# You told the agent to run `pnpm vitest run` twice this week.

_The Trace Letter, week of 2026-10-26._

17 sessions across 4 projects. 28 prompts, 5 corrections.

## What you worked on

**tidepool**

- Units setting (2 sessions)
- Cache tide predictions offline
- Widget for the next high tide
- Station search

**ledger**

- Recurring transactions detector
- Dark mode for charts
- Export a year as a tax summary
- Search transactions by amount
- Keyboard shortcuts

**fieldnotes**

- Fix broken RSS dates
- OG images (2 sessions)
- Pull quotes in posts

**relay**

- Rate limit deliveries per endpoint, 10/s default
- Filter the delivery log by status and endpoint (2 sessions)

## Where it went sideways

- Export a year as a tax summary (ledger): 1 correction
- Cache tide predictions offline (tidepool): 1 correction
- Pull quotes in posts (fieldnotes): 1 correction

## The habit that cost the most

> no watch mode. use `pnpm vitest run`

> no, `pnpm vitest run`

Trace drafted this for `/Users/you/code/ledger/CLAUDE.md`:

    Run tests with `pnpm vitest run`.

## What held

- Use pnpm for installs and scripts (`pnpm install`, `pnpm run <script>`). Never npm.

---

Counted from your local session files. Nothing here was written by a model.
