Meanings and articles (draft)

A design note, with a working draft. Nothing here changes what any argument proves. The modules it describes sit outside the arguments, and no argument's verdict reads them.

The library checks arguments it has encoded itself. This note is about checking an argument someone else has published — an article, a tract, a chapter that claims a position is unbiblical or fallacious — and saying, as the library's own commentary, whether its claim can be defended. Such an article does not come as a premise list. It quotes, attributes, asserts and infers, and it can be careful in one paragraph and careless in the next. To check it fairly is to check each of those acts for what it is, and then to weigh what it claims against everything the library already holds.

That needs three things the library did not have: a way to say what a claim means, so that two claims can be compared; a way to model what an article does, move by move; and a way to hear an article's claim as a position in a dispute, so that whether it can be defended is computed, like every other verdict here.

Where the parts live

Nothing in the generic layer names an argument. Every argument has an instance: sola fide is analysed; sola scriptura, the virgin birth, Spirit baptism and the canonical witness in part, the claims whose meanings bear on a join — and, in the virgin birth, the claims about the words of Isaiah 7:14 and its versions, the woman it describes and the birth it is read to promise; and born in Bethlehem, looked at and found to have no joins, is registered with its citations' labels and marked unanalysed. Each is counted from the start (each module's coverage_now). Testimony.Checks.Meanings fails the build for an argument with no instance, so a new argument is registered when it is added. It also fails the build for an argument whose joins are not listed and pinned, or whose meaning postulates come from anywhere but those lists (computed ratings).

What a claim means

Every argument names its atomic claims in a type of its own. dikaioIsForensic is a name with a docstring; to the library it is opaque. So it cannot answer the questions a reader asks of the whole collection — every claim about δικαιόω, every passage read two ways, the same claim in two arguments — and it cannot see an article move from one passage to another under one word.

A meaning is a term built from the shared vocabulary. Its parts were each added because a claim in the library needed them:

Three distinctions in it do most of the work.

• The word and what it means are kept apart. A textual claim is stated with the words as they stand; what a word denotes is a separate claim, means. Romans 3:28 says a person is justified by faith apart from works of the law; what those words mean in Paul is three further claims, each disputed. This is Barr's distinction between the word and the concept, which the sola fide argument already turns on.
• says is not teaches. What a passage says, in its own words, is textual. What it teaches, read rightly, is a reading, which a rival may read otherwise. Luke 7:47 says her sins are forgiven, for she loved much; that her love is the evidence of forgiveness and not its ground is what the Reformed say it teaches. The two are never the same claim, as quotation is never prediction.
• A sense can be the same, or not, at two places. sameSense states that a word means the same in two texts. Every contradiction between two texts depends on it, for every word they share, and most arguments never say so.

The atoms are not replaced. An argument keeps its own claim type and its proofs are untouched; its instance says what each atom means, as cite says who holds it. Ninety-five of the hundred and seven sola fide atoms are analysed; twelve — the stories of the thief and the tax collector, a curse, a command, a disjunction, and the claims about texts and their translations — are marked unanalysed and counted, not forced.

What the meanings already make computable, from the meanings alone, for any argument:

• The contested readings are found, not listed. For sola fide: δικαιόω in Paul, the believing of John 6:29, doing the will in Matthew 7, keeping the commandments in Matthew 19, the water of John 3:5 four ways, the fire of Matthew 3:11 three ways (contested_length). Nothing else turns up.
• Every claim about a word. Thirteen sola fide claims turn on δικαιόω (about_dikaioo).
• The same claim in two places. An article's quotation of James 2:24 means exactly what the library's atom for the verse means (jamesSays_is_the_library's), so the two are one claim. Once a second argument is analysed, the same holds between arguments.

The ontology, derived in articles

Two claims join — one excludes or entails the other — when a reader who holds the one is bound to deny, or to hold, the other. That depends on what their words commit a reader to, and the words are the vocabulary's. "Faith is the only means of justification" and "baptism is a means of justification" contradict each other only because of what only means.

