{"path":"research/a-part-and-a-reason.md","content":"# A part and a reason\n\n**Date**: 2026-10-12 · **Type**: a session's reading of the corpus, with three outside lookups ·\n**State**: nothing here is ruled; it sharpens ledger row T9 and adds no row of its own ·\n⚠ **corrected the same afternoon**, on his challenge (§ 5a): the first version cited the ruling of\n30 September on what a failed reason does without having read it. That ruling had already put a\npart and a reason on one device. § 0, § 2 and § 4 changed, and § 2a, § 2b and § 5a are new\n\n**His question**, on being told that the splitter files a claim's reasons as its parts:\n*\"What differentiates a part from a reason, truly? Ground deeply on what the corpus has reasoned\nhistorically on this and philosophize about it and think deeply. Use the corpus graph and think\nabout distant connections between different parts of the corpus here. Think about possible synonyms\nand perspectives on this question so we're not limited to one frame on it.\"*\n\n**And on the proposed change to the splitter** (ledger T9), the same hour: *\"Yeah sounds good but\nlet's await the deep dive and what it gives back before we make any such change.\"* Nothing in the\nsplitter was changed.\n\n---\n\n## 0. The short version\n\n- **A part is something the claim says. A reason is something that speaks for it.** The test, which\n  the notes have held since 11 October: *if this sentence fell, would the claim be false, or would it\n  only have lost an argument?* False: a part. Lost an argument: a reason (§ 3). *The line*, below,\n  means the boundary between the two.\n- **One device, two numbers, and the notes had already ruled it.** What a sentence *gives* the claim\n  if it holds is the link's relevance. How much the claim *needs* it is a requirement weight. **A\n  part is presumed needed** (half, and fully when tagged as required). **A reason is presumed not\n  needed, until somebody argues that it is.** That is the ruling of 30 September (ledger Q34), and\n  it is the corpus's own answer to the question (§ 2b). This note adds one check to it: the cap's\n  formula is the exact ceiling in probability's terms (§ 4).\n- **So the difference is a presumption and not a wall, and his challenge finds where it ends.** If\n  every argument for a claim has failed, and somebody can say they were all there could be, the\n  claim falls as it would with a part. Today's code cannot say that: it shows such a claim as\n  *unexamined* (§ 5a).\n- **A reason is a part too: of an argument, and not of the claim.** Three wholes sit inside each\n  other: what the claim says, one argument for it, and the whole case for it. The same question is\n  asked at each, and the answer names the whole (§ 5).\n- **From the other end they are one relation.** *P is a part of W* and *W is a conclusive reason for\n  P* say the same thing. The linking step stored one as the other for six months (§ 6).\n- **The line moves.** A reason can be written into a claim as a condition, and the test for a word\n  can be taken as its meaning or as a sign of it. So a filing is a judgement somebody can argue with,\n  and the two wrong filings are two attacks the notes already describe (§ 8).\n- **The notes had met this line in twelve places under different names** (§ 2), and how each solved\n  its case and how they fit together are in § 2a and § 2b. Fourteen names for the line, and what\n  each adds, are in § 10.\n- **For the splitter**: its gate asks whether the claim would be *weaker*, and a reason passes that.\n  *(Narrowed the same evening: a reason the sentence itself states passes. A reason the claim does not\n  state did not, on 30 of 30 parts people had attached:\n  [checks-on-machines-and-people.md](checks-on-machines-and-people.md) § 11.2.)*\n  T9 stands and is sharpened in four ways. One of them corrects its headline: the reasons in an\n  arguing sentence are parts of the sentence and reasons for its conclusion, both at once (§ 11).\n\n## 1. What set it off\n\nA made-up sentence, *the room should be warmer, because it feels cold and comfort matters more than\nthe bill*, had been called three pieces. He answered: *\"Three pieces? But it's arguing for a\nconclusion for two reasons.\"* The splitter, asked to propose a cut of a real sentence of that shape,\nreturned the conclusion, the two reasons and the whole sentence again, all as parts\n([microtasks-in-broad-daylight.md](microtasks-in-broad-daylight.md) § 3h). A part caps its whole.\nA reason that fails drops out. So the filing decides what a weak reason does to the claim.\n\n## 2. Twelve places where the notes met this line\n\nFound by asking the index of notes the question in eight wordings, then reading each home.