Decision Receipts

Public decisions deserve receipts.
What portland.gov and the city's mailed decision notices would look like if algorithmic involvement were disclosed natively — on the page, in plain language, with the full trace attached. Every page here is real Portland markup; every AI system is hypothetical and labeled.
The 30-second pitch When a city decision touches your life — a permit, a citation — and an algorithm was involved, you find out… never. Portland passed the rule this March: AI use must be disclosed, humans must review. But nobody's designed what a resident actually receives. So we built it. Every algorithmically-touched decision answers six questions in plain language, with the full machine-readable trace attached — built into real portland.gov pages and a real mailed decision notice, so adopting it means adding a section, not a system. And it's live: you're looking at it.

Every receipt answers six questions

  1. What happened to me?
  2. Was an automated system involved — and which one?
  3. What did it do — and what was it not allowed to do?
  4. What information did it use, and where did it come from?
  5. Who had final authority?
  6. How do I challenge or correct it?

Why this exists

The plain-language case, in three steps.

The rule is already here

Portland's Administrative Rule BTS-4.04, effective March 6, 2026, requires public disclosure of AI use, a public inventory of approved systems, and human review proportionate to risk. Automated decision-making without appropriate human review, where outcomes could significantly affect someone, is a prohibited use. The city has the rule.

The disclosure isn't

Governments publish registries about systems — Amsterdam, the UK, New York, Portland's own inventory. But when a decision lands on a person, nothing in the notice says what a machine did, on whose authority, or how to challenge it. France mandates a one-line version; U.S. credit denials have disclosed the scorer since 1970. Nobody sends the full account.

So we built the instrument

A decision receipt: six plain-language answers on the notice a resident already gets, backed by a machine-readable trace — on the web and in the mailed document. Built in Portland's own page and document language, so adopting it means adding a section, not a system. It makes an automated system's behavior legible to someone who isn't an engineer: what it did, on whose authority, and how to stop it.

The tour

Follow the numbers — this is the same path as the live demo.

1

The feed

Portland's real public-notices page. One entry is different — find the marker: ● Automated system disclosure

Open the notices feed →
2

The receipt

Land-use decision LU 26-880412 AD. The six questions answered in plain language, the city's own High risk tier, and the full timestamped trace — every step marked system or human.

Open the decision notice →
3

The correction

City records said the house was built in 1923. It was 1993. Because the owner could see the fact the system relied on, they could correct it — amended decision, condition dropped. Transparency with recourse.

Open the amended decision →
4

Today's example real program

Portland's speed safety cameras are real, present-day automated enforcement — a machine measures, a sworn officer decides, a citation lands in your mailbox. This is that citation as a receipt. Only the instance is hypothetical; the automation is not.

Open the citation receipt →
5

The system record

Behind every receipt: who runs the system, what it may and may not do, what data it touches, what human review is required — and every decision it touched.

Open the system record →
6

The inventory

Portland's real AI inventory has one entry today. Registries in Amsterdam, the UK, and NYC tell you about systems. None tell you about your case. Receipts are the citizen-facing end of an inventory.

Open the AI inventory →
7

The mail

The same case as it arrives in a mailbox: an 8-page decision notice with an Automated System Disclosure section and Exhibit H — the trace, riding the exhibits mechanism Portland's notices already use.

Open the PDF →
8

The spine

One trace, many renderers: the web receipt, the trace view, and the mailed Exhibit H all render from one JSON file conforming to decision-trace.schema.json. A hypothetical trace cannot pass as production — status is a required enum.

Schema & source on GitHub →

Three disclosure levels, one source of truth

L1 — the boxSix answers in plain language on the notice itself. Every load-bearing term defined where it's used.
L2 — factors & provenanceWhat was considered and what was not, how much each factor mattered, which city system each fact came from, and what the human reviewer did.
L3 — the traceThe full timestamped event log — on the web as a trace view, in the mail as Exhibit H.

Declared plainly: what's real, what's not

The judges are scoring honest demos. So is the public. Here's the full split.

Real
  • The portland.gov markup, styling, and page structure — cloned, not imitated
  • Every quote from Administrative Rule BTS-4.04, and its March 6, 2026 effective date
  • The land-use notice format, fields, and appeal mechanics ($250 fee, Hearings Officer, appeal windows)
  • Portland's AI inventory and its one real entry — Microsoft Copilot Chat Basic
  • Portland's photo enforcement program and its sworn-officer review requirement
  • Portland's seven-year policy lineage, from the 2019 privacy principles to the 2025 algorithmic rent-setting ban
Hypothetical — labeled on every surface
  • All three AI systems, including Development Review Triage — Portland does not use AI to review land-use cases
  • Case LU 26-880412 AD, citation TSC 26-004417, and every trace in the repo ("status": "hypothetical" is required by the schema)
  • Every person named — applicants, owners, the planner, the reviewing officer
  • Every statistic on the system record page
  • The disclosure format itself — that's the proposal

Fair questions

“Isn't this just a mockup?” The standard is the product: a JSON Schema under which a hypothetical trace cannot pass as production, two conforming traces, and three renderers — web receipt, trace view, mailed Exhibit H — reading the same file. The cloned pages are the delivery vehicle.
“Would the city actually do this?” It rides mechanisms Portland already has: the notices it already mails, the exhibits those notices already attach, the inventory it already publishes, and a rule in effect since March 2026. Adoption means adding a section, not procuring a system.
“What does this have to do with AI governance?” Governance a resident can't see isn't legible governance. The six questions answer the governance prompt for one person's case: what the system did, on whose authority, and how to stop it — readable by someone who isn't an engineer.
“Why build it before the AI arrives?” Retrofitted transparency is how you get registries nobody reads. And machine-initiated decisions already reach Portland mailboxes — the speed-camera citation on this tour is automated enforcement that exists today.
“Automated decision-making without an appropriate level of human review, proportionate to the risk, when outcomes could significantly affect an individual's health, safety, rights, or financial well-being” is a prohibited use. — City of Portland Administrative Rule BTS-4.04, “Artificial Intelligence Use and Governance,” effective March 6, 2026. Portland has the rule; receipts are the missing instrument.