A Dashboard for Every Report
Every report we publish comes with its own interactive dashboard and its own permanent link, so a claim about how a story spread can be checked instead of taken on trust. One dashboard per report, not one for the whole project.
What it shows
When several channels cover the same event, we group their posts into a cluster and keep measuring that cluster for an hour. The dashboard is how that measurement is read.
Publication chronology
Which channel published first, and how many minutes behind each of the others followed. This is the view that separates a story breaking naturally from one appearing near-simultaneously across channels.
View and reaction dynamics
Views and reactions sampled four times across the first hour, per channel. A post that arrives with its audience already attached looks different from one that accumulates.
Reaction mix per channel
Which emoji the audience of each channel actually used. The same event can land as anger in one audience and approval in another, and the breakdown shows that instead of averaging it away.
Comment sentiment over time
Sentiment of the comment threads as they develop, next to the reaction curve. Divergence between the two, approving reactions above hostile comments, is often the more interesting signal.
Linking to a specific report
Each cluster has a stable identifier, and the dashboard accepts it as a query parameter. That means any individual report can be cited directly:
https://aisbergpublicorganization-dashboard.hf.space/?cluster=cl_1787693694324_0The identifier is the same cluster_id used in the open dataset, so a dashboard view and the underlying clusters/cl_*.json file in the dataset always refer to the same thing. Cite the file in a paper, link the dashboard for readers.
Reports published in our Telegram channel link straight through, and open as a Telegram Mini App on mobile rather than kicking you out to a browser.
Good to know
- The dashboard is hosted on Hugging Face Spaces. If it has been idle it may take a few seconds to wake up.
- It reads from the public dataset rather than a private database, so everything it draws, you can download.
- The interface is in Ukrainian, since that's the language of the material it describes.
Where the numbers come from
Nothing in the dashboard is modelled or estimated. It is measurement. Views, reactions, comment counts and timestamps are captured directly from the public channels at fixed intervals after publication. Sentiment is the one derived layer, and it is computed, not curated.
If you want the detail of how collection and analysis work, including where the method is weak, that's on the methodology page.
News gets cleaner when somebody checks it
News clusters, fact-check verdicts and a link to the dashboard go out on the Telegram channel every hour. Reading it is free and needs no account. And if you want more of this work to happen, you can join it.
Subscribe to the channel Become a volunteer