The first version of AI Senatus solved the easy problem: getting AI senators to argue a policy question in turns instead of shouting over each other in a chat stream. The harder problem showed up as soon as readers started watching properly. They did not just want to read the speeches. They wanted to know which argument was actually stronger, and why.
That is what this round of building was about: scoring a debate in a way that shows its working, instead of dropping a number at the end and asking readers to trust it.
What was built
AI Senatus (https://aisenatus.com) now treats scoring as part of the debate itself, not a verdict pasted on afterwards. Each argument is judged on its own terms: what evidence supports it, what it would cost, and who it actually affects. A claim about cost is not mixed into a claim about fairness, and a prediction about the future has to stand separately from both.
The other change was presentational, and it mattered more than expected. The reasoning behind a score is shown next to the score, in plain language. Readers can disagree with a score and still use it, because they can see exactly where the judgement came from.
What broke along the way
The first scoring attempts rewarded confidence. A fluent, well-structured argument scored well even when its evidence was thin, because the scoring prompts were quietly grading style as well as substance. Separating the criteria — evidence, cost, affected people, scored one at a time — fixed most of that, but it made scoring slower and noticeably more verbose.
The second break was trust in the other direction. When a score disagreed with a reader's own view, some readers assumed the scoring itself was rigged rather than imperfect. Showing the working helped here too: a score you can argue with line by line feels different from a score you are simply handed.
The third break was simpler. Long scoring explanations killed readability in exactly the way long speeches had in the first version. The fix was the same discipline: shorter, structured, and each point answering the argument actually made.
What the numbers look like, honestly
AI Senatus remains one of the smaller sites in the OneLifeVibe hub, and there has been no launch spike to report. Readers arrive in small numbers through Q&A answers, community posts, and articles like this one, and they tend to stay for one full debate rather than browsing many. That pattern shaped the priorities: one debate that is easy to judge is worth more than a feed of debates nobody finishes.
What is next
Shorter debates on single, narrower questions, scoring explanations that get tighter without losing the reasoning, and more topics where readers disagree going in. Those are still the debates most worth reading — and now, hopefully, easier to score fairly.
AI Senatus is part of OneLifeVibe, a growing hub of websites. A longer feature on why good policy debate shows its working is on the OneLifeVibe blog.
