## What is a personal benchmark?

A personal benchmark is a test built from your own work. Every time you correct an AI, that correction can become a case: the task, plus the yes-or-no checks that capture what you were looking for. Run a new model on your cases and you can see how often it clears your bar. Public benchmarks are like SAT scores. They can tell a 300 from a 1600, but they can't help you choose between models that all land around 1570. A personal benchmark is a reference check instead: it tells you which model is best for your work, whoever makes it, and when a new one is worth switching to.

---

## Notes

**Where it goes:** the empty blocks under "What is a personal benchmark?", before "We believe that there is great human work to do with AI." It is about 105 words, so the full strategy doc stays under the 500-word cap (the current draft is roughly 300).

**Choices I made**
- **The SAT / reference-check analogy comes from your pitch to Rick.** It's the clearest thing you said about why generic benchmarks fall short. I used your "1,600 vs. 300 vs. 1570" version. The Rick transcript is an unverified Granola Chat extract, so check that the numbers match what you remember.
- **The paragraph doesn't make the "replace you vs. work for you" contrast.** Your closing lines right after it make that point, and this paragraph would repeat them.
- **I used "cases" and "checks," not "KateBench."** Both terms appear in your draft and the Rick demo. KateBench is a product name and reads oddly as a definition. Swap it in if you want the tie-in.
- **The paragraph doesn't say "one click, automatically."** Your bullet above it already promises that, and the definition should stand on its own. Every Checks is internal-only today, and the correction-to-case step is still an ambition. The editorial review flagged that the page should separate today from next.

**Optional closing sentence if you want to mark that line:** "Today we build these by hand for ourselves; we want it to happen as you work."

**Shorter version (about 60 words):**
A personal benchmark is a test built from your own work. Your corrections become cases, each with yes-or-no checks for what you were looking for. Public benchmarks are like SAT scores, which don't help when every model scores about the same. A personal benchmark is a reference check. It shows which model is best for your work, whoever makes it, and when to switch.

**Assumptions:** I assumed the paragraph goes in the "Personal benchmarks" section of the Journal draft, in your voice ("we") and matching the plain, short-sentence style there. I didn't touch the Notion page. This file is the only output.