[Jammy2211]

The --mark-unresolved pass — applied by hand rather than by the flag, and the
reason why is the point of this PR.

The residue split almost in half

After the retrieval fix (#64), 50 lines still would not resolve:

• 22 are your own annotations — Cant remember, CONTAINS SMBHB EVOLUTION EQUATIONS, Good overview of dual AGN and looking for them
• 28 are real papers whose lines lost content on the way in —
  A Super-Eddington, Lensing-Magnified Quasar at ␣␣observed with JWST is
  missing its redshift, not mangling it, and no query can search for words that
  are not there

--mark-unresolved marks everything unresolved, so running it here would
have NOTE-marked those 28 real papers too and dropped them off the queue. The
flag is built for a residue that is mostly annotations; this one is 44% papers.
So the 22 are marked by their exact text, one at a time — indentation is nearly
a signal in this file but not reliably one, and a mis-marked line silently loses
a paper.

One repair rather than a mark

Eryn: A multi-purpose sampler for Bayesian inference: https://arxiv.org/list/astro-ph/new
carried an arXiv listing URL welded onto its title, which no title search can
match. The URL is dropped; the paper stays a paper.

The queue now reads honestly

Papers waiting	186
With a 📄 PDF button	158
Search-only	28
Annotations (NOTE)	38
Sub-headings	7

Every one of the 28 remaining search-only lines is a real paper with a damaged
title — a queue-hygiene problem rather than a matcher one. They stay
unmarked deliberately: fix a line and the nightly picks it up.

make validate clean, 121 tests pass.

---

Generated by Claude Code