Home / Ops / More in this area

Affirmology PDF Report - working notes (current state + where we can take it)

Updated Jun 27, 2026 · Affirmology_PDFReport_Notes_v1.md

Affirmology PDF Report - working notes (current state + where we can take it)

Logged 2026-06-27. This is the single notes file for the per-person Sacred Audio PDF report (the printed/downloadable "blueprint" that ships with every audio). Read this before touching the PDF. No em dashes anywhere in output.

What the PDF is

A 5-page, print-quality PDF that accompanies each person's Sacred Audio. It explains, in warm precise prose, what the audio is, the person's cosmic blueprint across the four systems, the full script with chart-source labels, and how the script was built from their chart. The gold-standard target is the Josh Parini reference PDF (see below).

How it is built (architecture)

• Library: WeasyPrint (HTML + CSS rendered to PDF). On macOS it needs Homebrew pango cairo gdk-pixbuf libffi; the pipeline sets DYLD_FALLBACK_LIBRARY_PATH to find them.
• Content is AI-filled: a single Claude call (model claude-haiku-4-5-20251001, max 8000 tokens) takes the birth data + chart JSON + the finished script, and returns ONE structured JSON object (cover paragraphs, four pillar key/value lists, throughlines paragraph, script sections with chart-source subheads, five build-note bullets, closing blurb). Override the model with env ANTHROPIC_PDF_MODEL (or ANTHROPIC_MODEL_PDF).
• Template substitution: the JSON is escaped and slotted into the HTML template via regex replacements ({{NAME}}, {{ASTROLOGY}}, {{SCRIPT_SECTIONS}}, etc.), then WeasyPrint writes the PDF.
• Facts are chart-only: the Haiku prompt forbids inventing placements; if a value is missing it is omitted, not fabricated.

Canonical files (edit THESE - the top-level affirmology-agent/ is a stale copy)

• Generator + Haiku prompt: affirmology-studio/affirmology-agent/src/affirmology/agents/sacred_audio_report.py
• System prompt (tone, JSON schema, content rules): lines ~44-126
• Model choice: line ~226 (claude-haiku-4-5-20251001)
• HTML build / substitution: lines ~154-214
• WeasyPrint render: line ~324
• HTML + CSS template: affirmology-studio/affirmology-agent/templates/sacred_audio_report.html (307 lines; all styling is the inline <style> block, lines 6-231)
• CLI trigger: affirmology-studio/affirmology-agent/src/affirmology/cli.py lines ~820-836 (fires on --audio when a mixed MP3 exists and --no-pdf is not set; failure is non-fatal, audio still ships)
• Pipeline (Studio): affirmology-studio/api/pipeline.py lines ~225-301 (PDF produced in the audio phase, progress marker "Sacred Audio Report PDF" at 90%)
• Storage: affirmology-studio/api/storage.py (local dir or R2 upload, content-type application/pdf)
• Jobs / email: affirmology-studio/api/jobs.py lines ~801-808 (save to R2) and ~818-890 (email includes "Open your PDF report" link)

The 5 pages (current structure, verified from the template)

1. Cover - AFFIRMOLOGY brandmark, title "Your Personal Sacred Audio", italic subtitle naming the four systems, "For {Name}", birth line, then three short sections: What This Audio Is / How To Use It / The Purpose.
2. Your Cosmic Blueprint - four pillars (Astrology; Gene Keys Activation Sequence; Human Design; Numerology), each a key/value list, then "The Throughlines" synthesis paragraph. 3-4. Your Sacred Audio · The Script - full script, broken into sections; each section has a Roman-numeral heading, an italic gold chart-source subhead (e.g. "Gene Key 50.2 (Life's Work) · Sun in Libra 12th House · Soul Urge 2"), then the verbatim first-person lines.
3. How This Was Built From Your Blueprint - five diamond bullets (the spine, body integration, closing anchor, walking anchors, voice and pacing), an italic centered closing blurb, and a heart flourish.

Current styling (the inline CSS, verified)

• Page: US Letter, margins 0.85in top / 0.9in sides / 0.95in bottom.
• Colors: ink #1a1410, soft gray #555048, burgundy #5b1c20 (headings), gold #b08a3e (accents), rule #c9a85a (dividers).
• Type: body Georgia 10.5pt / line-height 1.45; headings Helvetica Neue; cover title Georgia 30pt bold; subtitle + script subheads Georgia italic gold.
• Brandmark: gold, 0.32em letterspacing, gold underline.
• All system fonts (Georgia / Helvetica Neue). No embedded brand fonts, no images, no chart wheels, no cover art.

Gold standard reference

• Affirmology_JoshParini_SacredAudio_v1.pdf (project root, also in baselines/). The Haiku prompt is explicitly told to match its warmth and precision.

Known gaps / opportunities (candidate work for the Cowork session)

These are observations, NOT decisions. Jeff drives which to pursue: - No visuals. No chart wheel, no Human Design bodygraph, no cover image, no per-pillar glyphs. Pure text. Biggest visual-impact lever. - No embedded fonts / brand identity. Falls back to system serifs. A real brand typeface (embedded) would lift it from "document" to "artifact." - Fixed length / density. Same 5-page skeleton regardless of how rich the chart or how long the script. Long scripts can overflow pages 3-4 awkwardly. - Model is Haiku. Fine for structured extraction, but the prose synthesis (throughlines, build notes, cover paragraphs) is where a stronger model (Opus 4.8) could noticeably raise the writing quality. One call, so cost impact is small. - No cover art / dedication page. A first-impression page (image + name + a single line) before the brandmark cover could feel more like a gift. - Print vs screen. Currently print-first (Letter, serif). A screen/mobile-optimized variant (the app opens it in player.html) may want a different layout. - Charts data depth. The pillars show top-line placements. Could include more (all gates, defined channels, full numerology stack) or a fold-out "full data appendix" page.

Open questions for Jeff (decide in the Cowork session)

1. Is the goal a VISUAL redesign (add chart wheel / bodygraph / cover art / brand font), a CONTENT upgrade (deeper data, better prose, more pages), or both?
2. Keep WeasyPrint (HTML+CSS, easy to restyle, weak on complex graphics) or move to a layout engine that handles chart graphics better?
3. Print-first (Letter PDF) or screen-first (the in-app reader)? Or both as two templates?
4. Upgrade the fill model from Haiku to Opus for the prose sections?
5. What exactly about the current PDF feels short of the Josh Parini gold standard?

How to render a test PDF (verify any change)

From affirmology-studio/affirmology-agent/, the CLI --audio path produces the PDF next to the MP3 as {FirstLast}_Affirmology_Demo_v{N}_SacredAudioReport.pdf, with the intermediate HTML + Claude payload saved under out/{job}/backend/ for inspection. The fastest visual iteration loop is editing the template HTML and re-rendering that saved backend HTML through WeasyPrint directly, without re-calling Claude.