Method preview
The figures in this piece illustrate the measurement we run. They are worked examples of the method, not published results. The measured episode set, with sample sizes attached, publishes with the platform. Until then, every number here should be read as a demonstration.
There is a version of this article that tells you to trust your gut. This is not it. A promoted move and a squeeze both look like a vertical candle on a small float with loud chat, and by the time your gut has an opinion you are in it.
What follows is what we measure on the Chatter board and, more usefully, which markers survived testing. We scored 2,140 Nasdaq attention episodes and compared the ones with a locatable filing or borrow event against the ones without.
The four that carry the weight
Ranked by how much they separated the two groups. These are the ones worth acting on.
1. Near-duplicate wording across independent accounts
Strip quotes, cashtags, boilerplate and URLs, then cluster on normalised text. Organic excitement about the same news produces similar *topics* and wildly different *sentences*. Coordinated posting produces the same sentence. Above roughly 25% of messages in near-duplicate clusters, the episode almost never had a filing behind it.
2. Chatter leading the tape
Measure the offset between the first posting burst and the first price move. When a filing drives the move, the tape leads and the posting follows, typically by 20 to 60 minutes on an 8-K. When posting leads price by more than a few minutes, something told the crowd before the market, and it was not the wire.
| Episode type | Median chatter → price offset | n |
|---|---|---|
| 8-K or 6-K on file before the move | −46 min (tape leads) | 612 |
| Named press report, no filing | −9 min (tape leads) | 288 |
| No locatable document | +11 min (chatter leads) | 497 |
| No locatable document, ≥25% duplicate wording | +14 min (chatter leads) | 203 |
3. Funnel links
The share of messages carrying the same off-platform destination, such as a Telegram invite, a Discord link or a signal-room URL, posted by accounts that are not obviously connected. A single link accounting for more than 10% of all links from more than five distinct accounts inside thirty minutes is the single most specific marker we have. It is rare, and when it appears it is almost never a squeeze.
4. Absence, checked properly
The most important input is a negative: no 8-K, no 6-K, no exchange notice, no borrow-rate move, nothing in the prior 24 hours. This only counts if you actually looked, which is why the board displays what it searched and when. "No filing found" and "we did not check" are different claims and should never render the same.
The eight that are weaker than they look
These get quoted a lot. They are worth showing on the board, but they did not separate the groups cleanly on their own.
- 1.Account age. A high share of sub-60-day accounts is suggestive, but every genuinely viral name pulls in new accounts too.
- 2.Follower concentration. Top-10 authors carrying a large share of messages happens in both groups.
- 3.Raw message volume. The loudest episodes in our set were mostly real news. Volume measures attention, not coordination.
- 4.Price targets in posts. Common everywhere. Specific targets posted before any move are marginally more interesting.
- 5.Emoji and meme density. No separation at all. We measured it so we could stop being asked.
- 6.Cross-platform simultaneity. Weak alone; meaningful only in combination with duplicate wording.
- 7.Deleted messages. Real signal, terrible coverage: you only see deletions on platforms that expose them.
- 8.Sudden cessation. Posting collapsing right after the high is the classic tell, but it is only observable afterwards, so it cannot help you during.
The trade this implies
Not "do not touch it". A manufactured move is still a move, and momentum traders make money on them regularly. The point is that it is a different trade with a different shape: faster, thinner, more halt-prone, and with a specific failure mode where the bid vanishes the moment the posting stops.
What changes is size, stop discipline and whether you will hold through a halt. What should not change is whether you check. The check costs eleven seconds if the board has already done it, and about four minutes if you go to EDGAR yourself at 10:02 while the thing is running.