{"path":"research/theorists-on-normic-rules-exceptions-numbers-and-many-minds.md","content":"# What the theorists would say: normic rules, exceptions, one number, and many minds\n\n*22 October 2026. His ask: \"Find a bunch of other theorists and philosophers and what they'd think about all\nthis.\" Claude's reading, nothing ruled. Session b66ddc4d, the outside reader of the night's arc.*\n\n## 0. What \"all this\" is, and what already stands\n\n**The night's moves, each with its home:**\n\n| | Move | Home |\n|---|---|---|\n| A | The rule under a link is a *reason* for the application, and the application sentence is the link's relevance claim | `what-relevance-is.md` § 3b; `retrofitting-the-reasons-channel.md` § 9h–9j and its correction |\n| B | An exception is a fact plus a rule aimed at the application, never at the conclusion | `what-relevance-is.md` § 3b; retrofit § 9g |\n| C | A rule's *usually* is a scope word: normic (a failure calls for an explanation) becomes the universal plus named exceptions; statistical becomes a stated proportion; a rule claim keeps one number | `what-relevance-is.md` § 3f; ledger AT1, AT2 (ruled) |\n| D | A failure with a named kind is an exception and leaves the rule's number alone; without one it counts against the rule | ledger AT3 |\n| E | Within one argument the weakest premise governs (ruled), or premises multiply (proposed); across arguments the best counts until independence is read, then they add | ledger Q38; retrofit § 9j |\n| F | Rules about groups of people start statistical | ledger AT4 |\n| G | A rule set held in supposition is tried out over cases, by computation, no model playing a person | ledger AT5, AT6 |\n| H | A lens includes which failures a person finds in need of explaining | ledger AT7 |\n| I | An obvious carry proceeds and the person is told; attribution partly evidence, partly etiquette | ledger AS1, AS2 |\n\n**What the corpus already has, and this note does not redo.** `philosophy-of-the-unwritten-step.md` § 1\nreads fifteen traditions on the step and its rule (Stoics to Frege; Sellars and Brandom on stating the\nrule; Kant and Wittgenstein on judgment; Toulmin and Hitchcock on the warrant; Pollock on the undercutter;\nDancy on favourers and enablers; Gadamer, Oakeshott and Polanyi on the background). `academic-foundations.md`\nreads twelve books, among them Gigerenzer on the weakest-part rule and Scott on legibility.\n`what-counts-as-evidence.md` § 2c–2d has Condorcet on independence, Haack's crossword, Brandom's\ndefault-and-challenge, and the Duhem–Quine point that blame localises once auxiliaries are claims.\n`weakest-link-arithmetic-and-the-merge-hunch.md` has Cohen, Clermont and the conjunction problem.\n`starting-values-and-the-problem-of-the-prior.md` § 2.2–2.3 has Hájek on reference classes and Joyce on\nweight, balance and imprecise priors. `normic-rules-normative-reasons-and-trying-rules-out.md` has Smith,\nSchurz, Horty, Bear and Knobe, Leslie. Where a name below is already there, this note adds only the angle\nthe night's moves open.\n\n**Read depth, said once:** every philosopher below is from memory, not checked this session; the corpus\nnotes named above were read this session. A sentence attributed to a thinker is a paraphrase unless quoted.\n\n## 1. On the rule as a reason and the exception that attacks the application (moves A, B)\n\n**Aristotle, *Nicomachean Ethics* V.10, on equity (epieikeia).** Law speaks universally, and cases arise\nwhere the universal is wrong; equity corrects the law there, *as the lawgiver would have said had he been\npresent*. That is move B in its oldest form: the fact of the case plus a rule the lawgiver would own,\naimed at this application, leaving the law standing. His test for the correction is the retrofit's test\nof reach (§ 9g): a correction the lawgiver would have written into the law is a narrower rule; one he\nwould not is this case only. **Takes:** the home-by-reach rule has a name and a pedigree; a note on it can\nsay *equity*.\n\n**H. L. A. Hart, \"The Ascription of Responsibility and Rights\" (1949).