# Build Your Own Gate

Take this home. Everything below is real, it's the actual system behind "139 things were true this morning, 5 of them needed me directly."

## The idea

Everything that could reach you (Slack, email, tickets, meeting notes, doc comments) goes through one gate before anything happens automatically. The gate asks three questions, in this exact order, before anything else runs:

**01. REVERSIBLE?**
If this goes wrong, can it be undone in a minute, or does someone have to apologise?

**02. MINE?**
Does it touch a person's name, a budget, or a date that only I can move?

**03. NOW?**
Does acting an hour from now cost anything real, or does it only feel urgent?

The answers sort everything into exactly three lanes. No fourth lane, no "it depends" bucket.

- **AUTOMATE** (reversible · not mine · can wait): it just happens, and you find out afterward.
- **DRAFT & WAIT** (reversible · might be mine): the work gets done, the decision doesn't. You read it, fix what's wrong, send.
- **ESCALATE** (not reversible · mine · now): no draft, no softening, it comes straight to you with the thread attached and nothing pre-decided.

## The actual policy, verbatim

This is the real permission-tier language the system runs on, copy it into your own setup if it's useful:

> **Prohibited (never do it, tell the person to do it themselves):** entering financial credentials or passwords, creating accounts, permanently deleting data, executing financial trades, bypassing CAPTCHAs, modifying security settings.
>
> **Explicit permission required (ask, wait for yes, then act):** sending any message on someone's behalf, publishing or modifying public content, purchasing with a saved payment method, accepting terms or granting OAuth access, changing account settings, entering personal data into a form.
>
> **Regular (just do it):** everything else.

## Build one this week

1. **Name one entry point.** Six channels can feed it. Only one queue exists.
2. **Write the checks first.** Decide the questions before you see a single case. Reversible, mine, now, whatever your version is, decide it in the abstract, not case by case.
3. **Give each answer a lane.** Automate, draft, escalate. No fourth lane, no "it depends" bucket, that's where judgment quietly leaks out.
4. **Log every route.** If you can't replay why something was automated, you have a habit, not a system.
5. **Let it say it doesn't know.** "I can't route this yet" is a valid output. A confident wrong route costs more than a pause.

## What it actually costs

About $0.50 to $0.65 a day to run. A cheap model handles the routine syncs; a better model is reserved for the one job that actually needs judgment, deciding what needs a person.

## Where it still fails

Being honest about this matters as much as the system working: on July 30th, a sync reported success and wrote nothing, silently. A manual audit caught it, not the system. It's fixed now, every sync records a before-and-after timestamp and fails loudly instead of reporting a clean run, but it failed quietly once, and that's worth saying out loud.

---
Built by Josh Stewart, PulsePoint. Full write-up and prompts: the-gate.stewart-m-joshua.workers.dev