\n\n| When | The line, as it was named | What was found | Home |\n|---|---|---|---|\n| Spring 2026 | a mother claim and its subclaims | His sentence used both words: the mother claim is *\"supported by … all the subclaims that it can and should be decomposed into\"*. And a scheme's critical questions are premises. | [self-similar-decomposition-and-claim-ontology.md](self-similar-decomposition-and-claim-ontology.md); [scheme-bounded-decomposition-and-evidence-as-subgraph.md](scheme-bounded-decomposition-and-evidence-as-subgraph.md) |\n| 17 August | jointly required against corroborating | The literature has two readings of support that are duals: *necessary* (accepting the whole means accepting this) and *deductive* (accepting this means accepting that). They differ in what attacking one does to the other. | [support-semantics-and-the-conjunction-problem.md](support-semantics-and-the-conjunction-problem.md) |\n| 31 August | parts against evidence | A real part is required by construction. A child that looks like independent support is usually evidence with the wrong label: *three studies are witnesses, not parts*. | [what-the-necessary-default-changes.md](what-the-necessary-default-changes.md) § 6 |\n| 17 to 19 September | how much the whole needs the part | A part caps by how strongly it is required, and that can be argued in a claim of its own. Whether the cap can be abused is open. | [weakest-link-arithmetic-and-the-merge-hunch.md](weakest-link-arithmetic-and-the-merge-hunch.md) § 9d, § 9e |\n| 21 September | parts against premises | A cut through a concept gives premises: the pieces do not give the claim back. Dummett's *analysis* against *decomposition*. The field has no settled test. | [core-periphery-and-the-concept-layer.md](core-periphery-and-the-concept-layer.md) § 5a-ii; [a-cut-through-a-concept-is-an-argument.md](a-cut-through-a-concept-is-an-argument.md) |\n| 22 September | the same, counted | Of the part links stored the right way round, 59% were read as not parts: the splitter's 15 of 44, extraction's 22 of 30, people's 30 of 39. | [the-parts-review.md](the-parts-review.md) § 4b; [linked-convergent-and-what-we-are-actually-asking.md](linked-convergent-and-what-we-are-actually-asking.md) |\n| 28 September | sufficiency and necessity | A link's relevance is sufficiency plus necessity minus one. *How much the whole needs the part* is the necessity term. | [strength-calculus-literature/bayesian-and-informal-logic.md](strength-calculus-literature/bayesian-and-informal-logic.md) § 2.1 |\n| 29 to 30 September | when a reason fails | **The place that had already joined them.** A reason that fails drops out, and failure counts against a claim only through a contribution that says why: an objection, a requirement claim, or a thorough search that found nothing. The requirement claim was built for parts (presumed weight 0.5) and extended to reasons (presumed 0). Relevance is the gain on the link, and the loss sits in the requirement. | [relevance-and-the-two-factors.md](relevance-and-the-two-factors.md) § 7; [session41-when-a-reason-fails.md](session41-when-a-reason-fails.md); [starting-values-and-the-problem-of-the-prior.md](starting-values-and-the-problem-of-the-prior.md) § 6.2, § 6.8 |\n| 7 October | a premise stored as a part | The linking step had written *this is a premise of that* as a part link, premise first, 261 times. It can no longer write part links. | [auto-connect-upgrade.md](auto-connect-upgrade.md) § 2c |\n| 7 to 8 October | premise against licence | The rule under a step is a licence and not one more premise. And one question is asked at three places: the parts of a claim, several reasons for it, the premises inside one reason. | [philosophy-of-the-unwritten-step.md](philosophy-of-the-unwritten-step.md) § 2, § 3; [the-squirrel-and-the-strength-calculus.md](the-squirrel-and-the-strength-calculus.md) § 2 |\n| 11 to 12 October | relations by content against relations by the world; a step written into the claim | Lakatos: a reason can become a condition of the claim, and his test tells them apart. Proposal: mark every kind of link as one or the other. | [twelve-books-read-against-deliberus.md](twelve-books-read-against-deliberus.md) § 3; [the-tractatus-and-deliberus.md](the-tractatus-and-deliberus.md) § 5, § 7, § 9 |\n| 12 October | reasons inside one sentence | The splitter files them as parts. | [microtasks-in-broad-daylight.md](microtasks-in-broad-daylight.md) § 3h |\n\nMost of these solved their own case. One, the ruling of 30 September, had already put parts and\nreasons on one device, and the first version of this note cited it by number without reading it.\nNobody had set all twelve side by side.\n\n### 2a. How each place solved its case\n\n1. **Spring.** Subclaims hang under a mother claim on a part link, and their strength flows up to it.\n2. **17 August.** One word on the part link says which reading holds: *necessary* (the weakest part\n   governs, with a ceiling), *corroborative* (contributions add), *alternative* (the strongest\n   decides).\n3. **31 August.** The default for an untagged part was flipped to *necessary* (ledger B10), and a\n   pass was proposed that sorts each existing part three ways: required, independent support, or\n   evidence with the wrong label.\n4. **17 to 19 September.** A weight on each part says how strongly the whole needs it: presumed 0.5\n   for a part a person or the source wrote, 0 for a machine's unconfirmed guess, 1 when tagged.\n   Anyone can argue it in a requirement claim, and the cap follows that claim's standing.\n5. **21 September.** A cut that fails to give the claim back is to be filed as support, on the\n   scheme for arguing from a classification, and the test it sets for the word goes to the concept\n   layer. Proposed; nothing acts on the check yet.\n6. **22 September.** A pass read all 197 part links and proposed a kind for each. A review page puts\n   the same question to a person in plain words. Nothing was written from either. On 27 September\n   the 29 miscast links the linking step had written were retired.\n7. **28 September.