← sources

I asked Claude to design a new system of government that resists corruption and makes cheating expensive.

source: reddit (reddit:ClaudeAI)published 2026-10-13 16:03ingested 2026-10-14 02:37

retained text

The work (webpage + ideas)  is essentially all its own; it was given a primer, sort of, in the form of a previous essay on a similar topic, though that essay was also primarily written by Claude given a similar ask. It otherwise built the whole thing end-to-end in the desktop app with just the publishing login as assistance.

Since this is a Claude sub and not a political one, what do you think of Claude's work? ;) Does it tell us whether Claude will behave appropriately, if we later give it extended powers...?

[https://proposals.engineeringhumanity.org/answerable-state/](https://proposals.engineeringhumanity.org/answerable-state/)

\[edit\] Borrowed the summary (again from Claude) from a commenter below:

**The one-line version:** make every government decision a checkable promise with a named owner, an independent measure and a deadline, and have the consequences kick in on a fixed clock.

* **Commitments:** Every use of public power becomes a public contract with two named owners, a measure, a budget, a deadline, a pre-agreed consequence and an expiry date.
* **Ledger:** A public, tamper-proof record logs each commitment within 48 hours.
* **Measurement:** An independent Evidence Office measures results, so officials don't mark their own work.
* **Consequences:** A miss triggers a public flag on day 1, an explanation by day 14, a review by day 30 and a keep/change/stop decision by day 60.
* **Honesty:** Disclosed failure is never punished. Concealment and gaming the numbers go to a fast court, with a first verdict within 90 days.
* **Citizens' Jury:** A lottery-chosen jury can summon anyone in power to answer in public.
* **Transparency:** Scrutiny scales with power, so ordinary residents stay private.
* **AI:** An AI "Clerk" flags anomalies, but humans make every decision.
* **Safeguards:** Equal votes stay, and the whole framework expires unless voters renew it.

**Rollout:** Start with a volunteer shadow ledger run by newsrooms and universities, then pilot it in councils, then pass national law when a political window opens.

The page is upfront that every number is a proposed, untested starting value. It also flags its own biggest risks: voters may punish honest failures anyway, and measuring things can crowd out what can't be counted.

\[\\edit\]

  
\[edit2\]

Or, based on the feedback in the comments, now there is an interactive version, instead of the long wall of text. This is the original content, but reworked by Astra. Don't shoot me... 

  
[https://proposals.engineeringhumanity.org/answerable-state-interactive/](https://proposals.engineeringhumanity.org/answerable-state-interactive/)

\[edit2\]

document fetched 2026-10-14 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1x50mx9/i_asked_claude_to_design_a_new_system_of