Suryansh Mishra
On a busy day in India, one event can produce a couple of hundred articles before evening. A train derails, a court rules, a state budget lands. National papers file, regional papers file, the wire agencies file, and dozens of sites carry the wire copy under their own mastheads. If you are the person who has to brief someone on it by morning, what you have is a pile, not a story.
I started building Alethman in September because I wanted to read that pile as one thing. Not a summary of it, and not a feed of the loudest headlines, but the whole story: who reported it, how many of them actually reported it rather than reprinting someone else, where the accounts agree, and where they split.
Most of what I build is plumbing for data: systems that take in a lot of messy input and have to be exact about what comes out. News turned out to be the messiest input I have worked with, and the one where being exact matters most.
The question a pile of articles cannot answer
The first thing that breaks when you read news at volume is counting. Forty articles about one event feel like forty sources. Often they are six. A wire story goes out once and is carried by thirty sites, sometimes with the agency credit stripped. If you count articles, you count the wire thirty times and the reporter who went to the scene once.
So the first question Alethman asks is not "how many articles" but "how many independent outlets". An outlet that reprinted a wire report is not an independent voice on that story. An outlet that did its own reporting is. That one distinction changes which stories look big, and it is the number we show beside every story.
The second thing that breaks is place. Most tools that map news put a dot in the middle of the country and stack the headlines on it. For India that is close to useless: a flood in Assam and a policy row in Delhi land on the same dot. We wanted coverage placed where it happened, in a state, a district or a city, read from the article itself, in English and in Indian languages.
Where coverage splits
The part I care about most is the split. Take a hypothetical protest. One group of outlets leads with the traffic and the arrests; another leads with the protesters' demands; a third barely covers it. Each article on its own may be accurate. Read together, they tell you something none of them says: that the coverage has divided, which outlets sit on each side, and whether the division follows language, region, ownership or timing.
Alethman shows that shape. For a story with enough independent voices, it reads what each article says about the central question and how it frames it, then groups the outlets that tell the story the same way. When two or more of those groups each have at least three independent outlets, that is a split, and the story page shows it with the outlets named.
What it does not do is tell you which side is right. That is a choice, not a gap we plan to fill.
What we learned from the first attempt
Alethman is not our first version of this idea. An earlier console we built set out to do more, including finding coordinated campaigns in the news. When we measured that against real coverage, we reached an uncomfortable conclusion: from news text alone, you cannot reliably tell a coordinated campaign from many outlets independently covering the same thing. Outlets share sources, press releases, wire copy and news cycles. Similarity is not coordination.
That finding shaped Alethman. Where the product talks about coordination at all, it reports a finding graded by the strength of the evidence, with the outlets and the evidence beside it and a line on what would change the reading. It never says a campaign is confirmed, and it does not attribute an action to anyone on its own: it says who the reporting names and how the outlets carried it. A false accusation is worse than a missed campaign, and a tool that hands out confident accusations will eventually hand out a wrong one.
The commitments that came with it
Building something that reads the news at scale means deciding early what it will refuse to do. Ours are on our About page, and they are enforced in the product rather than kept in a policy document:
- It reads open, lawful sources only.
- It does not profile private individuals.
- It does not write anyone's message for them.
- Every figure carries its source and how sure we are of it.
- A person decides; the interface never implies the machine did.
- Every query in a client workspace is logged against the person who ran it, refusals included.
These sound like restrictions. In practice they are what makes the output usable. A number you cannot trace back to articles is a number nobody in a serious meeting should repeat.
Who it is for
The people we built this for read news for a living without being journalists: policy teams, communications and corporate affairs teams, analysts who brief leadership. Their problem is not access to news. They have too much of it. Their problem is time and confidence: knowing, by nine in the morning, which things moved, how widely each was actually reported, and whether the coverage is telling one story or two.
That is what the dashboard gives them, and How it works walks through it. A team picks the states, sectors and countries it cares about. The engine reads all day, groups articles into stories, places them on the map and gives each sector a reading. Every number links back to the articles it came from.
Why India first
India is where the gap is widest. Coverage is spread across hundreds of outlets in many languages, the same place has three spellings, and much of what matters happens in state capitals that national coverage reaches late. Getting India right forced us to solve the hard parts early: Devanagari text, districts with the same name in two states, place names that are also ordinary words. A workspace can now follow other countries too, the United States among them, and the same engine reads them.
What comes next on this blog
I will write here about how the engine works and why it is built the way it is: how it decides that two articles describe the same event, why we count outlets instead of articles, and what we got wrong on the way. If you want to see it on your own questions, get in touch.
Posts on this blog are the signed views of their authors. Alethman News reports the news; this blog argues about it.
