Most re-crawl bytes buy nothing. The signed index tells the agent what moved.
The index lists every page with a SHA-256 digest, an ETag, token counts, and triage fields. An agent that stored digests from its last visit compares them locally and fetches only the pages whose digests changed.
| Scenario | Bytes transferred | vs HTML crawl |
|---|---|---|
| Full HTML crawl | 1,205,292 | baseline |
| Markdown conversion | 375,630 | −68.83% |
| Delta consumption | 51,408 | −95.73% |
On the repository's 60-page corpus those are the measured numbers. The index itself is small, signed, and cacheable, so the saving survives retries.