How it works

Price tells you what happened. Attention tells you what is about to.

Loud watches people talking about tokens on X, scores how much of that talk is real, and ranks a live board on it. Then it lets you trade straight from that board. This page explains exactly how each number is produced — including the ones we refuse to show.

1. Where the data comes from

A browser extension runs in a real, logged-in Chrome profile and reads x.com/search?q=$CASHTAG&f=live the way a person does — scroll, read, move on. It is not an API scraper and it is not a bot farm; it is a browsing session that happens to be patient.

The server decides everything: which token is due, how far back to read, and when to stop. Tokens are tiered by how fast they are actually being talked about, so a cashtag doing forty posts an hour gets looked at constantly and one doing four a day gets looked at rarely. Nothing is on a fixed schedule, because a fixed schedule either wastes reads on quiet tokens or misses fast ones.

2. How the attention score is built

The score is not a mention count. A hundred posts from three accounts is not a hundred people caring, so the input that matters most is unique posters, bucketed by hour.

Unique posters
Distinct accounts in the scoring window. The primary signal — one person posting twenty times counts once.
Velocity
Posters right now against the previous six-hour mean. Catches a token waking up, which is the entire point.
New posters
Accounts talking about it for the first time. Distinguishes a spreading story from the same circle repeating itself.
Reach
Views and engagement on the posts, so one large account is not treated the same as one anonymous reply.

Everything is timestamped in UTC and compared in UTC. That sounds like a footnote and is not: an hour of drift puts every recent post in the wrong bucket and makes the "last hour" column read zero forever.

3. Why the window is six hours, not one

Posts carry their original timestamp but only reach us when a scrape runs. A one-hour window on a token scraped every eight hours would read empty for hours, then spike the moment a scrape landed — the chart would be measuring our schedule, not the conversation. A six-hour window is wide enough to survive the gap between reads and short enough to still be news.

4. Market data

Prices, liquidity, volume and market cap come from DexScreener, filtered to Robinhood Chain (chain 4663). The board refreshes prices every 15 seconds and the slower fields every 60. Refreshes are shared: whichever request first notices data is stale updates it for everybody, so a hundred people watching the board costs the same upstream as one.

5. What we deliberately do not claim

DexScreener exposes no holder count and no holder distribution for this chain, and ERC-20 has no mint or freeze authority to revoke. So the token page shows only checks we actually ran — liquidity, our own organic score, age, and whether another tracked token shares the ticker — and renders an unknown holder count as an em dash.

Rendering a safety check we never ran is worse than running none. A green tick you cannot verify is more dangerous than a blank, because it buys confidence nothing earned.

The organic score is ours, not a third party's. It is computed from 24-hour turnover against pooled liquidity and the balance of buys to sells — a pool doing fifty times its own depth in a day is being cycled, not traded. It is labelled as our number everywhere it appears, because it is an estimate and not a fact.

6. Trading

Swaps route through LI.FI, which fills on Robinhood Chain through the chain's own venues. Loud is non-custodial: the app builds a transaction, your wallet signs and submits it, and we only ever see the resulting hash. We never hold your funds and cannot move them.

See Fees for what a trade costs and Risk for what can go wrong.