Principles, the Minting Rule, and Diagnostics

The schema inventory, grounding mechanism, and lexical-unit profiles covered so far are the what of the conceptual dimension. This document is the how — the standing rules a modeler applies when analyzing a frame, deciding whether a candidate structure deserves its own object, or working out which schema a frame actually belongs in.

Principles#

Nine standing commitments govern every schema, frame, and lexical unit in this layer. None of them is re-argued in an individual schema's own entry — they're assumed throughout.

One fact, one home. Each distinct fact is recorded by exactly one mechanism, in exactly one place. A derived fact is computed on demand, never stored a second time next to the fact it's derived from — because two homes for one fact can silently disagree with each other, and a single home never can.

Descriptive of language, normative of the record. The layer documents linguistic behavior; it doesn't legislate it. What is strict is how a fact gets recorded once a modeler has decided what it is — which mechanism carries it, which register it belongs to, that it has exactly one home.

Construal over classification. Where a fact could be recorded either as a single-valued classification or as a set-valued construal, record it as a construal. One situation genuinely admits several legitimate readings at once — o gelo derreteu ("the ice melted") is a change of state and a phenomenon at once — and a construal is the device built to hold both without forcing a choice.

Meaning, not syntax. Classification tracks what something means, never its grammatical form. O barulho me assustou ("the noise scared me") and eu temo o barulho ("I fear the noise") share one analysis despite realizing it with different syntax; a maintaining verb is a stative event and an attributive verb is an attribution, regardless of both being verbs.

Few primitives, many named objects. Exactly two primitives exist, and nothing is ever added to them. Named objects — schemas and microframes alike — are expected to be numerous and specific, each one traceable back to a primitive by a walk anyone can check. Generality belongs to the two primitives; specificity belongs everywhere else.

Diagnostics, not definitions. A rule a modeler can't actually apply to a real sentence isn't a usable rule. Every judgment this layer asks for comes with a test: a question about one concrete expression whose answer decides the case, reproducible by anyone who tries it on the same example.

Patterns are generative. A pattern holding in three places predicts a fourth. Recognizing that a pattern applies is what supplies new candidates worth checking — the system proposes, and a modeler accepts or rejects, rather than every candidate having to be found by inspection from scratch.

The FE-inventory criterion. An element exists exactly when some lexical frame might plausibly bind to it — no more, because an element nothing could ever bind to isn't earning its place, and no less, because an element some frame genuinely needs belongs in the inventory even where it looks like it should collapse into something else.

Anchoring. Every element must trace back to a primitive — through its parent, if it's derived, or through its own filler restriction, if it's native. Where a frame grounds in more than one schema at once, every core frame element only has to anchor across the union of all its groundings, not through any one grounding taken alone.

The minting rule#

The question a modeler actually faces, over and over, is simple to state and easy to get wrong by instinct: is this candidate a genuinely new object, or just a new way of looking at something that already exists? Getting it right matters, because an object is something you can inherit from, coerce to, and point at — a profile is none of those things, just a selection over an object that already is one.

Two questions have to be settled before running any test, because skipping them produces a confident wrong answer just as easily as a confident right one:

• What is the base? Name the actual structure the candidate would be compared against. "What is the base here?" is a far better question to start with than "should I split this into two things?" — a base is something you can go inspect; a split is only a preference until you have one.
• Which inventory would it belong to? A candidate whose entire content is a force between two regions belongs among the microframes; a candidate that carries genuine element structure belongs among the schemas. This is decided by what the base actually offers, not by which one seems tidier.

With the base fixed, five tests decide object or profile, asked in order — the first yes mints an object, and five nos make the candidate a profile of its base. A sixth test, on a deliberately softer bar, is asked only once those five are settled.

Test 1 — Inventory. Does the candidate need a part the base doesn't have — an element, or a force between elements the base leaves unrelated? PART_WHOLE over DIMENSION adds a Part and asserts a mereological force the bare dimension doesn't; CLASS over the same base adds a Member and Criterion and asserts a classification force instead. A roda é parte do carro ("the wheel is part of the car") and o gato é um mamífero ("the cat is a mammal") aren't two views of one relation — they're two different relations, and a base that leaves the regions unrelated has neither. EVENT itself, minted over bare REGION, is the sharpest case in the whole inventory: REGION has nothing at all to derive from, so both of EVENT's elements — Agonist, the exerter, and Event, the force itself — are native.

Test 2 — Constraint. Does the candidate pin down something the base leaves free? EXPERIENTIAL's Antagonist is constitutively sentient — there's no fearing without something capable of fear. Contrast AGENTIVE: it may read its exerter as sentient, but only optionally — a tempestade derrubou a árvore ("the storm knocked down the tree") is agentive with no sentience anywhere in it — so a restriction minted purely for that optional reading wouldn't fire (AGENTIVE itself mints on other grounds: adding Antagonist, and pinning its own exerting row to a specifically agentive one). The word doing the real work here is constitutive: if removing the restriction still leaves something meaningful, it was only ever one admissible reading.

Test 3 — Process. Does the candidate change the steps, rather than simply looking away from a step that's still there? PROCESS over CHANGE genuinely expands a single transition into an ordered chain of intermediate positions — a água ferveu lentamente ("the water boiled slowly") has a real part-way point plain CHANGE alone doesn't. A step that's merely backgrounded doesn't count: a janela quebrou ("the window broke") still entails an exerting step even with no exerter in view.

