Has the story changed?
Not whether people are talking, but whether what they are saying is different from what they were saying yesterday. A market everyone has been bullish about for a month is not news. The week that flips is.
Engine 2 of 2This one reads people. The other reads prices.
It reads the room
Not by counting how often a coin gets mentioned. By noticing when the story people tell about it changes, how fast that is spreading, and whether the accounts spreading it have ever been early before.
It reads in seven languages, because a narrative can move through Korean and Turkish communities days before it turns up anywhere in English. A system that only reads English is structurally late.
- Languages
- 7
- not English only
- Sources
- 4 kinds
- social, forums, chat, news
- Output
- A score
- never a write-up
- Included from
- Pro
- Starter is prices only
The four questions
Counting mentions tells you almost nothing
Every tool in this category counts how often something gets mentioned and calls it sentiment. Mentions go up when a thing is popular and they go up when it is collapsing. These are the four questions actually worth asking instead.
How fast is it moving?
A mood that shifts over three weeks and a mood that shifts over three hours are not the same signal, even if they end up in the same place. Speed is most of what separates a real turn from ordinary noise.
Is it spreading or stuck?
One loud corner of one platform is not a narrative. The same idea turning up independently in places that do not usually talk to each other is a different thing entirely, and much harder to manufacture.
Who is actually saying it?
Some accounts have historically been early. Most have historically been late, and a few are being paid. An engine that weighs all three the same is easy to fool, and fooling it is cheap.
Why seven languages
English is usually the last stop
Crypto does not happen in English. A story can run through Mandarin, Korean, Russian and Turkish communities for days before an English-language account picks it up and presents it as news.
A tool that only reads English will find that story eventually. It will find it at the point where it has already been priced in, which is the least useful moment to find anything. Reading across languages was a requirement here rather than a feature added later, and it is one of the reasons the model had to stay small enough to run on an ordinary graphics card.
What actually comes out
A score and the parts it was built from. Never a summary, never a paragraph explaining the mood, never anything anyone has to read and interpret.
That is not a limitation. A number can be weighed against the other engine's number, checked later against what the market actually did, and written to a ledger where anyone can go back and read it. A written summary can do none of that, and it invites you to trust the writing rather than the evidence.
market BTC/USDT
tone +0.44
shift rising, over about 6 hours
spread 5 of 7 languages, 4 source types
flagged 11 posts discarded as coordinated
model version recorded with the scoreThe model version travels with every score on purpose. When the model changes, old scores stay attached to the model that produced them, so nobody can quietly upgrade the engine and take credit for calls an older one made.