About
Policy Lab
Policy Lab is an experiment in transparent policy debate. Most policy proposals arrive fully formed, with their reasoning hidden and their tradeoffs obscured. Policy Lab runs the process in the open: you can see which constituencies were at the table, what each of them wanted, where they disagreed, and exactly what changed across rounds of negotiation before a bill was drafted.
Read the blog post explaining why I created it.
Each topic is handled by a pipeline of AI agents. The pipeline maps constituency grievances, identifies distinct policy areas, iterates reform proposals against objections, scores viability for each constituency across rounds, and finally drafts a bill from the policy areas that improve on the status quo. Policy areas that hit a value-conflict impasse stop iterating, but they are still evaluated during bill drafting instead of being papered over. What you read is a complete record of that process, not just the conclusion.
The goal is not to suggest AI should replace legislators. It is to show what a structured, auditable debate process looks like when the reasoning is preserved at every step, and to ask whether the result is more legible and more accountable than what conventional policy processes typically produce.
Design choices
Numbers and opinions
Policy Lab treats two kinds of claims differently.
For factual claims about the status quo, it uses web search to find real figures and cites the source by name and year. If a claim can't be sourced, it uses qualitative language instead. No statistics are invented.
For constituency opinions, there is no fact-checking. A constituency that believes immigration is the main driver of crime in their community gets that view represented accurately in the record, even if the statistics point elsewhere. This is intentional.
Policy is made by real voters with real beliefs, including wrong ones. A deliberation process that filters out inaccurate views before they enter the debate does not reflect how policy actually gets made. And a bill that ignores a major constituency's concern because it is factually disputed is a bill with a gap. The goal is to represent the political landscape as it is, not as we would prefer it to be.
A project by Luca Dellanna
Luca Dellanna is a writer and researcher focused on governance, institutional design, and complex systems. His work explores how rules and incentive structures shape collective outcomes, and what better-designed institutions could look like in practice.
More of his writing is at luca-dellanna.com.