# One-Shot Prompt

**Topic**: AI benchmark catalog design and what one-shot routes reveal about model capability
**Theme**: Crest Alpha Origin
**Generated**: 2026-04-23
**Model**: Crest Alpha Origin

## Prompt

Write a complete Node.js ES module `generate.mjs` using `pptxgenjs` that generates a professional 15-slide presentation titled **"Benchmarking the Builders"** about AI benchmark catalog design and what one-shot routes reveal about model capability.

The deck is for the Crest Alpha Origin oneshot catalog and should visually feel branded for **Crest Alpha Origin**: sharp hierarchy, dark-to-light contrast, editorial polish, and programmatic shapes and charts only. Do not use any external images, templates, screenshots, URLs as images, or base64 images. Build every slide from scratch with shapes, text, tables, and charts.

The deck must be clearly high-end and intentional, not tutorial-like. Use realistic but clearly illustrative benchmark data. Label the dataset as illustrative wherever appropriate. Include speaker notes on every slide with 2-3 talking points, a source cue noting that the benchmark figures are illustrative, and a transition to the next slide.

### Deck-wide design system

- Format: `LAYOUT_16x9`
- Fonts: sans-serif only, safe default `Arial`
- Theme name: `Crest Alpha Origin`
- Primary dark: `07111F`
- Secondary navy: `16345F`
- Signal cyan: `56E0FF`
- Origin gold: `F5B84B`
- Coral risk: `FF6B6B`
- Mint positive: `52C7A5`
- Slate text: `24364D`
- Muted slate: `7B8CA5`
- Light background: `F5F8FC`
- Panel fill: `EAF1FB`
- White: `FFFFFF`

### Layout principles

- Use a dark cinematic title slide and closing slide
- Use light editorial content slides with dark titles and clear accent bars
- Maintain generous margins around 0.45-0.60 inches
- Every slide must include a meaningful visual element
- Keep on-slide text terse; detail belongs in speaker notes
- Include subtle footer text on content slides: `Illustrative benchmark sample for https://crest-alpha-origin.pages.dev`
- Use rounded rectangles for cards and callouts
- Use charts with the deck palette only
- Use no more than three text sizes per slide

### Data model

Treat all metrics as **illustrative benchmark sample data** for a one-shot route catalog.

Use these data points:

1. Catalog overview metrics
   - Routes: `15`
   - Artifact types: `5`
   - Observable signals per run: `68`
   - Single-attempt policy: `1 prompt, 1 pass, no retries`

2. Slide 5 bar chart: first-pass completion rate by route
   - Website: `97`
   - Dashboard: `91`
   - Data story: `88`
   - PowerPoint: `86`
   - Maproom: `84`
   - Physics sandbox: `79`

3. Slide 6 doughnut chart: evidence mix
   - Functional correctness: `31`
   - Visual judgement: `24`
   - Instruction fidelity: `18`
   - Recovery behaviour: `15`
   - Documentation hygiene: `12`

4. Slide 7 timeline milestones
   - 2023: Prompt scoring dominated early evals
   - 2024: Artifact routes exposed execution gaps
   - 2025: Multi-route catalogs created richer capability signatures
   - Early 2026: Speaker notes and delivery assets improved review depth
   - Next: Route-weighted scoring and replay harnesses

5. Slide 8 comparison table rows
   - Static benchmark
   - Long-form writeup
   - One-shot route
   - Tool-assisted production run

   Use attributes:
   - Artifact proof
   - Failure clarity
   - Polish signal
   - Reproducibility
   - Typical blind spot

6. Slide 9 line chart: signal lift over benchmark waves
   - Periods: `Q2 2025`, `Q3 2025`, `Q4 2025`, `Q1 2026`, `Q2 2026`
   - Prompt adherence: `71, 75, 80, 84, 88`
   - Artifact completeness: `64, 68, 74, 79, 83`
   - Presentation polish: `58, 62, 69, 76, 82`

7. Slide 10 case study: Crest Alpha Origin sprint
   - Completed routes: `12`
   - Partial routes: `2`
   - Failed routes: `1`
   - Standout strength: information architecture
   - Most revealing miss: simulation tuning under tight constraints

8. Slide 11 risk matrix items
   - Route overfitting
   - Style bias
   - Hidden scaffolds
   - Scoring drift
   - Sparse failure taxonomy

9. Slide 13 projection bars
   - 2026 routes instrumented: `15`
   - 2027 routes instrumented: `24`
   - 2028 routes instrumented: `36`
   - 2026 automated checks: `68`
   - 2027 automated checks: `105`
   - 2028 automated checks: `160`