So a meaning is read for what its words mean (Testimony.Semantics.Ontology). It is taken apart into the facts it affirms and denies, and closed under definitions: what the vocabulary's own words mean, each with a source, as a premise has one. Doctrine — what is held to be so — is taken apart; what a passage says, what it teaches, what it is written against and what a voice held are kept whole, because a text may say a thing without saying all that follows from it.

The definitions are not a list kept beside the vocabulary. Each is derived in an article (Testimony.Ontology), which names no argument and does two things, both checked:

1. Define. What the words mean, with a source. A definition is what a word means, which no reader disputes; what one tradition holds of a word and another denies is a premise of a position, and the article leaves it apart and says why.
2. Show each definition at work on two statements of its own, built from the vocabulary alone: they exclude each other with the definition, and not without it.

What the definitions do in the arguments is computed, for every argument the library has, by Testimony.Checks.Ontology, from the generic queries of Testimony.Ontology.Search. It finds the arguments itself and pins, for each, how many of its joins are found with no definition at all, how many claims use each defined word as doctrine or of a text or voice, and the joins each definition contributes: those found with it and not with it taken out. It fails the build if any meaning contradicts itself.

The first article, Testimony.Ontology.Relations, writes out three definitions the vocabulary already gave in words: "not a ground" is the denial of "a ground"; the only means of a thing is a means of it, and nothing else is; the whole ground of a thing is a ground of it, and nothing else is. One join follows from them, in sola fide — faith alone against Trent's baptism as the instrumental cause of justification (the_only_means_at_work) — and one more was found by reading the meanings more carefully: "forensic only" against Trent's "justification includes sanctification", whose denial was always in the Reformed claim's own meaning (a_denial_inside_needs_no_definition). No dispute's table and no verdict moves; the named readings of sola fide take a side on both pairs.

Testimony.Ontology.Usage is about senses rather than relation words. A word's usage as a whole is the senses its passages give it, so a sense a word has in one passage of a text is a sense it has in that text's usage (a_sense_in_a_usage_at_work). That is all it does. It says nothing about any single passage, and it does not count a word used of a virgin as evidence that the word means "virgin". What a word means at a passage stays a premise, cited and rated, and the context it is read from is the premise's evidence. What it can show false is a claim that a usage lacked a sense, wherever a passage of it is held to have that sense. In the virgin birth argument it joins the laws of the Greek Pentateuch, and the status reading of Genesis 34:3, to the Septuagint's Greek: their sense "virgin" makes "virgin" one of the word's senses in the Septuagint, which is the claim encoded for Rico and Gentry, and excludes Wegner's "not yet 'virgin'". It says nothing of which sense the word has at Isaiah 7:14.

What the search found, not yet fixed

The search was run over every argument's meanings before any article was written. Besides the two joins, it found five things a stricter vocabulary has to answer, each the subject of an article to come.

• Paul's texts are stated in doctrinal words, James's "by works" is not. What Romans 3:28 says is encoded with "the means of" and "not a ground of"; what James 2:24 says, with saidBy, the word kept for a "by" that leaves open whether it is a ground, a means or the evidence. A text's own connective should be stated in a textual word for both, and what it denotes argued as a reading for both.
• One concept does two jobs. The vocabulary's baptism is water baptism. Spirit baptism's readings of 1 Corinthians 12:13 say what "the baptism" of that verse is, using the same concept, so that read literally they say what water baptism is. Whether Paul's word and Luke's name one baptism is the question the argument says is not yet encoded.
• isA does three jobs: a kind ("Scripture is a rule of faith"), a property ("Scripture is infallible"), and the identity of an event ("the baptism of 1 Corinthians 12:13 is Christ's baptizing in the Spirit"). Disjointness and subsumption cannot be stated over it until they are told apart.
• Faith alone is analysed two ways. Sola fide states it as "faith is the only means of justification"; the canonical witness, as faith necessary, faith sufficient and works not the ground. And the canonical witness's labels for the first two parts say saved, while its module, and its meanings, say justified.
• Senses and concepts overlap, and voices repeat the traditions. salvation and worksOfFaith are both a sense and a concept; the kinds of faith a word can mean — trust, faith formed by charity, bare assent — are senses of πίστις but not kinds of faith. And the vocabulary's voices name again what Tradition already names: the Reformed, Rome, Second Temple Judaism.

