[jepsontaylor]

wp_t0250_layout flow pass (wp_opus_layout) on main 058244645 (paper = ff80e94). Claim in AGENT_CHAT.md; it avoids wp_sol_writer's flywheel-figure work.

1. Figures 5 and 18 no longer print before their first citation

	main	this PR
Figure 5	p. 13, first cited p. 14 (drift −1)	p. 13, first cited p. 12 (drift +1)
Figure 18	p. 22, first cited p. 23 (drift −1)	p. 22, first cited p. 22 (drift 0)

Figure 5 (06_method/flywheel_status.tex):

• The opening sentence of §6.8, on p. 12, now cites it: "This is the adapter record (Figure 5) for the second learning task in Section 6."
• Lines 5–6 are rewrapped to the same two lines, so the P1 TBD stays at line 84 and every RESULTS_NEEDS.md anchor still lands on its \TBD{ line.
• Moving the figure itself would collide with Figures 6–8, which fill p. 14.

Figure 18 (appendix_latest_event_rates.tex, appendix_candidate_retained_trace.tex):

• Its discussion, "Admission frequency through a later cutoff", moves from D.13 to D.11, directly after D.10.
• The two admission sections (D.10, D.11) now precede the two program-event sections (D.12 "Program-event rate and round duration", D.13 "A later program-event window"). The moved section's text lands on p. 22 beside the figure.
• The \FloatBarrier that closed D.13 moves to the end of the appendix wrapper, so it still ends Appendix D before the references.
• Moving the figure later instead was tested in the t0150 pass: p. 23's left column ends 37 lines short.

No other float moves. The §6.8 pointer now renders "Appendices D, D.10, D.12 and D.13" (cleveref, from the same four labels).

2. Prose flow: Section 2

Every edit keeps the wording; there are no new claims, and §2 has no TBDs or anchors.

• §2.2: Campbell's three-mechanism paragraph ("every genuine gain in knowledge needs ...; the aim ... makes it the fourth pillar") now comes before the four pillar definitions it motivates. It used to sit after the Retention paragraph, whose last sentence already hands off to §2.3.
• §2.3: the one-sentence AlphaGo paragraph ("AlphaGo's policy network learned to select moves from human expert games, then refined that judgment through self-play") joins the chess-expertise paragraph it continues.
• Terminology: "using evaluation feedback to update the generator" becomes "update the proposer", matching §2.5 ("update its proposer", "improve the proposer"). "Generator" is kept for the random string generator, where it means something else.

Effect

• Words: the rendered word list differs from main only by:

  • "(Figure 5)";
  • "generator" → "proposer";
  • "D and D.10 to D.12" → "D, D.10, D.12 and D.13".

• Pages: only pp. 3, 4, 12, 14, 22 and 23 change.

• Labels: no figure, table, equation or main-text section label changes page or number. Only the three reordered appendix sections renumber:

  • candidate-retained-trace D.13 → D.11;
  • latest-event-rates D.11 → D.12;
  • latest-round-clock D.12 → D.13.

  All references to them use \cref.

• Short columns: unchanged (p. 20 right, below Table 10).

• Widows: none new.

Verification

Fresh git archive of exact head 9df058dd349eda0b320cd36b672f6cc4fec68673, nice -n 19 make pdf with SOURCE_DATE_EPOCH = parent commit time and the cm-super texmf:

• rc 0, 28 pages;
• 0 warnings, 0 boxes, 0 undefined references;
• 37 TBDs; 124/124 bib keys;
• all 35 RESULTS_NEEDS.md anchors on \TBD{ lines.

Further checks:

• Committed PDF: reproduces byte-for-byte. Fonts equal main's apart from object numbering.
• Record gates: five jope-records gates pass on a full export with the head's index.
• Whitespace: git diff --check is clean.
• Rendered text: no experiment names.
• CI: the paper CI lane builds docs/paper/innovationzero, not this paper.

No new lane (39/8). Receipt: ~/whitepaper_tmp_t541/layout-1102t0250/.

Merge: the owner brief wp_t0250_layout authorizes a self squash merge with --admin within 30 minutes of the passing build.

🤖 Generated with Claude Code

[jepsontaylor]

Same-agent review (wp_opus_layout, the author of this PR; not an independent review) at exact head 9df058d.

Scope: 5 files.

• 02_anatomy.tex: one paragraph moved, two paragraphs joined, one word changed.
• flywheel_status.tex: lines 5–6 rewrapped with "(\cref{fig:fw-latest-admissions})".
• appendix_latest_event_rates.tex: input order changed, plus the closing \FloatBarrier.
• appendix_candidate_retained_trace.tex: trailing \FloatBarrier removed.
• The PDF.

Findings, each checked:

1. Figure 5: first cited p. 12, prints p. 13. Figure 18: first cited p. 22, prints p. 22. No other float changes page.
2. Anchors: flywheel_status.tex keeps 102 lines, and the P1 TBD stays at line 84. All 35 RESULTS_NEEDS anchors land on \TBD{ lines.
3. Appendix order:
   • Content: the moved section has no order-dependent wording ("the earlier ledger", "this separate capture"). D.12 and D.13 mention neither D.11 nor each other.
   • Barrier: \FloatBarrier still closes Appendix D, so Figures 17–18 print before the references.
   • Numbering: only the three reordered subsections renumber, and every reference to them is \cref. A grep finds no hard-coded D.1x text.
4. §2 edits:
   • Words: the Campbell paragraph is moved verbatim, and the AlphaGo sentence is unchanged.
   • Meaning: "proposer" matches §2.5's usage. The random "generator" uses keep their separate meaning.
   • Flow: the new order sets out the three mechanisms and the aim as fourth pillar before defining the pillars.
5. Text: the word list differs only by "(Figure 5)", generator→proposer, and the cleveref list for the renumbered sections. No claim, number, citation or TBD changes.
6. Pages and gates:
   • 28 pages, 0 warnings, 0 boxes;
   • 37 TBDs, 124/124 keys;
   • five record gates pass, and the committed PDF reproduces;
   • git diff --check is clean; fonts equal main's apart from object numbering.

Disposition: no blocking finding.