** A link's effect became the supporter's own standing times the link's relevance.\n   The standing questions on a link were sorted into those about the premise and those about the\n   connection.\n8. **29 to 30 September.** The row in the table above. Ruled for now; its four pieces are unbuilt.\n9. **7 October.** The linking step lost the ability to write part links. Its 261 were retired and\n   read again as premise links.\n10. **7 to 8 October.** What a step rests on (a fact, the names it joins, the rule) goes under the\n    link, as the grounds of the link's relevance claim, filed as required parts of it. Proposed. Since\n    10 October a machine's sentence under a link is stored there as a claim (ruled).\n11. **11 to 12 October.** Proposed: every kind of link is marked as holding by content or by the\n    world, and no pass answers one kind of question with the other kind of link.\n12. **12 October.** Ledger T9, and this note.\n\n### 2b. How they fit together\n\nSide by side, the twelve describe one arrangement. Anything attached to a claim has two numbers, and\nthe kind of link sets what is presumed about each until somebody argues otherwise.\n\n| Attached as | What it gives the claim | How much the claim needs it | When it falls | State |\n|---|---|---|---|---|\n| a part of what the claim says | nothing alone; all the parts together give the claim back | presumed half; all of it when tagged; nothing for a machine's unconfirmed guess; argued in a requirement claim | the claim is capped by it | ruled and built |\n| a reason for the claim | its own standing times the link's relevance (presumed half from the source or a person, nothing from a machine's guess between other people's claims; argued in a relevance claim) | presumed nothing; argued in a requirement claim | it drops out, unless a requirement was argued | the gain is ruled and live; the need on a reason is ruled for now and unbuilt |\n| a piece inside one reason: the fact, the joined names, the rule | no number of its own | all of it, for that reason | that reason is gone | proposed |\n| all the reasons together | no number of its own | all of it, **if** they are all there could be | *nothing stands*; leaning against the claim takes an argued thorough search | ruled for now and unbuilt |\n\n**What was still missing after 30 September**: the third row, and a splitter whose gate had not\ncaught up with the arrangement (§ 3).\n\n## 3. The test, and why the splitter's gate cannot apply it\n\n**Three wordings of one test are already in use.**\n\n- The review page's answers: *a piece: take it away and the claim fails*, against *not a piece: it\n  is evidence about the claim*.\n- Lakatos's: *if this sentence falls, is the claim false, or is only this argument for it gone?*\n- The rebuilding check: can the claim be written back from the pieces alone?\n\n**The splitter's gate asks something wider.** Since version 2 it keeps a part only if the answer to\n*if this part were false, would the claim be weaker?* is yes\n(`deliberus/extraction/decompose_claim.py`). That question was written to catch what a claim denies\nor only grants, and it does. But a claim whose reason fails is weaker too. *Weaker* holds two\nanswers: *the claim would be false*, and *the claim would have lost support*. Read from the prompt;\nit agrees with the one run on record, where both reasons came back as parts.\n\n*Narrowed the same evening, by a run.* Put to the 39 parts people have attached to claims in the live\ngraph, the gate answered *weaker* for the 9 their claim asserts and *unaffected* or *cannot tell* for\nthe 30 it does not say, 24 of which are reasons for the claim. So the reader takes *weaker* more\nnarrowly than the prompt's words allow: a reason passes when the sentence itself states it, which is\nthe case of the next paragraph and of T9, and not when the claim leaves it unsaid. One model, one run\n([checks-on-machines-and-people.md](checks-on-machines-and-people.md) § 11.2).\n\n**The test needs to know what it is asked about.** Take *C because R*. Asked about the whole\nsentence, R is a part: the author said R, and if R is false the sentence said something false. Asked\nabout C, R is a reason. So the splitter has to find the conclusion before the test means anything.\nThat is the first question T9 proposed.\n\n## 4. One dial: what is left of the claim when this falls\n\nTake a claim C and a sentence X whose own standing is q. Two numbers say how X bears on C. Both are\nin the September review: **sufficiency**, how surely C holds when X holds, and **necessity**, how\nsurely C fails when X fails. Then, in probability's terms:\n\n    P(C) = sufficiency · q + (1 − necessity) · (1 − q)\n\n**The second term is what is left of C when X fails.** With necessity 1 nothing is left: X is a\npart, or something the claim needs in fact. With necessity below 1 something is left: X is a reason.\n\nTwo bounds follow, whatever else is true of C:\n\n    a floor:    P(C) ≥ sufficiency · q             a reason gives a floor\n    a ceiling:  P(C) ≤ 1 − necessity · (1 − q)     a part gives a ceiling\n\n**The ceiling is the shipped cap.** The cap a required part puts on its whole is\n`1 − w·(1 − strength)`, where w is how strongly the part is required\n(`deliberus/support_semantics.py`). That is the ceiling above, term for term, with w read as\nnecessity. And the link rule ruled on 28 September moves a claim by relevance times the supporter's\nstanding above neutral, where relevance, on the reading the September review recommends, is\nsufficiency plus necessity minus one.