Testimony.Semantics.Discourse models an article as a list of moves. Each move records who makes it, the article's own words, verbatim, and the act they perform: an assertion, a quotation, an attribution, or an inference from premises to a conclusion.

The first check is the one that needs the grammar. An inference that sets one text against another — or a text against a doctrine — assumes, for every word they share, that it means the same in both. Move.unstated computes, from the move alone, the senses it needs and does not state; Move.meets finds the claims of any argument that deny them. A move with unstated senses is not thereby fallacious. It is conditional, on exactly those senses.

Commentary: whether an article's claim can be defended

The library's verdicts already say, of a position, whether it can be defended. A position is indefensible when no admissible set holds it: something defeats it, and nothing the dispute can hold answers that defeat. Trent's definition, read as a claim about what Paul's word means, is indefensible in the sola fide dispute (trent_on_pauls_word_indefensible). That is the library's own judgment, checked, and an article that argued the same would meet the same verdict, for the same stated reason.

So an article's claim is judged the same way, in four steps:

1. Model the move, verbatim.
2. Make its unstated assumptions explicit as premises, from Move.unstated. A position is answerable for what it needs, not only for what it says.
3. Hear it as a party in the argument's dispute, alongside every existing party.
4. Report what the solver computes: indefensible, defensible but not forced, or forced — whichever way it goes, and with the defeats that decide it.

This is commentary, not reproduction: the verdict is the library's, and it is a checked result about the article's claim, not a summary of the article.

Whether a dispute makes sense

A premise is rated disputed when a cited source grants its grounds and denies its conclusion. That is a fact about the literature: someone denies it. Whether the denial, stated as a position and weighed against everything else, can still be defended is a further fact, and Testimony.Logic.Contest makes it a computed one.

• A claim is contested in a dispute (Contested) when one party holds it, another denies it, and each can be defended. The rating disputed is then what the weighing finds.
• A claim's denial is answered (DenialAnswered) when every party that denies it is indefensible. The claim is disputed in the literature, and the denial does not survive the weighing; a page should say both.

The weighing takes the cited ratings as its input, so a contest result does not make ratings out of nothing. It says what the ratings already in the library imply once the positions that hold and deny a claim are heard together. A rating still moves only with a new, cited argument.

A worked example

Kenneth Howell, "Aren't We Saved by Faith Alone?", Catholic Answers Magazine, 2003, is a dialogue between a Protestant objector and a Catholic. TestimonyArguments.Articles.Howell2003 models four of its moves, quoted from the publisher's page, and hears one of them.

The argument about works. "If Paul and James mean the same thing by works, then they contradict one another. Since you and I both believe that the Bible cannot contradict itself, we must agree that Paul and James mean two different things by the word works." Romans 4:2–5 and James 2:24 share three words, not one — works, justify and faith — and their clash needs all three the same. The Bible's consistency rules out only that all three agree; it does not choose works (consistency_does_not_choose_works, refuted by the reading on which James uses "justify" otherwise, as Moo and Calvin read him). The article does give a reason for its choice — Paul's "works of the law" — and that reason is weighed where the library weighs ἔργα νόμου.

The argument from the phrase. "The phrase 'faith alone' does occur in the New Testament: one time, in James 2:24. There the inspired apostle denies that justification is from faith alone." To deny sola fide by this, James's "faith" and "justify" must be the ones the Reformers' formula uses of Paul. The move does not say so (faithAlone_unstated), and those are exactly the two senses one claim in the library denies, james2_24Compatible (faithAlone_meets_the_library).

