[Maya-Kid]

A WebDAV href such as plan%23draft.txt or what%3Fnow.txt was decoded before GET, turning filename characters into a URL fragment or query. Literal percent escapes could fetch a different file, and the same problem in a collection name could abort recursive synchronization.

Keep escaped href paths for GET and recursive PROPFIND, separately from document identity. Ordinary existing identities retain their shape; proxy-rewritten absolute hrefs are parsed before their path segments are decoded and reserialized, so distinct #, ?, and literal % names cannot collapse to the same document key.

Validation:

• Both local HTTP regressions failed before the fix (1 of 4 files fetched; encoded collection returned 404). A review regression also reproduced proxy-key collisions before they were fixed.
• The local HTTP fixture checks exact request targets, nine distinct filenames across relative/same-host/proxy hrefs, stable ordinary identities, Unicode/spaces, and nested collection traversal/self-skip.
• All 10 local WebDAV tests pass. The existing live-server test requires UTOPIA_DAV_TEST_URL and was not exercised.
• Workspace formatting, clippy with -D warnings, tests with PostgreSQL 16/pgvector and required DB/PDF guards, and build pass using Rust 1.98.1.
• The unchanged web baseline passes its 223 tests and production build, including typecheck and style guard.

[WaylandYang]

Thanks @Maya-Kid. Keeping two paths, the escaped one for the request and the decoded one for the document's identity, is the right split: %23 and %3F stay filename bytes on the wire, and ordinary files keep the keys they already have. The proxy branch decoding after parsing and re-encoding per segment is what keeps #, ? and a literal % distinct there. The 11 WebDAV tests pass here. Landing it.