### Required 15-slide narrative

1. **Title slide**
   - Title: `Benchmarking the Builders`
   - Subtitle: `What one-shot routes reveal about model capability`
   - Small line: `Crest Alpha Origin | Illustrative benchmark design deck | 23 April 2026`
   - Add geometric accents and a strong dark background

2. **Agenda / overview**
   - Show five agenda blocks:
     - Why route catalogs matter
     - Where signal comes from
     - How route evidence differs from classic evals
     - What an illustrative Crest Alpha Origin sprint shows
     - How the catalog should evolve
   - Add a right-side vertical “signal stack” visual

3. **Context / why this matters**
   - Title should be active, such as `Benchmarks need artefacts, not just answers`
   - Left side: three short bullets on why text-only evals miss delivery quality
   - Right side: callout card showing `15 routes x 5 artifact types`
   - Add a thin evidence ladder or layered shape motif

4. **Key data point**
   - Make `68 observable signals per run` the hero number
   - Supporting caption: routes expose planning, execution, taste, and finish quality in one pass
   - Add four small chips for `planning`, `execution`, `presentation`, `recovery`

5. **Market / landscape overview**
   - Use a bar or column chart for first-pass completion rate by route
   - Chart title: `Different routes stress different parts of the model`
   - Add a short side note that websites and dashboards reward broad assembly, while physics and maproom routes punish brittle reasoning

6. **Breakdown / categories**
   - Use a doughnut chart for evidence mix
   - Centre label: `100% observable signal`
   - Add 2-3 short takeaways beside the chart

7. **Timeline / history**
   - Create a horizontal timeline using circles, lines, and milestone cards
   - Use the milestone text provided above
   - Make the final milestone visually brighter to imply momentum

8. **Comparison table**
   - Build a polished table comparing the four evaluation modes across the five attributes
   - Highlight `One-shot route` as the strongest balance of artefact proof and failure clarity
   - Keep the table highly readable with alternating fills and strong headers

9. **Trend analysis**
   - Use a multi-series line chart for the three trend lines
   - Title: `Catalog design is lifting the quality of observable output`
   - Add a short caption that polish is catching up as routes become more delivery-specific

10. **Case study / example**
   - Make this a Crest Alpha Origin slide
   - Show three stat cards for `12 complete`, `2 partial`, `1 failed`
   - Add two larger cards:
     - `Strongest route signal`: structured information architecture across presentation-heavy routes
     - `Most revealing miss`: simulations break when hidden assumptions are not surfaced early
   - Include a small badge: `Illustrative sprint snapshot`

11. **Challenges & risks**
   - Create a risk matrix with labelled points placed by likelihood and impact
   - Use coral, gold, and cyan markers
   - Include a short legend and a short sentence on why benchmark governance matters

12. **Opportunities / solutions**
   - Four opportunity cards:
     - Route-weighted scoring
     - Notes-aware review
     - Artifact diffing
     - Failure manifests
   - Each card should have a mini tag and one short line of benefit

13. **Future outlook**
   - Show a projection layout comparing route count and automated checks for 2026, 2027, and 2028
   - Use vertical bars or a forward staircase composition
   - Add a bold forecast statement such as `Richer routes will compress the gap between demo and dependable delivery`

14. **Key takeaways**
   - Four numbered takeaway cards:
     - Routes reveal capability in context
     - Artefacts make failure legible
     - Presentation routes expose taste and finish
     - Catalog design becomes a product, not just a score
   - Each takeaway gets a numbered circle and a one-line support phrase

15. **Thank you / Q&A**
   - Dark closing slide matching the title slide
   - Headline: `Build routes people can inspect`
   - Subline: `Questions, critiques, and replay ideas welcome`
   - Small footer: `https://crest-alpha-origin.pages.dev`
   - Add subtle geometric branding and a final accent rule

### Speaker notes requirements

- Every slide must call `slide.addNotes(...)`
- Notes should include:
  - what to say
  - source cue: `Illustrative benchmark sample prepared for Crest Alpha Origin`
  - transition to the next slide
- Use concise bullet-style note text

### Technical requirements

- Use ES modules and `import pptxgen from "pptxgenjs"`
- Script must be runnable with `node generate.mjs`
- Output filename must be `presentation.pptx`
- Use helper functions for repeated design primitives
- Keep all data constants near the top of the file
- Use only programmatic shapes, text, charts, lines, and tables
- No external images
- No code placeholders or omitted slides

## Notes

- Designed as a download-first oneshot catalog route for Crest Alpha Origin
- Data is intentionally realistic-looking but explicitly illustrative
- How to run: `node generate.mjs`
- Output filename: `presentation.pptx`