← graph

build-bog-v2-3

Build BoG v2-zoom deck — figures full-bleed for Erik's review pass

Metadata

Statusdone
Assignedagent-878
Agent identityf51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e
Modelcodex:gpt-5.5
Created2026-05-06T02:04:23.639141062+00:00
Started2026-05-06T02:05:08.501705235+00:00
Completed2026-05-06T02:17:26.264662400+00:00
Tagsbog-this-week, bog-deck, render, review-mode, eval-scheduled
Eval score0.78
└ blocking impact0.75
└ completeness0.85
└ constraint fidelity0.85
└ coordination overhead0.80
└ correctness0.80
└ downstream usability0.75
└ efficiency0.80
└ intent fidelity0.64
└ style adherence0.80

Description

GOAL: Build a SECOND deck PDF — slides/v2-zoom/BoG_2026_zoom.pdf — that is a REVIEW-MODE companion to the main deck. Erik wants to look at each figure at near-full slide size to evaluate composition / readability / density before presenting. The main deck (slides/v2/BoG_2026.pdf, produced by build-bog-v2-2) packs bullets + figure + speaker notes into each slide; that deck is the talking deck. THIS deck is the "show me each plot big" deck.

DESIGN INTENT (lead-author verbatim):

"expanded view where we have all the plots designed to fit as well as possible on the slide, near full height and width, for me to review. right now everything is smushed, crushed, too dense, and sometimes often even too sparse"

CRITICAL: every figure should fill ~85-95% of slide area. No bullets on the rendered page. No speaker notes on the rendered page. A tiny title (one line, top) and a tiny source-path footer (one line, bottom small font) are OK for orientation — but the figure is the slide.

INPUTS:

  • slides/v2/slide_01_title.md … slide_15_concerted_evolution_thesis.md (15 markdown briefs — "Primary figure" line in each tells you what figure to show)
  • slides/v2/figure_manifest.md (all figure paths + ready-to-render R scripts)
  • slides/v2/SLIDES_v2_PLAN.md (consolidated plan — read for narrative ordering)
  • slides/20260204_Subtelomics_overview_EG.pdf (v1 deck, now tracked)
  • paper_prep/figures/ (existing figure assets)
  • paper_prep/figures/nj_tree_arms/nj_tree_annotated.{pdf,png} (slide 07 NJ tree)
  • /moosefs/guarracino/HPRCv2/PHR_III/similarity/ (off-tree figure alternates referenced in manifest)
  • slides/v2/_typst/ (if build-bog-v2-2 has finished, may contain rendered ER callout, clade callouts, etc. — REUSE if present)

WHEN A SLIDE HAS NO FIGURE (slide 01 title, slide 15 thesis closer per task spec — some slides may be text-only): produce a slide that shows the slide TITLE in large type centered, no figure box. Mark it with "[no figure — text-only slide]" in small footer text. Do NOT skip these slides; the deck must have all 15 in order so the review traversal matches the talking deck.

WHEN A SLIDE HAS A COMPOSITE (e.g. slide 07 heatmap + NJ tree side-by-side, slide 08 PCA chrom + superpop combined, slide 10 Hi-C bulk + Mantel exclusion): SPLIT into separate review slides — one full-bleed slide per panel. So the review deck may have MORE than 15 slides. Number them 07a, 07b for splits; this signals "these are the two halves of slide 07 in the talking deck". Erik can review each half at full size.

PIPELINE:

  1. Render any READY-TO-RENDER figures called out in figure_manifest.md whose source R script is embedded in the slide markdown — UNLESS slides/v2/_typst/ already contains them from build-bog-v2-2 (check first, reuse if present). Specifically: slide 03 ER-callout, slide 06 clade callouts, slide 09 clade legend, slide 12 stage trajectory inset.

  2. Use typst (~/.local/bin/typst v0.13.1). For this deck use a MINIMAL theme — just figure + small title + small footer. Recommended: a custom typst function rather than polylux's chrome-heavy templates. Aspect ratio 16:9; A4-landscape works too. Working dir: slides/v2-zoom/_typst/.

  3. Build slides/v2-zoom/_typst/zoom_deck.typ. Per-slide layout:

    • Top: tiny title bar (1 line, ~24pt, slide number + slide title from the slide MD)
    • Middle: figure, sized with #image(width: 95%, height: 88%, fit: "contain") — fit:contain preserves aspect ratio, fills as much as possible without distorting
    • Bottom: tiny footer (1 line, ~10pt grey) showing the source figure path For composite slides, split into multiple pages (07a / 07b). For text-only slides (01, 15 thesis), show the title in large type + a one-paragraph excerpt of the slide's "## Bullets" content centered.

  4. Compile: typst compile slides/v2-zoom/_typst/zoom_deck.typ slides/v2-zoom/BoG_2026_zoom.pdf. Log to slides/v2-zoom/_typst/render.log.

  5. MANDATORY VERIFICATION (lessons from render-manuscript-draft-2/4 figureless-PDF disasters): a. file slides/v2-zoom/BoG_2026_zoom.pdf → "PDF document" b. pdfinfo → at least 15 pages (more if composites are split into separate pages) c. pdfimages -list slides/v2-zoom/BoG_2026_zoom.pdf | wc -l → at least 12 image XObjects (count should be roughly equal to slide count minus text-only slides) d. Per-page screenshots: pdftoppm -r 100 slides/v2-zoom/BoG_2026_zoom.pdf slides/v2-zoom/_typst/page → commit those alongside e. ALSO: extract pdfimages -list to a file and CHECK that no image is rendered at <50% of slide area — i.e., every figure should occupy substantial slide real estate (this is the whole point of the deck). If any image is rendered at <50% area, fail with "slide N figure shrunk to X% of slide area" so we can fix the layout.

  6. Record slides/v2-zoom/BoG_2026_zoom.pdf as wg artifact. Single commit: "feat: BoG v2-zoom review deck — figures full-bleed for lead-author review pass".

ACCEPTANCE:

  • slides/v2-zoom/BoG_2026_zoom.pdf exists; ≥15 pages; ≥12 images; per-page screenshots in slides/v2-zoom/_typst/; figures fill ≥50% of each slide; wg artifact recorded.
  • If anything fails irrecoverably, wg fail with the SPECIFIC error per pipeline step.

DO NOT:

  • Modify any slide_NN_*.md (read-only)
  • Modify paper_prep/figures/ (read-only)
  • Modify the main deck slides/v2/BoG_2026.pdf
  • Touch files outside slides/v2-zoom/
  • Add bullets, speaker notes, or other talking-deck chrome to the rendered pages — this is FIGURES, BIG.
  • Use SBATCH (everything light: maybe-render R scripts + typst compile)

Depends on

Required by

Log