part of an argument, the creator\nsays, is often the bridge between evidence and claim that nobody looked at.\n\n**An everyday case.** *It is raining, so take an umbrella.* What joins the two? *Rain makes you wet.*\nOr *wet clothes are unpleasant.* Or *you have a meeting and should look tidy.* Any of them would do.\n\n- **The rule under a link is one of several that would do.** A philosopher who defends Toulmin, David\n  Hitchcock, puts it this way: a step claims that *some* general rule covers it. A rule that someone\n  names is support for that. So if the named rule turns out false, the step may still stand on another.\n- **That is how Deliberus stores a machine's sentence under a link.** It is kept as a claim, it moves\n  no strength, and it starts to count only when a person argues that the step needs it. Hitchcock gives\n  the reason this is right. It follows that *the step needs this* can seldom be argued for a rule, and\n  often for a fact about the case, such as *it is in fact raining*.\n- **He tested whether such rules can be found at all.** He drew fifty arguments at random from a\n  library and could name the rule in forty-nine. In our own trial a machine wrote the rule for 29 links\n  of 30. Neither figure shows whether two readers would write the same rule, and his point is that they\n  need not.\n- **The video's bridge is two things.** From *they remember more* to *ban laptops* runs a value\n  (remembering matters) and a weighing (it matters more than access and the rest). Toulmin's one slot\n  holds both. Deliberus keeps them apart: the value is a claim, and the weighing has its own questions.\n- **Words like *probably* sit on the step.** They say how far the evidence carries, and are no part of\n  the conclusion. That matches a ruling of 28 September: a link has a relevance of its own, apart from\n  how well the claims at its two ends stand.\n- **Evidence is three things**: the observation, the permission to use it for this claim, and what\n  backs the permission. Each can be disputed by itself. That is the founder's position of March, that\n  evidence comes apart into pieces.\n\nA caution: the oldest objection is that in real texts one cannot tell the data from the warrant.\n\n### Book 7. Ashby, *An Introduction to Cybernetics* (08:12)\n\nThe video's passage and the main lesson are finding 3.\n\n- **Two ways to obey the law, and one of them can be done to the world.** More responses, or fewer\n  kinds of trouble. A state that cannot take in a forest's variety can learn more, or it can make the\n  forest simpler. It planted rows (book 9). Deliberus's rule goes the other way: when a form cannot\n  hold what a person said, the form gives way and the person's words stay.\n- **Buffers.** Ashby's critics note that something can soak up a disturbance with no response at all,\n  as a thick wall soaks up a cold night. Deliberus has three such walls. A claim nobody has examined\n  moves nothing, however it is linked. A part written by a machine weighs nothing. Splitting a claim\n  into three parts or twelve gives the same strength. Each takes the reward out of a trick, so nobody\n  has to catch the trick.\n- **To check a thing you need a picture as rich as the thing.** Ashby and a co-author put it as a\n  theorem: a good regulator of a system must be a model of it. A person asked to approve a machine's\n  proposal about a field they do not know can only check that it looks well made.\n- **It explains why our instruction files grow.** A session makes a mistake, and a rule goes into the\n  file that every session loads. Each rule is right. The file grows, a longer file is read less well,\n  and that makes new mistakes. This is the video's company with its approval steps (book 4). The law\n  says why rules multiply: mistakes come in many kinds. It also names the other way, which is to remove\n  the occasion. A guard in code that makes a mistake impossible needs no rule.\n- **A rule has four parts, and one is easily left out.** The creator's map (what to keep, what disturbs\n  it, what shows the disturbance, what you can do) fits a rule for a session exactly. The part that\n  gets left out is the third. A rule that fires when a certain word is about to be written has a\n  sensor. A rule that waits for the session to notice its own state of mind has none, and from outside\n  it looks the same as a rule that works.\n\n### Book 8. Paul, *Transformative Experience* (09:25)\n\n**In the video.** Ordinary advice on decisions says: picture each future, judge how much you would\nvalue it, choose. But some experiences cannot be known before they are lived, and living them can\nchange what you value. The creator's case is becoming a parent. You can collect reports and facts\nbeforehand and still not know it from inside, and it may change what you care about. So the one who\nchooses and the one who lives with the choice may judge life differently. His advice: ask who this\nchoice might make you, what you can know in advance, and whether you would choose partly for the\ndiscovery.