Pipeline

The grammar pipeline connects go-html rendering to go-i18n's reversal engine, enabling semantic verification of rendered output via GrammarImprint.

See also: Home | Architecture | Node-API | WebAssembly | Code-Generation

Pipeline Flow

Node Tree
    │
    ▼  Render(ctx)
HTML String
    │
    ▼  StripTags()
Plain Text
    │
    ▼  reversal.NewTokeniser().Tokenise()
[]Token
    │
    ▼  reversal.NewImprint().Build()
GrammarImprint

The pipeline transforms rendered HTML into a grammar fingerprint that captures the linguistic structure of the content without preserving the actual words.

StripTags

func StripTags(html string) string

Removes all HTML tags and collapses whitespace. Handles nested elements, void tags, and attribute content. Used as the bridge between HTML rendering and grammar analysis.

text := html.StripTags("<h1>Hello <em>World</em></h1>")
// "Hello World"

Imprint

func Imprint(node Node, ctx *Context) reversal.GrammarImprint

Runs the full pipeline: render → strip tags → tokenise → build imprint. Returns a GrammarImprint from go-i18n's reversal engine.

node := html.El("p", html.Text("page.greeting"))
imprint := html.Imprint(node, ctx)
// imprint is a lossy feature vector of the grammar structure

What GrammarImprint Captures

The imprint records grammar features without content:

Feature	Example
Verb tenses	"built" → past tense detected
Noun forms	"tests" → plural detected
Article usage	"an API" → indefinite article
Sentence structure	SVO ordering, clause boundaries
Token distribution	Ratio of verbs/nouns/articles/unknown

The imprint is lossy — you cannot reconstruct the original text from it. This is by design: it's a privacy-preserving grammar fingerprint.

CompareVariants

func CompareVariants(r *Responsive, ctx *Context) map[string]float64

Renders all variants of a Responsive layout, computes GrammarImprint for each, and returns pairwise similarity scores.

resp := html.NewResponsive().
    Variant("desktop", desktopLayout).
    Variant("mobile", mobileLayout)

scores := html.CompareVariants(resp, ctx)
// {"desktop:mobile": 0.85}

Use Cases

Content parity testing — Verify that mobile and desktop variants render linguistically equivalent content
A/B testing grammar — Compare how different layouts affect text structure
Regression detection — Alert when a layout change significantly alters the grammar profile
Poindexter integration — Feed imprint distances into the Lethean analysis layer

Integration with go-i18n

The pipeline depends on go-i18n's reversal engine:

go-i18n Component	Role in Pipeline
`i18n.Service`	Text lookup and grammar composition
`reversal.Tokeniser`	Classifies words as verb/noun/article/unknown
`reversal.GrammarImprint`	Lossy feature vector projection
`reversal.Multiplier`	Training data augmentation (not used in pipeline, but available)

The Context.Service field carries the i18n service. When nil, Text() nodes return the raw key string, and Imprint() still works (tokenising the key names instead of localised text).