Strokes

KanjiVG’s format note says the strokes follow the reference order, and that the same file contains groups which describe the structure of the character. When elements are not contiguous strokes, the project uses part and number. That is the sourced fact this page is built on. REAL

A. Spatial hierarchy — containment

Nested groups carry kvg:element (which formative the cluster resembles), kvg:position (left / right / top / bottom / middle / kamae / kamaec / nyo / nyoc / tare / tarec, plus Ideographic Description Character extras), and kvg:radical (general / jis / nelson / tradit). kvg:original names a semantic source when the drawn form is a variant — 休’s 亻 is original=人. kvg:phon marks a phoneticum; the project’s own notes call the values inconsistent. kvg:partial means the group does not exhaust the named element. kvg:part and kvg:number appear when one formative’s strokes are not a single block in writing order.

B. Linear sequence — ordered realization

<path> elements follow writing order. The id ends in -sN, N consecutive from 1. In the current format, document order is stroke order. Reverse it and you do not get “the same character, commutatively rearranged.” You get a different glyph tradition, or an error. That non-commutativity is why the sequence is a packaging of the form, not a bag of strokes.

Locked vocabulary for this dig, used as analysis rather than as an identity: ordered realization, hierarchical containment of formatives, valence = arity of component slots. A two-slot frame can be likened to a two-argument operator — cited from Geometric Syntax notes only as analogy. The likening is SPECULATIVE. It does not make the graph a syntactic head.

明 and 休 are both deep-2 left–right frames. Their surfaces are 8 and 6. 国 and 語 are both deep-2 as well — enclosure plus interior, semantic plus phonetic — at surfaces 8 and 14. Same slot count, different sequence lengths. REAL pattern. The echo on Valency is the clause-level slogan (surface arity ≠ geometry) pointed at graphs. It is still not linguistic arity, not on/kun counted as arguments, and not θ-roles.

Graph	Frame	Deep arity	Surface	Cross-cut
一	atomic	0-place	1	none
人	atomic	0-place	2	none
木	atomic	0-place	4	none at component level; H before V
明	left–right	2 (日, 月)	8	none
休	left–right	2 (亻, 木)	6	none; original=人 is not a third slot
国	enclosure (kamae)	2 (囗, 玉)	8	囗 leaves and returns (part 1 / part 2)
囲	enclosure (kamae)	2 (囗, 井)	7	double split: 囗, and 二/十 inside
間	門-enclosure	2 (門, 日)	12	mild; 門 is contiguous, no kvg:part
五	split radical	not a clean top-level 2	4	tradit 二 discontinuous; nelson = 一
圖	nested shells	part + number; not forced to 2	14	two split 囗 formatives, disambiguated
語	left–right	2 (言, 吾)	14	五’s 二-split inherited inside 吾
漢	left–right	2 (氵 + phonetic complex)	13	top contiguous; nesting stays inside the right slot

Deep arity in this table is the top-level component-slot count read off the KanjiVG groups. Where the file does not present a clean frame — 五, 圖 — the cell says so instead of inventing a number.

The identity cases. There is no sibling frame, so there is nothing for stroke count to be confused with except the paths themselves. 木 is the warning label: four strokes inside one formative are an ordering constraint (horizontal before vertical in the cross), not valence 4.

The clean Apply-shaped picture, offered as a SPECULATIVE local model and not as phrase structure: a left–right frame opens two slots and the pen saturates the left argument completely before the right. 明 surface 8 / deep 2 beside 休 surface 6 / deep 2 is the smallest demonstration that surface length and deep arity come apart. original=人 on 休 is variant packaging of the left slot.

国 and 囲 are the clearest kvg:part exhibits. One spatial 囗 is two stretches of the ordered realization: sides first, interior next, closing stroke last. The pen leaves the enclosure and comes back. That discontinuity is REAL as data. Calling it a rhizome — discontinuous, non-concatenative packaging — is a SPECULATIVE metaphor on top of the part-split. It adds no slot. 間 shows the contrast: a 門-enclosure often finishes both door-leaves before the interior, so the file needs no kvg:part. “Close the box last” is an ordering constraint on realization, family by family, not extra arity.

Dig #2 counted on the order of two thousand KanjiVG files with kvg:part= at the clone it read (HEAD 70a0b7a, 2026-09-30). Discontinuous packaging is ordinary in the source project. This site does not vendor that set; the diagrams here are the worked examples.

五 is the format documentation’s own picture of a radical that is not a block: traditional 二 wraps the middle strokes, while the Nelson index takes only the upper 一. 圖 is why kvg:number exists. Two different 囗 formatives are each split; number says which part-pair belongs together. The docs reserve number for that collision. Ordered realization here repeatedly leaves a shell and re-enters it. None of that is an X-bar projection.

語 is the inheritance case. Top-level deep arity is 2: semantic 言 on the left, phonetic 吾 on the right, surface 14. Inside the phonetic, 五 still splits 二. Discontinuity is inherited under composition rather than repaired. 漢 keeps a binary top frame (氵 + phonetic complex, surface 13) while the right argument is itself top–bottom. Higher nesting can raise internal arity without adding a third sibling at the character’s top. That last sentence is the SPECULATIVE analogy with a two-argument operator. kvg:phon is weak REAL: the attribute is in the files, and the project flags the values as inconsistent (issue #312).

KanjiVG’s Google Groups thread “Consistency rules” is a design argument inside the project, paraphrased here rather than quoted. One proposal: the XML hierarchy should reflect spatial containment — what element is part of what — even when that puts paths out of writing order. Writing sequence would then live on an explicit attribute (kvg:strokenum), consecutive from 1, possibly out of document order.

The live format did not take that fork. Document order is stroke order. Non-contiguous components are split with kvg:part, plus rare kvg:number. The thread is REAL as a project tension and as design history. It is not evidence that a later theory was “always” the file format. The narrow fact is enough: the project had to treat order and containment as separable packagings of the same graph. That supports reading stroke order as ordered realization over a containment tree, not as the composition operator itself. Later hierarchy-fix work in the project (pull requests that police group structure) keeps spatial nesting as a first-class object, not mere animation scaffolding.

Highlights from the Dig #2 map. A labeled analogy is not a load-bearing claim. Rejected rows are not available as interpretations of these files.

1. Frame inventory. Can kvg:position combinations be listed as a closed set of frames (left–right, top–bottom, kamae, tare, nyo, …) with stable arities? kamae is documented as an inconsistent grab-bag. Valence counts are not trustworthy until that sub-typology exists.
2. Arity greater than 2. Three-across characters, 品-style stacks, left–middle–right. How often is middle a true third argument, and how often is it a nested binary?
3. Phoneticum reliability. kvg:phon inconsistency (issue #312). Can phonetic versus semantic slots be recovered more cleanly from KANJIDIC2 or classical semantic–phonetic tables than from KanjiVG alone?
4. Variant files (-Kaisho and the like). Do stroke-order variants change deep arity, or only the ordered realization? The prediction is realization only. Not tested on this page.
5. Inheritance of splits. 語 inherits 五’s 二-split inside the phonetic. Is discontinuity inheritance systematic under composition?
6. Site scope. How far to carry stroke-order investigation without orphaning Dig #1. This page is the public track. Classifiers, VN-suru, and compounds stay where they are.
7. Citation hygiene. Geometric Syntax takeaways and the two-argument-operator note stay in the analogy rows. They are not promoted to load-bearing kanji theory.