\n\n**An everyday case.** Before moving abroad you ask friends who did. Their answers are honest and do\nnot help much, and a year later you see why.\n\n- **It supports a line Deliberus has ruled.** The experience is the person's alone. What surrounds it\n  can be mapped: its causes, what else was possible, what else would have met the same need.\n- **A report of a hard choice should say whether its author has made it.** Advice on having children\n  from someone who has them differs in kind from advice by someone who has not. The planned collection\n  of how people went through hard choices has no field for that.\n- **A person's meaning of a word has a date.** *Tired* means something else after a first child. If the\n  concept layer records whose sense a sense is, it should record when.\n- **A reviewer who has seen the machine's answer is a changed judge.** Shown a proposal, a person has\n  taken on its way of framing the matter before judging it, and cannot go back. That is why the plan\n  for the first review sitting asks for a part in which the person answers before seeing the machine's\n  answer.\n- **Deliberus is itself such an experience.** What it is like to see your own quarrel laid out as a map\n  cannot be told. That is the case for two people trying it side by side before it is explained to\n  anybody, and for the small hand-made example on the first page.\n- **It describes the founder's position too.** A ruling asked for before he has walked the ground is\n  Paul's problem. The walkthroughs exist for that.\n\nA caution from a critic (Bykvist): parts of a new experience can be compared with ones already known. A\nquarrel about a thermostat is such a part.\n\n### Book 9. Scott, *Seeing Like a State* (10:45)\n\n**In the video.** Large institutions make people, forests, cities and economies understandable through\nmaps, measures, categories and simplified models. That makes coordination on a large scale possible. It\ncan also wipe out the local knowledge and the relationships that made the thing work. A forest reduced\nto a count of saleable timber is easy to measure, and the count leaves out soil, animals, the variety\nof plants and what the neighbours used the forest for. The map gets tidier while the land gets poorer.\n\n**An everyday case.** A school judged by test scores starts teaching the test. The scores go up.\n\n- **Deliberus draws such a map, of reasoning.** The sharpest critical reading the project has of itself\n  opens with this book. What a map of an argument leaves out of the argument has not been measured.\n- **Scott's own advice is nearly how the graph is written.** As a reviewer gives it: take small steps,\n  prefer what can be undone, expect surprises, expect people to be inventive. Undoable: machine work is\n  marked and reversible, and a contribution is added and never overwritten (built). Surprises: the\n  *does not fit* answer (built). Inventiveness: a reader may say that the whole picture is framed wrong\n  (ruled, and no page lets a reader do it yet). Small steps: how the work is done, enforced by nothing.\n- **His four conditions for disaster give four clauses for a constitution.** Disaster needs a society\n  made readable, a faith that redesign by plan helps, an authority able to impose the plan, and people\n  unable to refuse it. Deliberus does the first on purpose and believes the second. So its safeguards\n  must sit in the other two: an operator with no power to impose a version of someone's view, and\n  contributors who can refuse one, leave, and take the map along. As four questions: who reads the map,\n  who can change it, who can refuse a version of their view, who can leave with it. Three of our notes\n  ask for a constitution and none has written it.\n- **Readable to whom.** Scott's villagers resisted because the map served the tax collector. People\n  seek out maps that serve themselves. A question for every page: does it make a person more visible to\n  us than we are to them?\n- **The map changes the land.** The forest was replanted to suit the count. Our form of that is people\n  learning to speak in the shapes the graph offers.\n- **A standard meaning is nobody's meaning.** Local measures carried local purposes, and the metre\n  replaced them with one that carried none *(recalled)*. The average sense of a word, as a model writes\n  it, is a metre for meanings. The remedy proposed for the concept layer is in Scott's spirit: show the\n  average as a reference point, and record how each person's meaning differs from it.\n- **Our own list of open decisions is such a map.** It makes positions visible so that they can be\n  ruled on. A thought with no row in it is the forest floor.