** Legal concepts are *defeasible*:\na contract is valid *unless* duress, fraud, incapacity. The defences do not deny the facts of the\nagreement; they defeat the ascription. Hart introduced the word that Pollock and the AI tradition later\nbuilt on, and his form is exactly his stream of thought on *unless* claims: the exceptions live beside the\nrule as an open list, and no finite list closes it. **Takes:** the connection questions `rule_exceptions`\nand `exception_absent` in `deliberus/extraction/schemes.py` are Hart's defences, asked of every\napplication.\n\n**Frederick Schauer, *Playing by the Rules* (1991).** A rule is an entrenched generalisation whose point is\nto decide the case even when its justification misfires there; a rule that yields to every good exception\nis particularism wearing a rule's clothes, and what rules buy (predictability, less trust needed in the\ndecider, errors of a known shape) is bought with that resistance. He would read AT3's guard as the price of\nrulehood and the normic form's *rule's number untouched* as entrenchment. **Takes:** count, per rule, how\noften its exceptions fire against how often the rule decides; a rule whose exceptions decide more cases\nthan it does has stopped being one, and that count is the signal AT3 says accumulates.\n\n**Ronald Dworkin, \"The Model of Rules\" (1967), and Robert Alexy's weight formula (2003).** Rules apply\nall-or-nothing and meet a conflict by an exception or by one rule's invalidity; principles have a dimension\nof weight and meet a conflict by weighing. Tonight's normic rule with exceptions is Dworkin's rule; the\nweighing scheme handles his principles; AT5's *conflict with no priority to settle it* is where a\nDworkinian would say the trial has reached principles. Alexy's formula for the principle side multiplies\nthe intensity of interference, the abstract weight and *the reliability of the empirical premises*, and\nthat last factor is a rule's credence inside a weighing: one number, in his arithmetic too. **Takes:** the\npass that writes a rule (AT2) asks one more question, whether a conflict with another rule calls for an\nexception or for weighing, and routes to the weighing scheme on the second answer.\n\n**Jonathan Dancy, *Ethics Without Principles* (2004).** The standing challenge. A reason in one case can be\nnone in another; favourers, enablers and intensifiers come apart; no rule survives contact with the\nparticulars. He would say a normic rule plus exceptions is case-by-case judgment with bookkeeping, that *the\nrule as written* has no credence apart from its cases, and that the exception list grows without end. Horty\n(the normic note § 3) answers him with priorities among defaults. **Takes:** AT1 stands only while AT3's\nguard holds, since an exception that fits one case is Dancy's particular judgment entered as a rule; the\nguard is what keeps the rule's number from being a fiction.\n\n**W. V. O. Quine, \"Two Dogmas\" (1951).** When a prediction fails, something in the web gives, and the maxim\nof minimum mutilation says revise the periphery before the centre. AT3's default (mint an exception claim,\nleave the rule's number alone) is that maxim; Popper's and Lakatos's warnings about saving a theory\nforever are the brake on it, and AT3's reach guard is the brake written down. **Takes:** nothing new to\nbuild; the default has a reason that can be cited when someone asks why the rule, not the exception, is\nprotected first.\n\n## 2. On normic against statistical, and the one number (moves C, D, F)\n\n**Carl Hempel, the requirement of maximal specificity (1965).** A statistical explanation must use the\nnarrowest reference class about which reliable statistics are known, or the same case gets a 0.9 under one\nclass and a 0.1 under another (his *ambiguity of inductive-statistical explanation*). The relevance\nsession's answer under § 3f (a stated proportion inherits the reference-class problem; named exceptions let\nthe case's particulars in one at a time) is Hempel's requirement met by the normic form: each named\nexception narrows the class by one particular. **Takes:** the statistical form must name its population\n(the testable-bets note already asks this of a scope word), and two populations giving two proportions are\ntwo claims shown side by side, never averaged. Hájek's point that the problem is everyone's\n(`starting-values` § 2.2) says the normic form does not escape it either; it only takes it one named\nparticular at a time.