Heard in the dispute, it cannot be defended. Stated as a position, with the senses it needs made explicit, and heard with every party to the sola fide dispute and with James's own words about faith, the move falls (howell_indefensible). James calls the faith he denies justifies dead (2:17, 2:26); Trent reads it so (Session VI, ch. 7); the Reformers' confession calls the faith that alone justifies "no dead faith" (Westminster XI.2). The denial the move needs does not survive the weighing (faithDenialAnswered). The article's other case from James — that works "actually justify" — is the Catholic reading of James's "by works", weighed in TestimonyArguments.SolaFide.James: as a claim about James's word it cannot be defended (trent_on_james_word_indefensible); as a claim about what works do before God it can, and is not forced (trent_on_works_defensible).

The same check, on the other side. The objector's reply — James 2:14 "is dealing with the problem of those who claim faith but who don't show it by their works" — takes "faith" in 2:14 for "faith" in 2:24 without saying so (deadFaith_unstated). That is the Reformed reading's own condition, and the claim that carries it, james2TargetsDeadFaith, is rated disputed too.

What each position rests on

Testimony.Logic.Warrant answers a reader's question about any position: where does this rest on someone's word alone? Each cited work has a role — a text or lexicon, argued scholarship, a binding decree or confession, or a theologian's own teaching — and a premise's warrant is the best of its references: Scripture, evidence, authority alone, or the library's own assertion. A claim about what someone holds is proved by their own text, so it rests on evidence.

The warrant is reported beside the verdicts and never weighed in them. A rule that discounted authority inside the solver — "Scripture first, by default" — would decide the sola scriptura dispute by fiat, since whether a council's or a confession's word binds is what that dispute is about. And it applies to every side: in the James dispute (james_warrants) Trent's readings rest on Trent's word, and the readings of the Greek on Pius XII's and Westminster's at "the original outweighs any translation". The Reformed harmony rested on Westminster's word at "works are fruit, not ground" until it was given the Assembly's own proof texts as a step (fruitLine); it rests on Scripture now, and the step is still disputed, because Trent grants the texts and calls the fruit merit.

The same check answers whether Rome contradicts itself over Trent's Latin texts (TestimonyArguments.SolaFide.Vulgate). Read as what the inspired authors wrote, Trent's "as it is written" at Revelation 22:11 cannot be held with Pius XII's teaching that the original outweighs any translation (trent_as_the_inspired_text_contradicts_pius). Read as Pius XII reads Trent's decree — the Vulgate juridically authentic and free from error in faith — Rome is consistent (pius_reading_is_consistent), and the two texts then rest on the Church's word (pius_reading_rests_on_authority), which Pius XII asks to be confirmed from the originals; at these verses the originals do not confirm it (pius_finds_no_confirmation).

Computed ratings

A rating today is asserted, with a citation: disputed when a cited source grants a step's grounds and denies its conclusion. That records that someone dissents, not whether the dissent is an argument or only "I disagree". Computed ratings sets out the design for telling them apart — five checks a dissent must pass to count as credible, under a stated standard of evidence — and its first phase is built. It finds that Howell's move is a disagreement, not a critique (howell_asserts_his_denial). Trent's denial that works are fruit, not ground, rests at one claim on the council's word, so that step stands under the evidence standard and is disputed under the tradition standard (fruit_turns_on_the_standard). And what James's δικαιόω denotes turns on the standard in the same way (james_word_turns_on_the_standard). Contested and DenialAnswered remain the checks of whether a dissent survives the weighing: the Howell hearing shows one that does not (faithDenialAnswered), the James dispute one that does (worksContested).

What it would take to decide James

The library says a reading is right in one sense only: the dispute forces it. A reading is forced when it is in the grounded extension — every rival reading is indefensible, given premises that are cited and rated. Paul's word is forced in that sense (only_pauls_word_prevails), because the lexical case rests on links rated above disputed and nothing that attacks it is as strong. For James, four things are missing.