\n\nSo the cap and the link rule are one formula read at its two ends. **The ruling of 30 September had\nsaid this in words**: relevance in probability is sufficiency plus necessity minus one, *\"and the\nontology splits it into the gain on the link and the loss in a requirement\"*\n([relevance-and-the-two-factors.md](relevance-and-the-two-factors.md) § 7). Under that ruling a\nlink's relevance means the gain alone, and the need is kept apart, in the requirement. What this\nnote adds is the check that the cap as built is the exact ceiling: three lines of algebra, and\n20,000 random cases ([the script](a-part-and-a-reason/check_the_identity.py)). *(The first version of\nthis section said no note had drawn the connection. It had searched for the cap's name in the\ncode, and the ruling names the weight.)*\n\n**What follows.**\n\n1. *Part* is not a kind of link. It is the corner where necessity is 1 by what the two sentences\n   mean. A link in the middle of the dial is a reason the claim leans on heavily.\n2. The two claims that can be argued about a link are about its two numbers. The requirement claim\n   (*W needs X, because …*) argues the need. The relevance claim argues the gain.\n3. A part link has no account of sufficiency: how far established parts lift their whole is the\n   open question the August note left (three parts at 0.9 lift a whole to 0.507). The formula says\n   why it must wait on coverage: the parts give the claim back only when they are all of them.\n4. A support link's need is presumed nothing, and mostly that is right. Where somebody says *this\n   theory stands only if that experiment comes out*, a reason carries a requirement. The ruling of\n   30 September gives it one, argued as a part's is; that half is unbuilt.\n\n**A limit found from outside, 2026-10-14** (`claim-parts-and-inferential-reasons-2026-10-14.md` § 4): the\nfalse-without-it test measures *need* and not *content*. Any claim needs, strictly, itself-or-anything (A\nentails A ∨ Z), and a machine that cannot run without fuel does not *say* fuel. So the test is the right\ndial for the cap, which runs on need, and the wrong gate for the lift, which needs the parts to be all of\nwhat the claim says: a content question (Yablo's aboutness; the rebuild check is the nearest instrument).\nThe bicycle example used in chat that day named parts of a bicycle, not parts of a sentence about it.\n\n**Limits.** Strengths here are not probabilities: 0.5 means undecided, and the calculus is an energy\nand not a product of chances. The formula is for one X at a time, and where parts are needed\ntogether no part has a sufficiency of its own (the September review's *not separable*). What a\nrefuted reason should take away is ruled, for now, as *nothing, unless a contribution says why*\n(ledger Q34), and the path through honest starting values is open (Q35, Q36): § 5a.\n\n## 5. Three wholes\n\nThe squirrel note found one question asked at three places. Each place is a whole, and the question\nat each is the same: *what falls with this?*\n\n| The whole | Its parts | When one falls | Today |\n|---|---|---|---|\n| **What the claim says** | what it states outright, what it takes for granted, what follows from it by meaning | the claim is false | the weakest part governs, and a required part caps (ruled) |\n| **One argument for the claim** | a fact about the case, the names it joins, the rule that lets the fact count | that argument is gone, and the claim may stand on another | no place until a step is written out; proposed as required parts under the link |\n| **The case for the claim** | its arguments | the others remain | links add, and nothing discounts arguments that share an origin (open); when all of them fall, § 5a |\n\n**So a reason is a part too.** It is a part of an argument, or of the case, and not of the claim.\n*Reason* names two of these: a whole argument, and one premise inside it. That is why *part against\nreason* never had a clean answer: the two words are not on the same level. The useful question is\n*a part of which whole?*\n\n**Both of his descriptions were right.** *The room should be warmer, because it feels cold and\ncomfort matters more than the bill* is three pieces of a sentence and two reasons for a conclusion.\nA sentence that argues is an argument: a small map stored in one box. Its author is bound to all of\nit, the conclusion, each reason and the step between them.\n\n**The weakest-part rule was right about the sentence all along.** The August ruling took that rule\nfrom the premises of one argument, where it is old (Pollock's weakest link), and applied it to the\nparts of a claim. An arguing sentence is the case the rule came from: the author's argument, as\ngiven, is as good as its weakest piece. The rule goes wrong only when the sentence stands in the\nconclusion's place, because other people argue the same conclusion on other grounds.\n\n**One consequence for reuse.** Two authors who reach one conclusion for different reasons share a\nclaim and differ in their arguments. Stored as two arguing sentences they look like two claims that\noverlap.