What is the function-word floor in TypeGlish 0.3.0?

0.3.0 makes the closed-class function words of English paint, so the compiler reads them as structure instead of passing over them as prose: articles (a, an, the), prepositions, the quantifiers every and each, and the bare copula be. Earlier releases filled in the open word classes (nouns, verbs, adjectives, adverbs); this is the floor beneath them, the point where the small words that hold a sentence together are read too.

Does TypeGlish read quantifiers like every and each?

Yes. In 0.3.0 every and each paint as the universal quantifier, the same corner of the square of opposition as all. So 'Every refund is approved' and 'No refund is approved' are read as opposite claims about the same set, and the checker proves they cannot both hold with logic/quantifier, a blocking error.

Do prepositions and adverbs change how a rule is checked?

They do, because they paint now too. 'Every refund is approved automatically' and 'No refund is approved without a manager' compile clean side by side: the adverb and the prepositional phrase give the two rules different predicates, so there is no conflict to report. The same small words that can prove a contradiction also keep the checker from inventing one.

Does upgrading to 0.3.0 change how my existing .tg files check?

The change is additive coverage, not a new gate. A prompt whose function words already lined up keeps compiling; where two quantified rules genuinely conflict on the same predicate, 0.3.0 now sees it. Run npx typeglish check after upgrading to confirm, and reconcile any quantifier conflict it surfaces.

← Tidelines/ReleasesJul 21, 2026

TypeGlish 0.3.0: the function-word floor

The small words carry the logic. 0.3.0 makes articles, prepositions, the quantifiers every and each, and the bare copula paint, so the compiler reads the words that hold a sentence together instead of stepping over them.

by TypeGlish team5 min read#releasestypeglish 0.3.0

Every word counts.

TL;DR 0.3.0 lays the function-word floor: articles, prepositions, the quantifiers every and each, and the bare copula be now paint, so the compiler reads the closed-class words instead of skimming them. The payoff is logic that lives in the small words: Every refund is approved and No refund is approved now provably conflict (logic/quantifier), while an adverb or a preposition that splits two rules apart keeps a false conflict from firing. Upgrade with npm i -D typeglish@latest.

For four releases the world model has been filling in the open word classes: nouns, verbs, adjectives, adverbs, the words that carry the topic of a sentence. 0.3.0 lays the floor underneath all of it. The closed classes, the function words that never change and do the structural work, now paint too: the articles a / an / the, the prepositions, the quantifiers every and each, and the bare copula be. This is the point in the coverage map where the small words stop being scenery and start being read. Upgrade with npm i -D typeglish@latest, or run it with no install at all: npx typeglish@latest check **/*.tg. Here is why it matters.

§1The small words paint now

A function word is small, fixed, and load-bearing. English has only a few hundred of them and you never coin a new one, but they are where the logic of a sentence lives: every against no, the against a, with against without. Until now the checker read the topic words and let most of these pass as prose. 0.3.0 paints them, and the first place you feel it is the quantifiers: every and each now read as the universal, the same corner of the square of opposition as all. That is enough to prove a contradiction that hides entirely in the small words.

refunds.tglogic/quantifier

You are a refund assistant.

Every refund is approved.
No refund is approved.

Nothing in those two rules is a keyword in the old sense: no MUST, no WHEN, no numbers. The only words that disagree are every and no over the same predicate, and 0.3.0 reads them as the universal and its denial, so the checker blocks the build with logic/quantifier and names the line to keep. The CI guard on this post re-proves that exact code against the real checker on every build. It is the same square of opposition behind Say it once, or say it's optional, now reaching the two quantifiers people reach for most.

This is the kind of conflict that survives review because it reads fine. "Every refund is approved" and "No refund is approved" sit ten lines apart in a real policy, one in the happy path and one in an exceptions block, and a human skims past both. The compiler does not, because it now reads every and no as the operators they are.

§2And they keep the checker honest

The same painting that catches a real conflict is what stops a false one. A quantifier only contradicts another when they claim the same thing, and whether two rules claim the same thing often turns on a preposition or an adverb, another set of function words 0.3.0 now reads. Change the small words after the verb and the two rules stop being about the same predicate, so they coexist:

refunds-scoped.tg✓ compiles

You are a refund assistant.

Every refund is approved automatically.
No refund is approved without a manager.

Same subject, same quantifiers, and yet this compiles clean, because the adverb automatically and the prepositional phrase without a manager give the two rules different predicates. Approving automatically and approving without a manager are not the same act, so every and no are talking past each other, not fighting. A checker that skipped the small words would either miss the conflict in §1 or invent one here; reading them is what tells the two cases apart. The guard on this post asserts this file compiles with no blocking error.

That is the whole point of a floor: not a new rule to obey, but the last layer of words the compiler reads, so its verdicts turn on the same details you meant when you wrote the sentence. It is the closed-class complement to the open word classes the 0.2.3 notes laid out, and it is the piece that lets a definite article bind an object, a preposition scope a phrase, and a quantifier mean exactly one thing.

Field note

0.3.0 is a focused release: one change, the function-word floor, and it is a coverage milestone rather than a new feature to configure. It fires only what it can prove, the same as always, and there are no benchmark or adoption numbers to report. If npx typeglish check surfaces a quantifier conflict on a file that used to pass, that is the floor doing its job, reading a disagreement that was always there. The full changelog lives on GitHub. For the object system a definite article builds on, read the 0.2.2 notes; for why a proven contradiction is worth blocking on, read Your prompt argues with itself.

FAQQuestions about 0.3.0

What is the function-word floor?: The closed-class words of English, the ones that never change and do the structural work, now paint: articles (a, an, the), prepositions, the quantifiers every and each, and the bare copula be. Earlier releases filled in the open classes (nouns, verbs, adjectives, adverbs); this is the floor beneath them, where the small words that hold a sentence together are read too.
Does TypeGlish read every and each now?: Yes. They paint as the universal quantifier, the same corner of the square of opposition as all. So Every refund is approved and No refund is approved are opposite claims about the same set, and the checker proves they cannot both hold with logic/quantifier, a blocking error.
Do prepositions and adverbs change how a rule is checked?: They do, because they paint now too. Every refund is approved automatically and No refund is approved without a manager compile clean side by side: the adverb and the prepositional phrase give the rules different predicates, so there is nothing to report. The same small words that can prove a contradiction also keep the checker from inventing one.
Does upgrading change how my existing files check?: It is additive coverage, not a new gate. A prompt whose function words already lined up keeps compiling; where two quantified rules genuinely conflict on one predicate, 0.3.0 now sees it. Run npx typeglish check after upgrading, and reconcile any quantifier conflict it surfaces.

∿ washed up Jul 21, 2026 ∿