Launched this week
The Bias
The synthesis engine for multi-perspective news
64 followers
The synthesis engine for multi-perspective news
64 followers
Most big news stories don’t have one version — they have many. The Bias is a perspective synthesis engine that reconstructs coverage across outlets into one structured read, showing what’s corroborated, what’s contested, what’s still unclear, and how framing differs. Built for clarity without tab-hopping. PH feedback wanted: If you were improving this tomorrow, what’s the first thing you’d change in the reading experience?
The Bias
The corroborated / contested / unclear distinction is the feature that matters most here. Most news aggregators just show you more sources, which doesn't solve the problem. Knowing which claims have cross-source agreement versus which ones are disputed versus which ones nobody actually knows yet is a fundamentally different reading experience.
The closest comparison is Ground News, which does a solid job of showing left/center/right framing differences. But it still sends you to individual articles rather than synthesizing them. AllSides does the same, side by side rather than structured. The Bias is trying to solve a harder problem: not just showing you where sources differ but telling you what they actually agree and disagree on. That's a much more useful output if you pull it off consistently.
To answer your question on what I'd change first: I'd want a way to track how a story's corroboration status evolves over time. Something that looks contested on day one often gets clarified within 48 hours, and being able to see that arc would add a lot of value. Congrats on the launch! 📰
The Bias
@joao_seabra This is such a thoughtful take — thank you. You articulated the core problem better than I usually do.
You’re exactly right: just adding more sources doesn’t reduce uncertainty. The goal is to shift from “here are 12 takes” to “here’s what’s actually agreed on, what’s disputed, and what’s still unclear.” That synthesis layer is the hard part, but it’s also where the real value is.
We’re actively working on the temporal evolution piece now. The current approach is to update articles as new facts emerge, so if something is contested on day one but clarified 48 hours later, the status changes accordingly.
I’m curious how you’d want to see that expressed. Is a living, updated article the right format, or would something like a visible status timeline make the progression clearer?
@charlie_ehlen A visible status timeline would be more powerful, I think. A living article that silently updates loses the narrative of how the story developed, which is actually part of the value. Seeing "contested on day 1, clarified on day 3, one claim still unresolved on day 7" tells you something important about the story itself, not just its current state. It also builds trust in the synthesis; you can trace why the status changed and what new information triggered it. A changelog approach, even a lightweight one, would make the platform feel more like a source of record than a snapshot. Any, congrats on this - very much needed in this world!
The Bias
@joao_seabra You’re right that the arc of corroboration is part of the story itself, and could really help with trust. I'll have a think about where best to implement it. Thank you!
That’s awesome! How many sources are you tracking? We’re doing similar classification at MediaThrive. If you want, ping me and we can exchange ideas. 🙂
The Bias
@pavelbg Thanks! Right now I’m we’re tracking about 600 sources globally, but looking to expand to more as soon as we can.
That’s great you’re doing something similar at MediaThrive — I’d definitely be up for comparing notes. What’s the main idea behind MediaThrive? I’ll ping you so we can swap ideas. 🙂
@charlie_ehlen Hi Charlie, hope your product launch is successful! We’re automating the full publisher workflow, with a main focus on generating audio and video from a text article. One of our products is global news monitoring (we have access to ~1M articles per day), and to avoid cluttering our customers’ feeds we do news clustering based on the intent of the story. This way, we group coverage of the same event and include different viewpoints from different media outlets. But as I said, our solution is B2B for publishers — so I hope your product reaches more people. We need more truth in the world :)
Love the concept, feels very timely. How do you decide which sources to include, and how do you prevent your own synthesis from introducing a new layer of bias?
I get nervous when skills turn into a grab-bag of styles, agents start feeling random. How are you scoring Skills Refiner's benchmark, by running a small task suite or just grading the refactored text, including translate/refine cases? That's what makes a score like this trustworthy.
love the concept of synthesizing different perspectives! quick question though: how real-time is the synthesis? does it constantly update as a breaking news story develops throughout the day?