Finalize the deck for delivery

Use when production-qa is the current gate or when a deck needs final delivery. Author to the engine's technical rules (from its adapter).

Engine

This gate renders and exports, so it needs the slide engine chosen at the prototype gate. Resolve it before QA:

• Read the engine from the status tracker via node .agents/skills/make-presentation/scripts/status.mjs next <work-root> (the JSON block in STATUS.md carries an engine field set at the prototype gate). Fall back to the ## Engine line in PRESENTATION-BRIEF.md if the tracker has no engine recorded.
• If no engine is recorded anywhere, ask the user now, one question, with a recommended default (see .agents/adapters/: open-slide, genoffice, marp, pptxgenjs; the bundled reference adapter is open-slide, or genoffice for native byte-preserving Office .pptx).
• Read .agents/adapters/<engine>/qa.md (or adapters/<engine>/qa.md) for that engine's preview surface, export command, export format(s), and inspection steps. The sections below are the engine-agnostic backbone; the adapter supplies the engine-specific surface and exports.

Choose context

• Use pipeline mode when presentation-work/<deck-id>/ exists. Validate its status with the shipped helper. Read the current mode from STATUS.md.
• Use standalone mode when pipeline artifacts and STATUS.md are absent.
• In standalone mode, derive a minimal QA context from the deck and notes: purpose, likely audience, material claims, main/backup range, timing assumptions, and unresolved risks.
• Do not create a pipeline work area solely for finalization; only create or update STATUS.md when a pipeline work area exists.

Align content

• Read repo instructions, the full deck, notes, and unresolved QA risks. In pipeline mode also read available PRESENTATION-BRIEF.md, EVIDENCE-LEDGER.md, STORY.md, STORYBOARD.md, GRILL-LOG.md, and SPEAKER-SCRIPT.md.
• Check visible claims, notes, timing, and delivery wording for consistency with the approved brief, evidence ledger, story, and script.
• Reject stale, unsupported, or contradictory claims instead of polishing around them.
• Strict 1:1 Parity: Confirm slide count equals note count, including backup pages (Slide Count === Speaker Note Count === PDF Page Count === PPTX Slide Count).
• Language Purity: For target-language presentations (e.g. 100% English summit), verify zero stray secondary-language characters across all slides and notes.
• Check factual consistency across slides, notes, references, and exports.
• Run node .agents/skills/make-presentation/scripts/unslop-check.mjs <slide-folder-or-notes> --strict. Fix every finding or record an exception with its reason in QA-REPORT.md. See ../make-presentation/references/unslop-copy.md.
• Confirm the main-talk range, backup range, total timing, and per-slide timing.
• Confirm one necessary job and one dominant message per slide.
• In research-enhanced mode, require a per-slide fact check plus a reference and Q&A backup plan before export approval.

Verify the preview & visual integrity

• Run the relevant repo checks and target build.
• Review the complete deck in the engine's live preview surface.
• Inspect full-page renders or a contact sheet, then inspect changed, dense, and image-heavy pages at full size.
• Reject verbose paragraphs and nested text tiers. Split the slide rather than shrink text or hide overflow.
• Aesthetic & Rendering Contract Verification:
  • Verify compliance with the declared Visual Direction & Rendering DNA in PRESENTATION-BRIEF.md.
  • Verify typography hierarchy (Hero titles reach 120px+, section headers 72px+, prominent display scale).
  • If Sticker Pop / Sketch archetype was declared, verify all cards carry 3px ink borders and hard offset shadows (8px 8px 0px), not generic diffuse soft shadows or 1px faint borders.
  • Verify visual diagram craft (SVGs match the aesthetic DNA, no raw generic boxes).
• Treat clipping, unreadable type, generic AI slop styling, inconsistent page grammar, or notes drift as failed QA.

Export and inspect

• After deck or exporter changes, produce a fresh export in each chosen delivery format; old downloads are not repaired.
• Unified Export: Run node .agents/skills/make-presentation/scripts/export-deck.mjs <deck-id> --format=pdf,pptx (or npx open-slide export <deck-id> --format=all) to automatically detect the engine and headlessly export fresh deliverables to exports/.
• Automated QA Audit Pipeline: Run node .agents/skills/make-presentation/scripts/qa-deck.mjs <deck-id> --export (or npx open-slide qa <deck-id> --export) to execute an automated audit verifying 1:1 strict parity (Slide Count === Note Count === PDF Page Count === PPTX Slide Count), copy slop, text density, and PPTX OOXML structure. Pass --record to automatically update STATUS.md with the verified deliverables upon passing.
• Follow adapters/<engine>/qa.md for the engine's export steps, archive integrity, geometry, and any native-application inspection.
• Visual-Faithful Verification: Verify that the exported PPTX preserves visual layout without text box re-wrapping or bounding-box drift when opened in Microsoft PowerPoint.
• Speaker Notes Verification: Confirm speaker notes are properly embedded in the exported PPTX notes pane for every slide.
• If the preview and any export disagree, the export is not complete.

Record and hand off

• Sort remaining findings into Act on, Consider, Noted, or Dismissed using ../make-presentation/references/review-buckets.md. Any Act on finding blocks the gate.
• Complete templates/QA-REPORT.md with evidence for every section and any remaining caveats.
• In pipeline mode, update the production-qa gate in STATUS.md through .agents/skills/make-presentation/scripts/status.mjs.
• Provide direct links to every exported file, page/note counts, timing, inspected pages, and explicit export verification evidence.