> ## Documentation Index
> Fetch the complete documentation index at: https://differens.ossl.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Tier Adapters

> How non-code file formats are diffed

## Tier system

Every file that Differens processes goes through a tier adapter. The content router classifies files by extension and dispatches them to the right tier. Each tier can fall back to the tier below it. If the code parser fails, the line diff catches it. If the file is too large for the line diff, the hash fallback catches it.

### Tier reference

| Tier | Name   | Handles                                                  | Detects                                                            | Fallback |
| ---- | ------ | -------------------------------------------------------- | ------------------------------------------------------------------ | -------- |
| T5   | Code   | `.ts`, `.tsx`, `.js`, `.jsx`, `.py`, `.rs`, `.go`        | Functions, classes, methods, imports, renames, moves               | T1       |
| T4   | Data   | `.json`, `.yaml`, `.yml`, `.toml`                        | Key additions, removals, value changes, nesting changes            | T1       |
| T3   | Markup | `.html`, `.htm`, `.xml`, `.svg`                          | Element additions, removals, attribute changes, tree restructuring | T1       |
| T2   | Prose  | `.txt`, `.log`, `LICENSE`, `README`, `CHANGELOG`         | Word-level additions, deletions                                    | T1       |
| T1   | Raw    | All text files, markdown (`.md`, `.mdx`), reST, AsciiDoc | Line-level additions, deletions                                    | T0       |
| T0   | Binary | All binary files (terminal)                              | File-level hash comparison, byte-size delta                        | (none)   |

### Classification rules

The content router (`classifyFile` in `@ossl-dev/differens-tiers`) maps extensions:

* **Code extensions** (`.ts`, `.py`, `.rs`, `.go`, etc.) → T5, with a tree-sitter grammar lookup. If no grammar is installed for the language, the file falls back to T1.
* **Data extensions** (`.json`, `.yaml`, `.toml`) → T4. Unparseable data (malformed JSON) falls to T1.
* **Markup** (`.html`, `.htm`, `.xml`, `.svg`) → T3. Element tree with attribute diffs.
* **Markdown and doc formats** (`.md`, `.mdx`, `.rst`, `.adoc`) → T1 line diff.
* **Prose** (`.txt`, `.log`, files named `LICENSE`, `README`, `CHANGELOG`, `AUTHORS`) → T2 word-level diff.
* **Everything else** → T1 (line diff) or T0 (binary detection via null bytes).

### Graceful degradation

The pipeline degrades instead of failing. At every level:

1. Try the most specific parser for the file type.
2. If it throws, try the next tier down.
3. T0 is the terminal tier. It compares file hashes, and there is no tier below it.

Run `differens` on any file, in any format, at any size. The result ranges from a full semantic diff down to a hash comparison: "file changed, 1423 bytes different."

### Adding a tier adapter

New formats get new tiers. See [Contributing](/guides/contributing) for the step-by-step guide. Every adapter must produce a `Node` tree the diff core can consume. What the tree contains is up to the adapter. It can be a tree-sitter CST, a JSON path tree, an HTML element tree, or any other structured representation.