\n\n**Paul Pietroski and Georges Rey, \"When Other Things Aren't Equal\" (1995).** A *ceteris paribus* law is\nnon-vacuous only if each failure is explained by an interfering factor that is independently specifiable\nand does explanatory work elsewhere. That is AT3's guard (an exception must name a kind that applies to\nother cases and can fail on its own) in another vocabulary, published thirty years ago against the charge\nthat cp laws are empty. **Takes:** cite it as the guard's home, and take its *independently* seriously: an\nexception claim should carry evidence of its own beyond the case it rescues. Nancy Cartwright (*How the\nLaws of Physics Lie*, 1983) adds that most laws people actually write hold only under shielding\nconditions, so most rules in the graph will be normic rules with an open list, and the guard will be\nexercised constantly.\n\n**Nelson Goodman, *Fact, Fiction, and Forecast* (1955).** Which generalisations may be projected depends on\nthe record of their predicates (the unwritten-step table already has this). **Takes:** the pass that writes\na rule under a link should prefer entrenched predicates, and the rule's count across links (Mill's\nregister, same table) is the record it reads.\n\n**Thomas Kuhn, *Structure* (1962).** An anomaly is a failure that calls for explanation *against a\nparadigm*, and normal science shelves most of them. So the explanation question (AT2) is answered from a\nbackground, not from the world alone, which is § 2 of the normic note said from the science side.\n**Takes:** AT2's clause that a contested kind is a claim and never a setting has Kuhn's support; and a\npass that shelves failures the way normal science does should be visible doing it.\n\n**Charles Peirce, \"The Fixation of Belief\" (1877) and the schema of abduction.** Inquiry begins in surprise,\n*the surprising fact C is observed*. A normic failure is a surprising fact; a statistical failure is not.\nSo AT7's plain question, *would that call for an explanation?*, is Peirce's trigger of inquiry, and asking\nit of a person is asking where their inquiry would start. **Takes:** AT7 and the descent's *why does that\nneed explaining?* are one question; the normic note's § 5 can say so.\n\n**Philippa Foot, *Natural Goodness* (2001), and Michael Thompson, *Life and Action* (2008).** *The lion has\nfour legs* is true though some lions have three; the three-legged one is defective, and the defect calls\nfor an explanation. Natural-historical judgments are normic by their logic, which is the deepest home for\nthe normic form. It is also the hazard AT4 names: reading a social group as a life-form is the essentialist\nmove (Leslie, in the normic note). Foot and Thompson would draw the line where AT4 draws it: normic for\nkinds that have functions or norms of their own (organisms, artefacts, institutions with purposes),\nstatistical for aggregates of people. **Takes:** AT4 has a principled basis beyond stereotype avoidance; the\npass can ask whether the subject is a kind with a function or a population.\n\n## 3. On weakest premise against product, and the start (move E)\n\n**E. T. Jaynes, and the Fréchet bounds (a standard result).** For a conjunction of\npremises the probability lies between max(0, a + b − 1) and min(a, b); independence gives the product a × b.\nSo *the weakest premise governs* is the **upper** Fréchet bound, the most favourable number a conjunction\ncan have, and the product sits below it. Jaynes would add that min is not a probability rule at all but\nthe limit of total positive dependence, and that dependence is handled by conditioning, never by a minimum.\n**Takes:** Q38's ruled first sentence takes the most generous reading of an argument, and § 9j's proposal\n(multiply by default) the less generous one, which is what § 9j's own principle, *both defaults should err\nlow*, asks for; the ruling and the proposal differ by which end of the bounds is the default when\ndependence is unread, and that is the sentence to put to him.\n\n**Susan Haack, *Evidence and Inquiry* (1993), the three dimensions.