Test 4 — Invariant. Does the candidate hold something fixed that the base lets vary? STATIVE asserts that the position before and after are held equal, against a force that would otherwise change them — permanecer ("to remain") — where CHANGE on its own leaves those two free to differ.

Test 5 — Reference. Does something else in the layer need to point at the candidate specifically — some other object that mints on its own account and isn't a descendant of it? DIMENSION is licensed exactly this way: Attributes.Attribute and CHANGE.Ground both need a type to coerce to, and neither one descends from DIMENSION, so both are legitimate licensers. A candidate's own child pointing back at its own parent doesn't count — that's just the parent-child relationship restated, not a genuine outside need.

Test 6 — Reusable Naming. Does naming the candidate make it a recognizable, reusable part inside other composite schemas, even though its structure is identical to its base's? SOURCE is exactly a VALUE — one element, no new row — but "the starting point" is a role independent of any one frame's own traversal, and giving it a name lets a frame that names an origin as a thing in its own right ground somewhere purpose-built. This bar is genuinely softer than test 5's: it's licensed by definitional clarity and reuse alone, which is exactly why it's asked last, after everything else is settled, so it can't quietly stand in for a test-5 case that doesn't actually hold.

Every object this rule produces walks back to a primitive#

Running the five tests over and over, across the whole inventory, produces a structure that always terminates the same way: following any object's parentage backward always reaches one of the two primitives, and nothing else. This is principle five and principle nine made visible at once — few primitives, many named objects, every one of them anchored.

The two roots used to split the whole inventory cleanly by what an object fundamentally is: everything reaching REGION a thing or a space, everything reaching LINK a manifested force, with the static/dynamic split sitting entirely inside the LINK branch. EVENT moving to mint directly over REGION — rather than over LINK, the way it once did — flips that: LINK's own remaining children (Attributes, FIGURE_GROUND) are both static now, and it's REGION's own children that split between static (DIMENSION, OBJECT, VALUE and their own families) and dynamic (EVENT and its whole family). Either root is still sufficient, and the walk only ever needs to reach one of them — only which root the dynamic branch happens to reach has changed.

Two further, narrower rules#

The microframe rule is the same question, asked about a relation instead of a schema. Because a microframe can only ever carry one thing — which force holds between its two arguments — it has just one positive test (does some row genuinely need to name a force nothing already in the inventory names?) and a short list of exclusions that catch a converse of an existing relation, a narrower filler on an existing relation, two existing relations chained together, and a relation that already exists under an unfamiliar name. See Microframes for the worked version of this rule.

The identification rule answers a different question: does an existing lexical frame already state what a candidate schema would state, under a name that doesn't look like a match? Three questions about the cast — never the name — decide it: does every schema element have a counterpart core frame element, does every core frame element reach an element or resolve to a setting, and do the matched elements stand in the same relations. Where all three hold, the frame and the schema are one object, and the frame is what survives, keeping its own name, namespace, and lexical units. Attributes is exactly this case — a schema stating that a bearer holds a value on a dimension turned out to already be a lexical frame built for that purpose.

Diagnostics#

A diagnostic is a question about a real, concrete expression whose answer decides some assignment — never an abstract definition. The most-used diagnostic in the whole layer is the cascade that sorts a state-like or relation-like frame into its right place, asked in a fixed order, because each gate would give the wrong answer if asked before the one in front of it.

① comes first because a constitutive filler restriction is independent of both displacement and force — a frame can be experiential and dynamic (perceber, "to perceive") or experiential and static (saber, "to know"), and asking about displacement or force before this gate would misfile it either way.

② comes before ③ because the cascade would otherwise assume the very thing it's trying to sort out. A frame that isn't a state at all — dying, maturing — answers no at ③ and would be misfiled as an attribution if ② hadn't already caught it as dynamic first.

③ comes before ⑤ because a maintained state genuinely is a configuration of two regions, and would pass the exchange test at ⑤ too, incorrectly routing a stative frame into the relational branch.

④ guards ⑤ rather than deciding anything on its own. A one-role frame has nothing to exchange, so ⑤ is skipped entirely; a three-or-more-role frame asks whether the role left out of the exchange carries the value — if so, it's an attribution re-profiled rather than a relation reversed.

The test at ③ is what happens if it stops, never can you name what's holding it. Maintenance is very often asserted with no maintainer anywhere in the cast — permanecer ("to remain"), the schema's own defining example, has no exerter at all. Naming a maintainer, where the cast happens to supply one, is corroborating evidence; it is never the test itself.

Choosing among the relational schemas#

Once gate ⑤ settles that a frame is relational, one further fact — the converse predicate itself — decides which relational schema it belongs in, at no extra cost: the predicate that reverses the relation names a microframe, and that microframe's own home names the schema.

Pertencer ("to belong to") is genuinely ambiguous between two of these rows, and needs one further question to settle: does the second argument exist independently of what groups its members? A club does, surviving its members changing — is member of, PART_WHOLE. A class doesn't; it simply is what its criterion picks out — is classified by, CLASS.

These diagnostics decide meaning, never grammatical category, throughout — a maintaining verb is stative regardless of being a verb, and an attributive verb is an attribution regardless of being one too.