[d-morrison]

🤖 Generated with Claude Code

https://claude.ai/code/session_019xfc9qrViJfWgYsyX5QcEX

Closes #617.

Why

Ezra asked why the odds-multiplier results from sds#614 wrote out \diff \vpredvar instead of using a shorthand macro, as rme uses \diffvx.
macros#426 added \diffvpredvar.

Changes

• Moves the latex-macros pin from e445125 to ae1d822, the merge of macros#426.
  The bump spans 115 commits; besides \diffvpredvar, they add macros (for example \ccprod, \quadmodel and \sqprederr) and change the CI wiring in macros.
  Only two existing definitions change in macros.qmd, and neither changes what it prints: \design is now \mdesign, which is \matr{D} as before (sds uses it in _rem-cross-product-matrix.qmd), and \stprod now routes through \ccprod, which is the same \tp{#1} #2. The PR preview is the full check.
• Replaces \diff \vpredvar with \diffvpredvar in the six odds-multiplier subfiles of logistic-inference.qmd. That's 19 occurrences, and it changes no rendered output: the macro prints the same Δx̰.

\difflogodds is not used here.
The sds proofs never write a difference of log-odds, so adopting it would mean adding a new definition, which is beyond this swap.

Also: fix the failing publish (9678341)

Every Quarto Publish run on main has failed since 2026-11-12 04:25 UTC (for example run 43370603721), so the site has not republished, and sds#614's new anchors are not live yet.
pdflatex stops on exponential-families.qmd with "Extra }, or forgotten $".
Quarto puts the title of #exm-uniform-not-exponential-family in the optional argument of \begin{example}[...], and the ] in $[0, \highend]$ closed that argument early.
Writing the math as ${[0, \highend]}$ keeps the ] inside a group, and the rendered title is unchanged.
Locally, quarto render exponential-families.qmd --to pdf fails with that error before the change and succeeds after it.
It is in this PR, not a separate one, because it blocks the publish that the odds-multiplier links in lds#2430 and rme#1497 need.

Checks

• quarto render logistic-inference.qmd --to html gives no warnings, and the rendered theorem shows \Delta followed by the vector x̰, as before.

---

Generated by Claude Code

[github-actions]

Claude finished review — View run

---

I fetched PR #618 live just now. It is open, not a draft, and not merged. Its head is still 96783418993df47632213ceb2e2421b3ce7363d2, the commit I reviewed. My review therefore still applies.

This round's change is one line. The heading in _subfiles/exponential-families/_exm-uniform-not-exponential-family.qmd now writes the interval as ${[0, \highend]}$, which keeps the semantic \highend macro. I did not render the PDF to confirm it compiles. I found no new issues. The prior review, posted 2026-11-13 at #618 (comment), gave Ready for merge on commit e031f966. This one-line change doesn't touch what it checked, so that verdict still stands.

Verdict

Ready for merge

Structured Review Data (JSON)

{
  "schema_version": "1.1",
  "reviewer": "claude",
  "commit_sha": "96783418993df47632213ceb2e2421b3ce7363d2",
  "verdict": "CLEAN",
  "findings": [],
  "detailed_assessment": "The only new change, in _exm-uniform-not-exponential-family.qmd line 2, wraps the interval in braces within the heading math, which is valid LaTeX and keeps the semantic highend macro.",
  "holistic_assessment": "The change is scoped to the PDF compile fix, leaves the prior notation review untouched, and affects no page paths or anchors; a PDF render was not run."
}

Reviewed commit: 9678341

[github-actions]

💰 Cost: $0.1918 (review) — run

[github-actions]

PR Preview Action v1.8.2
🚀 Preview available at: https://Morrison-Lab.github.io/sds/pr-preview/pr-618/

_{Built to branch gh-pages at 2026-11-13 06:33 UTC.}

[d-morrison]

Annotation dispositions for head 9678341:

I read the 55 annotations on all 51 check runs of 9678341, in 5 groups.

• Failure, false positive: "Canceling since a higher priority waiting request for preview-refs/pull/618/merge exists", together with "The operation was canceled.": both come from the build run that started at 06:00 UTC. Adding the preview:pdf label started a new preview build, which superseded it. That rerun passed, and its PDF deploy landed on gh-pages as df6bef7, with pr-preview/pr-618/exponential-families-handout.pdf. The cancellation came from concurrency, not from this diff.
• Notice, routine: "Skipping element-level diff highlighting: 754 old / 778 new candidate elements exceed the 500-element cap": the tracked-changes step cut back its highlighting. That limit is in the step's tooling, and nothing here is a defect.
• Notice, routine: "SUBMODULES_TOKEN is empty; cloning submodules anonymously": latex-macros is public, so the anonymous clone works.
• Notice, routine (51 runs): "The ubuntu-latest label will migrate to Ubuntu 26": a runner-image notice from GitHub that applies to every repository, not to this diff.

Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[d-morrison]

Cross-harness review (agy Flash) on 9678341: VERDICT: CLEAN

Every \diff \vpredvar in the six odds-multiplier subfiles is replaced by \diffvpredvar (same rendered math), the submodule pointer moves to ae1d822f, the macros head where \diffvpredvar is defined, and the title ${[0, \highend]}$ puts the ] inside a group so it no longer ends the \begin{...}[title] argument. Beyond the diff, I cloned sds main (9f42c4e) and grepped every #### title in .qmd files for a ]: the only match is the title this PR fixes, so no other title can fail the same way. I did not render the PDF; I rely on the owner's green preview:pdf.

Posted by Claude Code (AI agent) --- not written by a human.

[d-morrison]

Narrowing a claim in my review above

I wrote that no other title can fail the same way. What I measured is narrower: on sds main (9f42c4e, tracked files only, no submodules), a heading of any level in the two lines after a theorem-type div opener (thm-, def-, lem-, cor-, prp-, exm-, exr-, rem-, cnj-, axm-, alg-) with a ] matched only the title this PR fixes, and a title= attribute with a ] matched none. I did not scan the submodules, a title placed further from its opener, or other title forms, and I did not render the PDF. The verdict and head are unchanged.

Posted by Claude Code (AI agent) --- not written by a human.