\n\n## 5a. When every argument falls\n\n**His challenge**, on Lakatos's test: *\"What's the real distinction here? The implication is that\nthere could be other arguments for the same claim, such that it could be evaluated as a strong\nclaim, still, if we came up with such an argument? And if it was deemed relevant. An\nargument/subgraph does bear on the mother claim's strength/estimated truth, right? Because it is\nimbued with the truth-value/strength of the claims that support it and are deemed relevant (read up\ndeeply on relevance). If all conceivable arguments that are relevant falter/are refuted, then the\ncentral/mother claim must have a very low truth-value/strength by extension, that is, \"be false\" or\nas close to false as modest uncertainty let's us say inside Deliberus. Right? No?\"*\n\n**He had said it before.** On 29 September, asked whether a refuted reason should pull its claim\ndown: *\"if the implication is that there's SOME truth value to a claim if it's in the middle then\nthat'd be misguiding if there's no good and relevant reasons whatsoever for believing it's true.\nBut ofc all of this depends on whether it's been examined deeply and exhaustively and it all depends\non how you see/judge \\*relevance\\*. Right?\"*\n\n**Yes to each step, and the last step needs two things said out loud.**\n\n1. *An argument bears on its claim by the standing of what supports it, where that is relevant.*\n   Yes. That is the link rule, live since 30 September: the supporter's own standing times the\n   link's relevance.\n2. *Lakatos's test implies there could be other arguments.* Yes, and that is all it implies. *Only\n   this argument is gone* quietly takes it that others may exist. For a part no such thing is taken:\n   the claim said it. So the distinction is in what is presumed when the sentence falls. A part is\n   presumed needed. A reason is presumed to be one of several.\n3. *If every relevant argument fails, the claim should read very low.* Yes, given two premises, and\n   the first is his own from 29 September.\n   - **That these were all there could be.** The graph holds the arguments somebody offered. That\n     none is missing is a claim, and a machine cannot make it: what nobody has contributed is the\n     one thing no reading of the graph sees. A person can make it, as an argument from a thorough\n     search that found nothing. The scheme is in the catalogue, and its question is *was the search\n     thorough?*\n   - **That the claim would have a good argument if it were true.** For *there is a draught in this\n     room* it would. For *the number of stars is even* it would not, and that claim stays unknown\n     however many arguments for it fail. The catalogue's question for this is *is the absence\n     meaningful?*\n\n**Where both hold, the set of all reasons behaves as a part does**: the claim needs at least one of\nthem, none stands, and the claim falls. That is where the two sides of the line meet. In § 4's\nterms the need of the whole set is 1, though the need of each reason alone was small.\n\n**Why the ruling does not do it automatically.** Anyone can attach a poor reason to a claim they\noppose and refute it. If each such failure pulled the claim down, a straw man would pay. So the\nruling of 30 September lets a failed reason drop out, and lets failure count against a claim only\nthrough a contribution that says why: a finding filed as an objection, a need argued as a\nrequirement, or the thorough search above. With it came a third state for a badge, *nothing stands\n(N checked)*, *\"so the middle never implies some truth where every checked reason failed\"*. He ruled\nit *\"for now\"*, and questioned its basis the same hour (ledger Q35).\n\n**What the code does today** ([the script](a-part-and-a-reason/when_every_reason_falls.py), fed by\nhand, with no graph and no model):\n\n| The claim has | Strength | Shown as |\n|---|---|---|\n| nothing attached | 0.500 | no data |\n| three reasons, each standing | 0.632 | supported |\n| three reasons, two refuted and one standing | 0.519 | unsettled |\n| **three reasons, every one refuted** | **none** | **unexamined** |\n| the same, and an objection that stands | 0.481 | unsettled |\n| three parts, one refuted, presumed half needed | 0.493 | unsettled |\n| three parts, one refuted, tagged as required | 0.100 | unsupported |\n\nSo the case he describes reads today as *unexamined*, the same as a claim nobody has looked at. The\nrepair is in the ruling and is unbuilt: a refuted reason counts as examined, and the badge says\n*nothing stands (3 checked)*. And even with the thorough search argued and standing, one objection\nmoves the claim only to 0.481. The calculus is nearly flat close to the middle, which the September\nreview flagged and nobody has ruled on.\n\n**Low standing is not falsehood.** *Nothing stands* says nobody is entitled to lean on the claim.\n*False* says its denial stands, and the denial needs a case of its own. Where absence is meaningful,\nthe thorough search is that case.\n\n**The author's own reason is the case nearest to a part.** A reason attached from outside may have\nbeen chosen because it fails. A reason the claim's own author gave was relied on. Whether its\nfailure may then count against the conclusion is open (ledger Q36), and it bears on § 11: the reasons\nin an arguing sentence are the author's own.\n\n## 6. The same relation from the other end\n\n**Duals.** *X is necessary for W* says what *W is sufficient for X* says. So *P is a part of W* and\n*W is a conclusive reason for P* are one fact: a whole gives each of its parts for free. The linking\nstep's backwards links were this fact written down from the wrong end.