1. The dispute now exists (TestimonyArguments.SolaFide.James), with the Catholic reading as two parties: James's word as the increase of justice, which cannot be defended; and what works do before God (canon 24), which can, against the Reformed harmony. What is left to decide is the second.
2. The question splits by word and by verse. What δικαιόω denotes at 2:21, 2:24 and 2:25; what πίστις denotes at 2:14, 2:17, 2:19 and 2:24; and whether each keeps its sense through the paragraph. The grammar can now state each as its own claim, and sameSense the claim that one word means the same at two verses.
3. Evidence that discriminates between the readings, each a cited premise:
   • 2:14 asks whether "the faith" can save — the faith just described, said and not shown. The article is in the Greek; that it points back is a reading, to be rated.
   • 2:21–23: Abraham is "justified by works when he offered up Isaac" (Genesis 22), and 2:23 says this fulfilled Genesis 15:6, said of him years before. That one justification follows the other is in the texts; whether "justify" then means a second verdict, a showing, or an increase is where the readings part.
   • 2:18, "show me your faith", and the uses of δικαιόω elsewhere for being shown to be in the right (to be verified against a lexicon before any is cited).
4. Ratings audited, not assumed. Each step stays disputed until the rating auditor finds no cited reader who grants its grounds and denies its conclusion. If every step on both sides stays disputed, the solver will report that James is contested — as it now does — and the page will say so. A forced reading needs one side to rest on something the other cannot match.

The textual facts in the third item can also be checked against an annotated Greek text — that the article stands in 2:14, which words recur through 2:14–26 — rather than cited alone (MACULA Greek carries the syntax and Louw–Nida sense domains).

The checks an article audit needs

Each common charge of fallacy is a question the library can make precise. Two are built; the rest reuse machinery that exists, or that the open issues propose.

Two of these are worth a word, because Howell's article already supplies them.

• An attribution has readings, as a claim does. The objector asks how the Church can teach "that we have to earn our salvation". Against the library's Trent, that is ambiguous: if "earn" means by our own efforts, Trent denies it, and the Catholic says so; if it means works done in grace meriting an increase of justification, Trent holds it (worksMeritIncreaseOfJustification). The grammar tells these apart — works and worksOfFaith are different concepts — so an attribution check can report which reading of the attribution is fair, rather than calling it a straw man or not.
• An implicature is not an assertion. The Catholic says Luther "added the word alone—but alone is not in the original Greek text". Both claims are true of the texts, and the move is modelled as asserting exactly them. What it may be taken to imply — that the reading was new with Luther — is a different act, and a check must not charge an article with what it only implies.

Rules for checking someone else's argument

The library's own rules carry over, and four more follow from them.

• Quote, do not paraphrase. A move records the article's words exactly. A model of what an article "basically says" is a straw man in waiting.
• Every voice gets the same check. Howell's objector is checked as his Catholic is. An audit that only finds faults on one side proves nothing about the side it spared.
• Report the verdict the solver computes. An article's claim is indefensible only when the dispute says so. Howell's move is not, and the page says that.
• Apply the article's principles to the article. When an article states a principle — that the Bible cannot contradict itself, so two authors must mean different things by one word — the same principle applies to every word it uses that way.

What comes next

1. Meanings for every argument. Every argument is now analysed where a join turns on it; the rest of each argument's claims are registered and unanalysed.
2. A James dispute, with the Catholic reading as a party, so that the question the Howell hearing leaves contested is asked directly.
3. The remaining checks, in the order of the table, each as a checked certificate that reuses Because, Dilemma and the solver.
4. Articles as pages. An audit, or an ontology article, is published prose like an argument page, generated from its docstrings. Today an ontology article's prose is its module's docstrings, and the generators render arguments only.
5. The next ontology articles, one for each finding above: textual words for textual claims; baptism kept apart from the baptisms; kinds and properties told apart; faith alone stated once; and voices taken from the traditions the library already cites.