---
title: "Architecture"
description: "How conversion, acquisition and execution fit together."
---

Webfold has distinct conversion, acquisition and execution paths. Local HTML conversion needs no browser process.

## Conversion

```text
HTML string -> to_blocks -> Markdown blocks -> Template page counts
                                    \                    |
                                     +---- render <------+  -> Markdown string
```

`to_blocks` scans UTF-8 input into blocks (`md`, a 64-bit `sig`, and `links`). `norm_sig` normalises whitespace and case before a non-cryptographic 64-bit fold. `Template::fold` counts each signature at most once per page. `render` walks blocks in source order and omits those whose frequency meets the threshold.

Tables are accumulated as a single block. Rectangular output, including generated header and separator cells, is allowed only when its cell slots fit a bound relative to the source cells and maximum width. Otherwise source rows become unpadded pipe lines.

## Acquisition

`Webfold::scrape` calls a `FetchBackend`. `scrape_tiered` adds three acquisition actions under a budget. `scrape --render` connects to a Chromium you run and observes computed CSS visibility and same-origin frames, behind a validating proxy. See [Scraping](/docs/guides/scraping) and [Rendered pages](/docs/guides/rendering).

## Execution

- An in-process DOM and JavaScript engine lifts and runs a supported script fragment. It is not V8.
- A CDP subset server exposes that engine over WebSockets. Every method is implemented, a named no-op, or unsupported.
- Screenshots use Webfold's own layout and paint pipeline rather than Chrome's.

## Layers

| Layer | Responsibility |
| --- | --- |
| L0 | Block signature hashing, template frequency counting, string extraction. |
| L1 | DOM arena, tag tokenizer, CSS cascade, structural fingerprints. |
| L2 | Execution runtime, sandboxing, origin-isolated contexts, stealth patches. |
| L3 | Crawl frontier, MCP and CDP servers, SERP parsing, compliance checks. |
| L4 | CLI binary and test suites. |

The full description, including known limits of each path, is in `ARCHITECTURE.md` and `docs/EXTRACTION-PIPELINE.md` in the repository.