AI reads and recommends. People decide.

Not as a policy statement, as a property of the code. The service that records an AI output refuses at runtime to store one that claims to perform a prohibited action, and high-risk recommendations cannot take effect without a recorded human review.

AI outputs recorded

Every one, with model and prompt version

Human reviews

Overrule rate

A rate near zero means review is not real

High-risk, unreviewed

Cannot take effect

What it does

Classify an issueSuggest which category a report belongs to, with a confidence figure and the exact rules that fired.
Detect duplicatesSpot that a new report is the same problem as an existing one, so the real scale becomes visible.
Estimate urgencyWeight signals like reported injury, affected numbers and how long it has gone on.
Identify missing informationTell the reporter what a volunteer will need that they did not include.
Find clustersPoint out constituencies with rising issue density and no volunteer coverage.
Flag anomalies

What it is forbidden to do

These are refused by assertNotProhibited() before any output is stored. Not discouraged, refused.

Appoint a leader
Remove a leader
Declare anyone guilty of anything
Select an election candidate
Disqualify a citizen

Why we publish the overrule rate

Human review becomes a rubber stamp very easily. The honest way to expose that is to publish how often reviewers actually disagreed with the recommendation. If that number sits near zero for long, the review is theatre, and we have committed in advance to saying so rather than presenting it as accuracy.

What happens to your data

Does anything you write leave this platform?: No. Nothing at all.
How recommendations are produced: Rules written down in this codebase, not a language model.

Deterministic rules. Every recommendation lists the exact rules that produced it. No data leaves this deployment.

Where this is written down

Section 33 of the Party Constitution sets out the rules. The manifesto item on automated decisions about citizens applies the same standard to government, which is only credible because we apply it to ourselves first.