** Supportiveness, independent security,\nand *comprehensiveness*: how much of the relevant evidence the structure includes. Comprehensiveness\ncannot be reduced to the other two. The parts channel's exhaustiveness discount (the whisper\n`weakest-link-arithmetic-and-the-merge-hunch.md` § 7 calls an accident of inherited constants) is a\ncomprehensiveness term: the parts may not exhaust the whole. **Takes:** give the discount Haack's name and\na stated reason, and since she holds the dimension irreducible, the proposed coverage-gated lifting is the\nright shape, a separate number and not a constant.\n\n**Gerd Gigerenzer, take-the-best.** Already read as support for the weakest-part rule (`academic-foundations`).\nThe across-argument rule, *the best argument counts until independence is read*, is take-the-best itself:\na lexicographic stop at the first discriminating cue, which beats weighing when data is scarce and cue\ndependence is unknown. **Takes:** the rule is the right default exactly while independence is unread, and\nthe *then add* step is the moment data arrives; Gigerenzer's known failure condition (cues of similar\nvalidity) is where the default should be measured.\n\n**Judea Pearl.** Independence between two arguments is a question about a graph: two premises with a shared\norigin have a common cause, and conditioning on it is what Q4's provenance node does when it becomes a\nnode. **Takes:** the keeper reading independence (Q40) is reading whether the two arguments are\nd-separated once their sources are drawn; the corpus's rule that a known shared source becomes a node, never\na blind discount (`what-counts-as-evidence` § 2c), is Pearl's answer.\n\n**John Maynard Keynes, the weight of argument (1921), and Glenn Shafer (1976).** Both already in the corpus:\nweight apart from balance in `evidential_weight.py`; the 0.5 start as a neutral point rather than a vacuous\ninterval (`starting-values` § 2.3 answers Shafer's objection with the inertia problem). **Takes:** nothing\nnew; case 6 of the gate script, *unexamined, no lean*, is a weight statement and should read as one.\n\n## 4. On trying rule sets out (move G)\n\n**Karl Popper, piecemeal social engineering and theories dying in our stead.** A trial over a rule set in\nsupposition is a policy's consequences dying in the mind of Deliberus instead of in the world, and his\npiecemeal method (change one thing, watch what it refutes) is the shape of AT5's outputs: the conflicts and\nthe undecided cases are what the rule set fails to settle. **Takes:** the trial reports first what would\nrefute the rule set, then what it decides.\n\n**Elinor Ostrom and Sue Crawford, \"A Grammar of Institutions\" (1995).** Every institutional statement\nparses into Attributes (to whom), Deontic (may, must, must not), Aim (do what), Conditions (when), and\nOr-else (the sanction). A rule written for AT5 has these five slots, and the fifth is what AT6's\nconsequence trial needs. Ostrom also separates rules-in-form from rules-in-use. **Takes:** adopt the five\nslots as the shape of a rule claim written for a trial; and report the gap between the written rule set\nand what people are seen to do as a finding of its own, since a trial can only run rules-in-form.\n\n**Friedrich Hayek, \"The Use of Knowledge in Society\" (1945), with James Scott, Michael Polanyi and Michael\nOakeshott already in the corpus.** Rules evolve, and much of what makes them work is dispersed and tacit; a\ntrial over written rules misses what their practitioners know. Yet he would prize a deterministic trial with\nno model in it, since predictability of rules is his rule of law. **Takes:** AT5's line, *no model plays a\nperson*, and its reporting of undecided cases as gaps rather than filling them, are the two things that keep\nHayek's and Scott's objections at bay; neither may be relaxed for convenience.\n\n**Herbert Simon and Charles Lindblom.** Policy is made by successive limited comparisons from where things\nstand, not from first principles, and people satisfice in the cases no rule decides. **Takes:** show a trial\nas a difference between two rule sets (the standing one and the proposed one), not as a run from nothing;\nthe undecided cases are where satisficing happens today and are worth listing by themselves.\n\n## 5. On lenses and attunement (move H)\n\n**Hannah Arendt, representative thinking.