diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2ec2d78..5d85a45 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -93,6 +93,14 @@ _No changes yet._
   suffix-less API/RSS paths like `news.ycombinator.com/rss` still
   route through raw passthrough (`fetcher_used: passthrough-probed`).
 
+- **Pluggable PDF extraction backends**
+  (`src/trawl/fetchers/pdf_backends.py`) — the PDF fetcher now routes
+  through a backend layer (`pymupdf` / `markitdown` / `unstructured` /
+  `docling` / `mineru`, heavy parsers behind lazy imports). The
+  production default is unchanged (PyMuPDF); the layer powers the R5
+  backend comparison harness, which concluded keep-PyMuPDF (see
+  Tests / research below).
+
 ### Fixed
 
 - **Wikipedia fetcher heading preservation** — modern MediaWiki HTML
diff --git a/CLAUDE.md b/CLAUDE.md
index 9878c09..cc39f1e 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -37,8 +37,9 @@ trawl directory. Humans should read `README.md` first, then
 - **Longform retrieval cost (default on, 2026-04-22)**: `TRAWL_CHUNK_BUDGET`
   default flipped from `0` to `100` after re-validating on curl.se
   manpage (275 KB / 760 chunks, p95 25149 ms → 3065 ms). Parity 15/15
-  + agent_patterns coding 23/24 (pre-existing unrelated `arxiv_pdf_lora`
-  fetcher fail) preserved. Opt out via `TRAWL_CHUNK_BUDGET=0`.
+  + agent_patterns coding 23/24 at that time preserved (the unrelated
+  `arxiv_pdf_lora` fetcher flake has since recovered — coding is 24/24
+  as of v0.4.5). Opt out via `TRAWL_CHUNK_BUDGET=0`.
 
 ### What a new session should do first