Style Review

Review and update documents to match the voice and standards defined in style_guide.md.

Quick Reference

Core Voice

• •Direct, calm, grounded
• •Confidence from specificity and outcomes, not tone
• •Plain language over jargon
• •No hype, theatrics, or "thought leadership" language

Writing Rules

1. •Lead with outcome and impact, not activity
2. •Describe systems, not tasks
3. •Mention scale, constraints, and context when relevant
4. •Prefer concrete verbs; avoid vague or inflated language
5. •Emphasize lifecycle ownership, not handoffs
6. •Treat evaluation as first-class
7. •Be honest about what failed or was hard
8. •Cut anything that doesn't change the reader's understanding

Review Checklist

Copy and complete this checklist:

Style Review:
- [ ] No hype or performative language
- [ ] Outcomes before activities
- [ ] Systems, not tasks
- [ ] Concrete verbs (built, shipped, designed, reduced)
- [ ] Scale/context included where relevant
- [ ] Honest about challenges
- [ ] Every sentence earns its place
- [ ] Sounds like Evan, not a marketing department

Language Examples

Phrases That Sound Right

• •"Design models, evaluation, and deployment together to move fast"
• •"Make it cheap to experiment and safe to fail"
• •"Unexpected behavior in prod is information"
• •"Simplify as much as possible, but no more"
• •"If you can't evaluate it, you don't understand it"

Red Flags to Rewrite

Before/After Examples

Before (activity-focused):

Worked on fraud detection systems and helped improve performance.

After (outcome-focused):

Built daily-refit fraud detection systems processing >1TB/day; prevented ~$14M in annual fraud losses.

Before (hype):

Passionate thought leader driving transformative AI initiatives across the enterprise.

After (grounded):

Build production ML systems; focus on evaluation, reliability, and real-world impact.

Review Process

1. •Read the document completely
2. •Run through the checklist above
3. •Flag sentences that violate the rules
4. •Rewrite flagged content using the patterns above
5. •Verify all claims have appropriate specificity (numbers, scale, context)
6. •Cut anything that doesn't change understanding
7. •Read aloud — if it sounds like it's trying to impress someone, rewrite it

Source of Truth

The canonical style guide is at style_guide.md in the project root. If this skill conflicts with that document, the style guide takes precedence.