\n\n### Book 10. Gigerenzer and Selten (editors), *Bounded Rationality: The Adaptive Toolbox* (11:44)\n\n**In the video.** Theories of rational choice picture a thinker with unlimited information, time and\ncomputing power. Nobody has those. So the book stops treating every shortcut as a lapse, and asks how a\nsimple rule can make use of the way a situation is built. The creator adds that shortcuts have to be\nused with care, and says the authors set out how.\n\n**An everyday case.** Choosing a restaurant in a strange town by which one is full of locals. One clue,\nno weighing, usually right.\n\n- **Less can be more, and why.** One of the programme's rules, *take the best*, looks at clues in order\n  of how often each has been right, and decides on the first clue that separates the options. In\n  prediction it matched or beat methods that weigh everything. The reason given: with little data, a\n  rule that ignores how the clues depend on each other does better than one that guesses the dependence\n  badly.\n- **That supports a rule Deliberus already has.** A claim with required parts is judged by its weakest\n  part. One bad part decides, and nothing makes up for it. The argument for the rule was that it stays\n  right when the parts share an origin and nobody knows how much. Gigerenzer found the same in another\n  field.\n- **A caution for our note on evidence.** That note found that answers to the standard questions are\n  counted where every tradition says to weigh them. This tradition says that counting does about as\n  well as weighing when there is too little data to set the weights. So keep counting until a weight\n  can be measured.\n- **Every limit works somewhere.** Deliberus has several quick rules: ask first the question on which\n  most hangs; stop when no answer could change what is shown; compare each claim with its twelve\n  nearest neighbours; show about ten boxes. The third was tested against doing the whole job and kept\n  20 of 24 objection links, on one pair of opposed texts. A limit should say where it was measured.\n- **His situations do not answer back.** A simple rule that becomes known can be gamed. Advertising\n  buys recognition, which defeats *choose the one you recognise*. Ours will meet people who answer\n  back.\n- **Where a person's effort should go.** His critics found that putting several clues together is easy\n  for people, and that deciding how to go about a decision is the hard part. So the machine should\n  choose the next question and the person should answer it. That is Deliberus's principle that the\n  machine leads the way down, with an experiment under it.\n\nA caution: people do not reliably use such rules even where they would work, and the toolbox does not\nsay how a tool gets chosen.\n\n### Book 11. Jullien, *A Treatise on Efficacy* (12:34)\n\n**In the video.** One familiar Western picture of acting is to design the ideal result and then impose\nit. Jullien sets against it ideas that he rebuilds from Chinese thought, in which the strategist\nstudies how things already lie and changes the conditions, so that the wanted result comes more and\nmore easily. (The creator says *Efficiency*. The title is *A Treatise on Efficacy*.)\n\n**An everyday case.** To get a child to read you can order it to, or you can leave good books around\nthe house and read yourself.\n\n- **It describes how this project moves.** Stay cheap, keep going, and be in place before things shift.\n  A push for fast growth would be the other picture.\n- **It describes how the ontology has been grown.** The project calls settling the philosophy first a\n  closed road. Categories are left to come out of runs on real material, and each list is expected to\n  break on the next kind of text.\n- **It describes how sessions came to be steered.** Earlier instruction files said *must* and *never*.\n  Later ones place a short cue at the spot where the act happens.\n- **It carries a warning, from Jullien's own last chapters.** Arranging conditions so that people do a\n  thing as if of themselves is also how manipulation works. Deliberus is built against that, so the\n  cues that steer a session are written down where the founder can read every one.\n- **Three of the video's books say one thing from three sides.** The structure is in the situation and\n  not in the plan. Scott says it from the side of those planned upon, Gigerenzer from the side of the\n  one deciding, Jullien from the side of the one acting. Ashby gives the arithmetic.\n\n### Book 12. Mercier and Sperber, *The Enigma of Reason* (13:01)\n\n**In the video.