\n\n**One scale.** The Tractatus reading set the seven answers of the sameness check beside that book's\nrelations of containment between sentences. Laid in a row they are one scale: *same* (each gives the\nother), *contains* (one gives the other outright), *overlaps* (one makes the other likelier),\n*unrelated*, *opposes*. Part and reason are two stretches of it:\nhow much of one sentence is already in the other. A step down to a part adds nothing and risks\nnothing. A step from a reason to a conclusion adds something and can fail. Peirce's words for the\ntwo are *explicative* and *ampliative* *(recalled)*. Mill's old charge against the syllogism, that\nits conclusion is already in its premises *(recalled)*, is the limit: where a reason is conclusive,\nthe conclusion is a part of it.\n\n**Two lines, four corners.** The proposal to mark each kind of link as holding by content or by the\nworld (ledger AJ3) draws a real line, but it is not the line between part and reason. A conclusive\nreason holds by content. A need in fact holds by the world.\n\n| | holds by what the sentences mean | holds because of how the world is |\n|---|---|---|\n| **the claim needs it** | a part: stated, taken for granted, or following by meaning | a need in fact: *the engine runs* needs fuel. It has to be argued, which is what the requirement claim is for. |\n| **it gives the claim** | all the parts together; a conclusive set of premises | a reason: evidence, a sign, somebody's word |\n\n**A part proper is the one place where both directions hold by meaning.** The claim gives each\npart, and the parts together give the claim back. Those are the two tests the notes already use:\ntake it away, and rebuild.\n\n## 7. What can stand beside a claim\n\n| What it is | Example | If it falls | Where it goes |\n|---|---|---|---|\n| stated outright | *the room is cold and damp* says *the room is damp* | the claim is false | a part |\n| taken for granted | *she has stopped smoking* takes it that *she smoked* | the claim is false, or about nothing | a part; whether it should cap is open |\n| following by meaning | *he is a bachelor* gives *he is unmarried* | the claim is false | a part, seldom worth writing |\n| needed in fact | *the engine runs* needs *there is fuel* | the claim is false, if the need is real | a part whose requirement is argued |\n| a test the word is taken to set | *she is an expert*, so *she has published in the field* | false for a reader who holds that test; for another, one sign is gone | the concept layer, and an argument from classification |\n| a fact in a step | *the squirrel smells* | this argument is gone | the supporter on a link |\n| the rule of a step | *dead squirrels smell* | this argument is gone | under the link |\n| a separate reason | a second study; a second witness | it drops out and the rest stand | a link of its own |\n| a cause | *the bridge fell because the bolts had rusted* | the explanation is false; the bridge still fell | a claim of its own (*the rust is why*), and no support link |\n| a purpose | *we should heat the room, because comfort matters more than the bill* | this argument is gone | a step from a goal, with the weighing as its rule |\n\n**The word *because* covers four of these rows.** A reason to believe, a cause, a test the word\nsets, and a purpose are different things said with one word. Informal logic separates the first two\nas argument and explanation: an explanation takes what it explains as settled and says why it is\nso, and the same two sentences can be read either way depending on who doubts what (Mayes 2010,\nlooked up). The older names are the reason of knowing and the reason of being *(recalled)*.\n\n## 8. Where the line moves\n\n**A reason becomes a part.** In Lakatos's account of how a theorem is repaired, a step of the proof\nthat a counterexample breaks is written into the theorem as a condition. A sentence that was a\nreason for the claim is now a part of it. The claim becomes safer and says less.\n\n**A sign becomes a test, and back.** Wittgenstein separates the *criterion* that defines a word\nfrom a *symptom* that experience ties to it: the bacillus in the blood against the inflamed throat.\nHe adds that people use such words without ever settling which is which, and that asked to say, one\ncould mostly answer only by *\"an arbitrary decision ad hoc\"* (the Blue Book, looked up). For a\ncontested word, then, one reader's part is another reader's reason. This is the corpus's core and\nperiphery again, and it is why a cut through a concept is filed as an argument: filed as a part, a\ndefinition has been written into the graph where nothing can attack it.\n\n**So a filing is a move, and each wrong filing is an attack already on record.**\n\n- **A reason filed as a part** makes the claim hostage to something it does not need. That is the\n  bogus requirement of the weakest-link note (§ 9d, § 9e): add a part nobody has established, and\n  the whole is held down.\n- **A part filed as a reason** lets a claim outlive something it said. That is the retreat in the\n  motte note: when the strong sense falls, *that was only one argument for it*. In the graph's terms\n  a retreat re-files a part as a reason after it has fallen. The motte note's four tests tell a\n  refinement from a retreat; this says what the retreat does to the structure.