** To form an opinion is to consider the matter from the standpoints\nof those who are absent, *going visiting* in thought. A lens computed from a holder's priorities and from\nwhat they find in need of explaining is that visit, offered. She would warn that a lens read from someone's\nwords can replace visiting them with a model of them; AN3's line, that a lens is made of holders' words and\nno model plays a holder, is her guard. **Takes:** a lens is a prompt for the user's own representative\nthinking, shown as *given their weights, this follows*, never as *they would say*.\n\n**Iain McGilchrist, already read in `coupling-the-poles.md`.** The explanation question is the right\nhemisphere's attention to the particular that does not fit; the pass that writes a rule is the left's. His\npathology is the emissary keeping the output. **Takes:** AT2's pass returns its rule to the cases it came\nfrom and re-reads them, the return edge the coupling note names; a rule that never meets its cases again is\nthe non-return.\n\n**Hartmut Rosa, *Resonance* (2016).** Trials and lenses are modes of reaching out; resonance is being reached\nand answering, and it needs the possibility of being changed. The exchange the normic note § 5 imagines\n(*we agree on what happened; you see an exception, I see the ordinary run*) is a resonance moment only if\neither side can come away different. **Takes:** an attunement feature leaves a way for the lens's holder to\nbe moved, not only seen.\n\n**John Vervaeke, relevance realization.** Relevance is realised, not computed; the frame problem says no\ngeneral procedure exists. Making relevance a claim people argue hands realisation to them, and the 0.5\nstart is relevance not yet realised. His opponent processing, compression against particularisation, is the\nstatistical form against the normic form: a proportion compresses the cases, named exceptions particularise\nthem. **Takes:** the pass choosing a form (AT2) is choosing a point on that trade-off, and a mixed rule is\nboth at once.\n\n**Daniel Schmachtenberger and Stafford Beer, his other lenses.** Schmachtenberger would read *who thought of\nit first* as rivalrous dynamics in an epistemic commons and the sameness keeper (one claim, many holders) as\nthe third-attractor shape. Beer would place the trial as a society's System Four, its model of the\nenvironment, and ask what the trial's outputs *do*, since the purpose of a system is what it does.\n**Takes:** nothing to build; two ways to say why AT5 is not a curiosity.\n\n## 6. On never bottlenecked, and what attribution carries (move I)\n\n**Robert Brandom, *Making It Explicit* (1994), deontic scorekeeping.** The graph is a scorekeeper of\ncommitments and entitlements. Who asserted a claim fixes who is committed by it; entitlement to reuse it is\nsocial and can be inherited by anyone, as in testimony. So for the sorting AS1 leaves open (which\nprovenance rules carry evidence, which are etiquette): tracking commitment, who stands behind a claim, is\nstructural bookkeeping; tracking priority, who said it first, is etiquette. **Takes:** keep the first,\nloosen the second; the starting-values note's *who asserted a claim is its first evidence* is the first\nkind.\n\n**Niklas Luhmann.** Society is communication, persons are addresses communication attributes itself to, and\nthe same thought arising in many heads is communication reproducing itself. He would endorse the\nnever-bottlenecked ruling without hesitation and add a warning: a system that processes communication\ndevelops its own code and becomes opaque to its participants, and its selectivity, what it lets count, is\nthe thing to watch. **Takes:** the outside checks and the gate scripts are that watch; keep them.\n\n**Bruno Latour, *Science in Action* (1987).** A claim's standing is the allies it enrols, machine passes are\nactants in the network, and an obvious carry is a black-boxing. He would ask who can reopen the box. **Takes:**\nAS2's *visible and restorable* is the answer, and it must stay true of every carry, not only the contested\nones.\n\n**Jürgen Habermas, already read in `habermas-machine.md`.** A carry that proceeds without the author's yes is,\nin his terms, a system imperative standing in for communicative action, acceptable only while the person can\ncontest it. **Takes:** AS1's notice to the person is not optional interface; it is the condition under which\nthe ruling is Habermasian at all, and its channel (unbuilt) is owed before the carry is.