[d-morrison]

Closes #106.

Seven fragments wrote the Shannon entropy, cross-entropy and divergence as H_2(p), H_2(p, q) and D_2(p \,\|\, q), where the 2 means base-2 logarithms. The same site writes the Rényi entropy of order 2 as \renyientropyf{2}{X}, which prints the same glyph, so a reader could not tell the two apart. They now use the existing macros \entropyf, \crossentropyf and \kldivf, and the prose says "base-2 logarithms" where each result is stated.

Files: _thm-source-coding, _thm-wrong-code, _rem-cross-entropy-wrong-code, _exm-cross-entropy-decomposition, _rem-cross-entropy-loss, _thm-cross-entropy-decomposition, and _thm-kl-divergence-nonnegative. The last is a seventh file beyond the six that #106 lists; it writes D_2 the same way, and its proof uses the factor ln 2, so the statement and that proof line now say the divergence uses base-2 logarithms.

Edited lines gained links for the terms the term-link check found there: bits, binary digits, uniquely decodable code, loss and logarithms.

Checks run locally. Rendered entropy.qmd and source-coding.qmd to HTML: no H_2 or D_2 remains in source-coding.html, and the expanded math shows the macros' output. The only warning is downlit and xml2 missing from this machine. I looked at the expanded math in the built HTML, not a browser render. The term-link check passes through the push hook. 7 files, 21 lines changed, plus the term links.

Not changed: the collision entropy fragments, which ids#100 converts to \renyientropyf. The two PRs touch different files and can merge in either order.

---

---

Generated by Claude Code

[d-morrison]

/review

Requesting the Claude review on the current head.
Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[d-morrison]

/review

Merged main into the branch (head 3747d44); the merge adds check-gradient-lhs.yml and changes no fragment. Requesting the Claude review on the current head.
Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[d-morrison]

/review

Fixed the blocking finding in eeb5cec: the proof line at _thm-kl-divergence-nonnegative.qmd:41 now breaks after each of its three clauses, and so does the statement line at line 10 of that file. I also moved the loss link in _rem-cross-entropy-loss.qmd onto its own line, as the review suggested. I left the binary digits link: the term-link check requires a link on that edited line, and #def-bit is the closest definition. Requesting the Claude review on the current head.
Posted by Claude Code (AI agent) --- not written by a human.

---

Generated by Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[github-actions]

Claude finished review — View run

---

I reviewed commit eeb5cec, the only unreviewed commit since round 2. It only re-wraps lines and changes no wording, math or links.

Prior blocking finding is fixed.

• _subfiles/_thm-kl-divergence-nonnegative.qmd:43-45: this was the blocking finding from the round 2 review. The three-clause sentence is now broken after each clause, as that review suggested.
• _subfiles/_thm-kl-divergence-nonnegative.qmd:10-12: the "Then the … divergence, with base-2 logarithms, satisfies" sentence is now split into clause-level lines. That sentence was already in an earlier version of this PR and wasn't flagged before.
• _subfiles/_rem-cross-entropy-loss.qmd:15-17: the long line holding the [loss] link is now split. This was the non-blocking suggestion from round 2.

Rule 12 check. Each remaining line holds at most one clause. The split …$\crossentropyf{e}{q}$ / equals $\kldivf{e}{q}$: is a single main clause. The link and the "satisfies" fragment are short phrases, not clauses.

Markup. Link targets are unchanged. The split Markdown link text in _rem-cross-entropy-loss.qmd:15-16 still sits on one line, with the [loss](…) link intact. The edit adds no math and no macro changes, so the macro findings from rounds 1 and 2 still hold.

Carried-over suggestion, non-blocking. _subfiles/_thm-source-coding.qmd:19 links "binary digits" to #def-bit. Check that the bit definition equates the two terms.

Verdict

Ready for merge

Structured Review Data (JSON)

{
  "schema_version": "1.1",
  "reviewer": "claude",
  "commit_sha": "eeb5ceca144e71773ac2041707598914dc97a60f",
  "verdict": "CLEAN",
  "findings": [],
  "detailed_assessment": "The new commit breaks the three-clause proof line in _thm-kl-divergence-nonnegative.qmd after each clause and splits the long loss-link line in _rem-cross-entropy-loss.qmd; links and macros are unchanged.",
  "holistic_assessment": "The commit addresses the one blocking line-break finding from the previous round with no wording or scope change, so the earlier macro, anchor and base-2 checks still stand."
}

Reviewed commit: eeb5cec

[github-actions]

💰 Cost: $0.1806 (review) — run

[github-actions]

PR Preview Action v1.8.2
🚀 Preview available at: https://Morrison-Lab.github.io/ids/pr-preview/pr-108/

_{Built to branch gh-pages at 2026-11-08 07:05 UTC.}

[d-morrison]

Annotation review for head eeb5ceca: 23 passing check runs carry annotations, and I read the text of every one through the check-runs annotations API. The classes present are: the runner-wide ubuntu-latest migration notice; the SUBMODULES_TOKEN is empty notice. None concerns this PR's files. I am passing the ids to check-pr-fully-clean.py --annotations-read.

agy's quota is exhausted, so I ran the checker with --reviewer-out-of-quota agy; the Claude CI review on this head is the substitute.

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