\n\nEach has an honest form, and the graph has a place for both: narrowing a claim in the open, and\narguing a requirement in the open.\n\n## 9. The house example, read again\n\n*Dr Smith is a credible expert in labour economics* has been the corpus's example of parts that are\nall required: her publications, her post, her field. Put the test to each.\n\n- *She has published 47 papers.* If false, is she not credible? She could be, on other grounds. A\n  sign.\n- *She holds a chair.* The same. A sign.\n- *Her field is labour economics.* If false, the claim is false. A part.\n\nSo the standing example of required parts holds one part and several signs. A reader for whom\nexpertise *means* a record of publications would call them parts, which is § 8's point.\n\n**Where they came from explains it.** They are the critical questions of the argument from expert\nopinion, and that note's heading is *Critical Questions ARE Premises*. They are the parts of an\nargument: the one that runs from her say-so to what she says. *All required* is true of them\nthere. If she is not credible, the argument from her opinion is gone, and what she vouched for may\nstill stand. They sit in the middle row of § 5 and were given the first row's rule.\n\nThe count agrees. Of the 39 part links people wrote, 30 were read as not parts.\n\n*This is a session's reading. It does not undo the pattern of a mother claim with its case beneath\nit. It says which arithmetic fits which child.*\n\n## 10. Other names for the line\n\n| The pair | From | In the notes | What it adds |\n|---|---|---|---|\n| constituent and premise | Dummett on Frege | yes | A thought has one analysis into constituents and many decompositions. |\n| necessary and deductive support | Nouioua and Risch; Boella | yes | Duals. They differ in what attacking one does to the other. |\n| linked and convergent | Beardsley, Thomas, Freeman, Walton | yes | Forty years, a dozen rival tests, and one respected case for dropping the distinction. |\n| by content and by the world | the Tractatus | yes | Which pass may answer which question. |\n| premise and licence | Carroll, Ryle, Toulmin | yes | The rule cannot be one more premise. |\n| a local and a global counterexample | Lakatos | yes | The test itself. |\n| sufficiency and necessity | Hunter; the form of Youden's index | yes | Two numbers on one link. |\n| core and periphery | the corpus | yes | Whose part it is depends on how the word is taken. |\n| criterion and symptom | Wittgenstein | **no**, looked up | The two trade places, and usage never settles which is which. |\n| a content part and a mere consequence | Yablo; Gemes; Fine | **no**, looked up | A part is implied *and* raises no new subject: *p or q* follows from *p* and is no part of it. This is the structural signal the parts review uses (same subject, a fragment of what is said) and the splitter's rule that a part adds nothing. |\n| argument and explanation | informal logic | **no**, looked up | One word, *because*; not a relation between two sentences alone. |\n| commitment and entitlement | Brandom | named, the pair is not *(recalled)* | Downstream and upstream: what asserting a claim binds you to, and what earns the right to assert it. |\n| explicative and ampliative | Peirce; Kant's analytic and synthetic | one mention *(recalled)* | A part adds nothing and risks nothing. A reason adds and can fail. |\n| what may be drawn from it and what it may be asserted on | Gentzen; Dummett's harmony | **no** *(recalled)* | For a plain *and*, the two are the same sentences. That is the one case where parts and reasons coincide. |\n\nOne more, at a distance. Gemes built his account of content parts to stop an irrelevant addition\nriding along when a hypothesis is confirmed, the *tacking* problem that the evidence dossier names\n*(the link to Gemes is recalled)*. The padding ruling met the same shape from the strength side.\n\n## 11. What it gives back for the splitter\n\nRefinements of ledger T9. Unruled, and no new row.\n\n1. **Find the conclusion first.** T9's first question stands, and § 3 says why it is needed.\n2. **Split the gate's *weaker* in two**: *the claim would be false*, and *the claim would have lost\n   support*. Only the first makes a part of the conclusion.\n3. **T9's headline was too wide.** It said the reasons are filed *never as parts*. They are parts of\n   the sentence and reasons for its conclusion, both. The cut the splitter made was right as far as\n   it went. What it lacked was the second filing, the relations among the pieces: which piece is the\n   conclusion, a support link from each reason to it, and the piece that repeated the whole filed as\n   what it was, the step between them, under that link. The sentence stays, as its author's\n   argument, and the weakest-part rule on it is then correct. The conclusion becomes the claim that\n   is argued about.\n4. **Two reasons or one.** Ask of each: *if the other fell, would this one still count for the\n   conclusion?* Yes: two links. No: one link, with the other sentence under it. Readers will differ\n   here, and that is the expected case ([a-cut-through-a-concept-is-an-argument.md](a-cut-through-a-concept-is-an-argument.md) § 5).\n5. **A *because* that explains** is written as a claim of its own and gets no support link. The test\n   is whether what is explained is treated as already settled.\n\n6. **The author's own *because*.