\n\n**Hugo Mercier and Dan Sperber, already read.** Reasoning's evaluative side is good and its productive side\nbiased, so distributing evaluation across many minds is where the theory predicts reliability. **Takes:**\nAS1's bet is the one their results support; the limit they state (strong results for problems with a\nshowable answer) marks where an obvious carry may not be obvious.\n\n**Peirce and Dewey, the pragmatist line the threads synthesis says the corpus skipped.** Truth as what the\ncommunity of inquiry converges on, and Dewey's public formed around shared consequences. The\nnever-bottlenecked ruling is theirs. **Takes:** `threads-synthesis-and-gaps.md` already names the Dewey gap;\nthis is one more reason to read *The Public and Its Problems* against AS1.\n\n**Richard Rorty, for completeness.** He would call Horty's logic of defaults a mirror of nature for ethics,\nsince a reason is what one's peers let one get away with, and then feel at home in a graph where justification\nis many minds in conversation and the rules are people's own words. The bookkeeping and the conversation are\nnot rivals here; the extensions computation is the bookkeeping of the conversation.\n\n## 7. What Deliberus takes from them, in order of bite (all unruled)\n\n1. **Q38's two defaults are the two ends of the Fréchet bounds** (§ 3): *weakest governs* is the upper\n   bound, the product sits below it. The question to him becomes which end is the default while dependence\n   is unread, given the ruled principle that defaults err low.\n2. **The exhaustiveness discount is Haack's comprehensiveness** (§ 3): name it, give it a reason, and make it\n   a separate number, as the coverage-gated proposal already does.\n3. **AT3's guard has a home, Pietroski and Rey 1995** (§ 2), and its *independently* asks that an exception\n   claim carry evidence beyond the case it rescues.\n4. **The statistical form names its population, and two populations are two claims** (Hempel, § 2).\n5. **AT2 asks one more question: exception or weighing** (Dworkin, § 1), and routes to the weighing scheme\n   on the second.\n6. **Per-rule counts of exceptions firing against the rule deciding** (Schauer, § 1) are the accumulating\n   signal AT3 names; cheap to keep.\n7. **AT4's line is principled** (Foot and Thompson, § 2): normic for kinds with a function, statistical for\n   aggregates of people.\n8. **A rule claim for a trial has five slots** (Ostrom and Crawford, § 4), and a trial is a difference between\n   two rule sets (Simon and Lindblom).\n9. **AS1's notice is the condition, not the interface** (Habermas, § 6); **commitment-tracking stays,\n   priority is etiquette** (Brandom, § 6).\n10. **AT7's question is Peirce's surprise** (§ 2), one question with the descent's.\n\n## Sources\n\nAll from memory, not checked this session: Aristotle, *Nicomachean Ethics* V.10; Hart 1949; Schauer 1991;\nDworkin 1967; Alexy 2003 (*On Balancing and Subsumption*); Dancy 2004; Quine 1951; Hempel 1965 (*Aspects\nof Scientific Explanation*); Pietroski and Rey 1995; Cartwright 1983; Goodman 1955; Kuhn 1962; Peirce 1877;\nFoot 2001; Thompson 2008; Jaynes 2003; the Fréchet bounds; Haack 1993; Gigerenzer and Goldstein 1996;\nPearl 1988; Keynes 1921; Shafer 1976; Popper 1945 and 1972; Crawford and Ostrom 1995; Hayek 1945; Simon\n1955; Lindblom 1959; Arendt, *Lectures on Kant's Political Philosophy*; McGilchrist 2009; Rosa 2016;\nVervaeke, Lillicrap and Richards 2012; Brandom 1994; Luhmann 1984; Latour 1987; Habermas 1981; Mercier and\nSperber 2017; Dewey 1927; Rorty 1979. Read this session: the corpus notes named in § 0.\n\n## Cross-references\n\n`normic-rules-normative-reasons-and-trying-rules-out.md` (the arc this reads against) · `what-relevance-is.md`\n§ 3b, § 3f · `retrofitting-the-reasons-channel.md` § 9g, § 9j · `philosophy-of-the-unwritten-step.md` § 1\n(the fifteen angles this does not repeat) · `weakest-link-arithmetic-and-the-merge-hunch.md` § 7 ·\n`what-counts-as-evidence.md` § 2c–2d · `starting-values-and-the-problem-of-the-prior.md` § 2.2–2.3 ·\n`a-part-and-a-reason.md` § 10 · `coupling-the-poles.md` · `../worldview-lenses.md` · `habermas-machine.md` ·\n`threads-synthesis-and-gaps.md` · `../academic-foundations.md`.\n"}