** We picture reasoning as a private road to truth. The authors propose that reason grew\nmainly for justifying our claims to others and for judging theirs. The creator notes that this is\ndisputed and that thinking alone keeps its worth. But it explains a pattern: alone we are good at\nbuilding a case for our first hunch, and under the right conditions other people spot the weak points\nwe missed. His advice: after reasoning through an important decision, do not ask someone whether they\nagree. Split the roles. One builds the strongest case for the conclusion, another judges it without\nhaving to defend it, and then they swap. A group whose disagreement is well arranged can be cleverer\nthan any of its members.\n\n**An everyday case.** You proofread your own letter three times, and a friend finds the error in ten\nseconds.\n\n- **For machine work this is already how Deliberus is built.** Every split of a claim is checked by a\n  separate reader. A reader of our notes is never their writer. No step grades its own work. A verdict\n  is a record, and a challenge arrives as a new claim, so a judge cannot be talked round in\n  conversation.\n- **For people it is a script.** Two people at one screen: each writes the strongest case for the\n  other's claim, the other accepts or refuses that version, and then they swap. Refusing a version of\n  your own view is the one act the project reserves for the person whose view it is. The two-person\n  trial is planned and has not been held.\n- **A right answer convinces where it lands on the faulty step.** In one study the authors cite, people\n  who had gone wrong were not moved by the correct working until it addressed the step where they went\n  wrong. That is the case for being able to point at one step: it turns *something is off* into a\n  change of mind.\n- **A reason given on request is made for a listener.** It need not be what caused the belief. That is\n  enough here, since the design needs something that can be answered and does not need the cause.\n- **Who brings the evidence.** The reasons under a claim are brought by its friends. Their\n  one-sidedness is to be expected, and nothing records it.\n- **What may not be borrowed.** The strong results are for problems whose answer can be shown, as with\n  a puzzle. The project's central bet is about values, and *the best argument wins* does not carry over\n  to it untested.\n\nA caution: critics ask what would refute the theory, and doubt its story about evolution. Nothing above\nrests on that story.\n\n---\n\n## Part 3. What the reading of Wittgenstein adds\n\nThe founder merged another reading on 2026-10-12:\n[the-tractatus-and-deliberus.md](the-tractatus-and-deliberus.md), on Wittgenstein's first book, the\n*Tractatus*. In three sentences: that book wanted what Deliberus wants, which is to make thinking clear\nby taking statements apart. It assumed three things Deliberus has given up: that the taking apart ends\nin final pieces, that the limit of what can be said is fixed beforehand, and that one mind does the\nwork alone. Its author called it final, and outside readers broke it within eleven years.\n\nNeither note mentions the other. Ten places where they meet:\n\n| In the twelve books | In the Wittgenstein note | Together |\n|---|---|---|\n| Lakatos: a reason is written into the claim and becomes a part (finding 2) | Two kinds of relation. One holds because of what two sentences say (*is this a part of that?*). The other holds because of how the world is (*is this a reason for that?*). Its proposal AJ3: never answer a question of one kind with a link of the other | Lakatos explains why the two keep getting mixed: one sentence changes kind over time. The Wittgenstein note gives the rule, and Lakatos gives the test and the history |\n| A claim can stand well by saying little (finding 1) | What a statement says is what it rules out. *It rains or it does not rain* tells nothing about the weather. Its warning AJ5: values that everyone shares may be shared because they rule nothing out | One measure for both: what does this claim, or this shared ground, exclude? |\n| Ashby: a list with too few answers must misfile (finding 3) | A notation must be able to tell apart as many things as there are to tell apart, and it cannot see its own shortfall from inside | One law, stated once for regulators and once for notations. It is why the way out has to be built in beforehand |\n| Hofstadter and Sander: my glass and yours are one thing in the dishwasher (book 4) | The later Wittgenstein on a chair: whether its simple parts are bits of wood or atoms depends on what you are asking | How far to take something apart depends on the purpose, for categories and for parts alike |\n| Lakoff: *mother* as several tests (finding 4) | The corpus cites the later Wittgenstein (family resemblance) against taking things apart, while its own splitter runs the earlier one | Lakoff's second chapter is titled *From Wittgenstein to Rosch*. Finding 4 is family resemblance given a working form: tests, and edge cases to ask about |\n| Lakoff: both sides agree on the tests and differ on which one governs | A person who says the picture is framed wrong could answer with a second map of the same question. Its proposal AJ2 | A dispute over which test governs *fair* could be shown as two maps of one dinner, one under each test |\n| Toulmin: the rule under a link (book 6) | (not in that note) | Toulmin was Wittgenstein's student at Cambridge, in 1941 and in 1946 to 1947. A paper of 2024 traces his *backing for a warrant* to an idea in the *Tractatus* (read as a search excerpt) |\n| The friend's design: one line with fixed slots for each claim about what is the case (finding 5) | The *Tractatus*'s picture of a statement: names standing in a fixed arrangement. In the story told of how it broke, the economist Piero Sraffa brushed his fingertips under his chin and asked what the logical form of that was (an anecdote, told in two versions) | The design is the early Wittgenstein in small. It is safe for two reasons that note supplies: it is kept to claims about what is the case, and its list of predicates has a *does not fit* answer. A refusal, a hedge, or *you framed this wrong* will fit no slot |\n| The design: the one who says a thing is never its subject | *A says p* does not make p true or false | Both keep the report apart from what is reported |\n| Mercier and Sperber: nobody checks their own case well (book 12) | The book stood for years because nobody was attacking it. Its proposal AJ6: count the mistakes found per attack, not per month | A quiet period with no outside reader is no evidence of soundness. In Ashby's terms it is a regulator with no sensor |\n\nOne of the ten is worth more than a row. **Finding 2 and that note's third proposal are one proposal\nseen from two sides.** Suppose every kind of link is marked as being *about what the two sentences say*\nor *about the world*. Then Lakatos's repair is the one lawful way for a sentence to move from the\nsecond kind to the first, and a record of how a claim answered an attack is what would show that the\nmove was made.\n\n---\n\n## Part 4. Where the friend's idea came from\n\nThe founder asked for the video to be read against *\"[a friend]'s recent idea of predicate logic-like\ntracking\"*. The long note took that to be a design in a friend's mail, and said it might be aimed\nwrong. On 2026-10-12 he placed it: the friend's idea, from a visit the week before.\n\nThe record from that week holds one design that fits, and it is the one the long note read\n([store-the-uses-read-the-meaning.md](store-the-uses-read-the-meaning.md) § 4b). Its own word for the\nmiddle slot is *predicate*, and its predicates come from a short fixed list with a *does not fit*\nanswer. It was written out by the friend's coding assistant, on a question the founder put to it, in\nanswer to the friend's own wish: something exact and compact, where a store of sentences would be a\ntangle. So finding 5 was aimed right.\n\nWhat the record cannot show is whether the friend said more in conversation than the mail holds.\n\n---\n\n## What was not done\n\n- **No book was read**, as before. The long note lists what was read about each.\n- **Nothing was built or run.**\n- **The video's words are not quoted here.** Each passage is retold with its time code.\n- **Part 3 is one session's reading.** The Wittgenstein note was read in full. Two outside facts were\n  looked up for it and are in the sources below. Nobody else has checked the ten rows.\n- **The everyday cases are made up**, and the six answers to the penguin are this note's illustration,\n  not Lakatos's own examples.\n\n## Sources added for this note\n\nRead 2026-10-12 as search excerpts.\n\n- Toulmin as Wittgenstein's student, and his model's debt to the *Tractatus*: *Wittgenstein and\n  Toulmin's Model of Argument*, Argumentation, 2024 (its author's name was not in the excerpt read)\n  (https://link.springer.com/article/10.1007/s10503-024-09637-w); the New York Times obituary of\n  2009-12-11 (https://www.nytimes.com/2009/12/11/education/11toulmin.html).\n- Lakoff's chapter *From Wittgenstein to Rosch*: the publisher's page with the table of contents\n  (http://www.press.uchicago.edu/ucp/books/book/isbn/9780226471013.html).\n\n## Cross-references\n\n- [twelve-books-read-against-deliberus.md](twelve-books-read-against-deliberus.md): the long reading,\n  with its sources, its nineteen candidates (§ 16) and what was not done (§ 17).\n- [four-books-the-corpus-had-not-read.md](four-books-the-corpus-had-not-read.md): the first reading of\n  four of the twelve.\n- [the-tractatus-and-deliberus.md](the-tractatus-and-deliberus.md): the reading Part 3 is set beside.\n- [store-the-uses-read-the-meaning.md](store-the-uses-read-the-meaning.md) § 4b: the friend's design.\n- [the-analogy-daemon.md](the-analogy-daemon.md) § 8: the test run on our own notes.\n"}