** The links such a cut writes run from reasons the author gave,\n   the one case where reliance is known from the text. If ledger Q36 is answered yes, these are the\n   links on which a failed reason may count against the conclusion. Until then a conclusion whose\n   only stated reasons fail would read *nothing stands*, once that state is built.\n\nNot proposed for building: the two-number reading of § 4. It is a way to see three open rows as one\n(the requirement weight, the lift from established parts, and the marking of links as by content or\nby the world).\n\n## 12. What this does not settle\n\n**Next examination (2026-10-16):** the founder reaffirms separating inline reasons and asks how\nconcept senses and duplicate challenge routes interact. The [Lins worked cases](one-challenge-one-home.md)\nkeep the utterance, conclusion and reasons distinct, compare three representations, and test premise\nchallenges against challenges to the inference. The representations remain alternatives.\n\n- **Strengths are not probabilities.** § 4 reads the rules in probability's terms and proves nothing\n  about the calculus as built.\n- **The field has no agreed test** for premises that work together against premises that stand\n  alone. Lakatos's question is one criterion.\n- **Evidence bears on webs of claims, not on one claim.** *What is left when this falls* holds the\n  rest still, and the rest does not hold still\n  ([evidence-division-and-the-foundation.md](evidence-division-and-the-foundation.md) § 1).\n- **A reason in one case can be none in another** (Dancy), so sufficiency does not travel with a\n  sentence.\n- **What a claim is worth before anyone examines it** (ledger Q35) is the question under § 5a, and\n  it is open.\n- **No person has read a cut of an arguing sentence.** There is one machine run of one sentence.\n\n*Settled further 2026-10-15: four parts rulings had a reason as their example, and on the reasons channel they read as weakest premise within a line of reasoning (his 2013 argument bundle), best line across lines; the three tag words get exact jobs on support links, and 24 reasons people filed as parts sit under min today. [retrofitting-the-reasons-channel.md](retrofitting-the-reasons-channel.md) § 1; ledger Q38.*\n\n## 13. What was read, and how far\n\nThe index of notes was asked in eight wordings (part against reason; required against\ncorroborating; part against evidence; a cut through a concept; necessary against evidential\nsupport; argument against explanation; what a claim asserts against what makes it believable;\nrelations by content against relations by the world). Read whole: the notes on support semantics,\non linked and convergent structure, on evidence division, on a cut through a concept, and the parts\nreview. Read by section: the weakest-link note (§ 1d, § 1e, § 8, § 9, § 9d, § 9e), the note on the\nnecessary default (§ 6 to § 9), the unwritten step (§ 0 to § 3), the squirrel (§ 0 to § 5), the\nSeptember review (§ 0 to § 2.2), the Tractatus reading (§ 5, § 7 to § 9), the twelve books (§ 3),\nthe decomposition axes (§ 3 to § 5 and the parked section on separability), the concept layer\n(§ 5a, § 5a-ii), the microtasks note (§ 3e, § 3h), and the splitter's prompts.\n\nRead after his challenge, and owed before it: the note on relevance and a link's two factors (§ 0\nto § 3, § 5 to § 7), the session note on what a failed reason does (§ 4, § 9), and the note on\nstarting values (§ 0, § 6 to § 9).\n\nLooked up, as search extracts and not in full: Wittgenstein's Blue Book on criteria and symptoms;\nYablo's *Aboutness* and his 2014 paper on content parts; Mayes, *Argument-Explanation\nComplementarity* (Informal Logic, 2010), and a 2025 paper on the same distinction. Items marked\n*(recalled)* are from memory and are not to be cited outward before somebody opens them.\n\n## Cross-references\n\n[microtasks-in-broad-daylight.md](microtasks-in-broad-daylight.md) § 3h (the splitter and ledger T9) ·\n[relevance-and-the-two-factors.md](relevance-and-the-two-factors.md) § 7 (what a failed reason does;\nledger Q34 to Q36) ·\n[starting-values-and-the-problem-of-the-prior.md](starting-values-and-the-problem-of-the-prior.md)\n§ 6.2, § 6.8 ·\n[the-squirrel-and-the-strength-calculus.md](the-squirrel-and-the-strength-calculus.md) § 2 (one\nquestion at three places) ·\n[weakest-link-arithmetic-and-the-merge-hunch.md](weakest-link-arithmetic-and-the-merge-hunch.md)\n§ 9d, § 9e (the cap, and the bogus requirement) ·\n[strength-calculus-literature/bayesian-and-informal-logic.md](strength-calculus-literature/bayesian-and-informal-logic.md)\n§ 2.1 (the identity) ·\n[support-semantics-and-the-conjunction-problem.md](support-semantics-and-the-conjunction-problem.md)\n(the duals) · [the-tractatus-and-deliberus.md](the-tractatus-and-deliberus.md) § 5, § 7 (ledger\nAJ3) · [twelve-books-read-against-deliberus.md](twelve-books-read-against-deliberus.md) § 3\n(Lakatos) · [the-motte-and-the-periphery.md](the-motte-and-the-periphery.md) (the retreat) ·\n[core-periphery-and-the-concept-layer.md](core-periphery-and-the-concept-layer.md) § 5a-ii ·\n[a-cut-through-a-concept-is-an-argument.md](a-cut-through-a-concept-is-an-argument.md) ·\n[the-parts-review.md](the-parts-review.md) ·\n[the-chain-inside-a-link.md](the-chain-inside-a-link.md) § 10 (2026-10-13: the floor for a step\ninside a link is of the same kind as the floor for splitting, in his words; this note's test says\nwhich of the two directions a cut runs in)\n"}