Every probe coin's image loads through our server. Each load is counted with request headers only — it works even in mobile apps, where scripts never run.
HTML coins ping home every time their animation plays. Two hooks, one instrument.
Aggregates only. No accounts, no raw IPs, no cross-coin tracking. We count views, not people.
gmgn.ai, basedbot.app and debot.ai hit a probe within minutes of launch — no marketing, no announcement. First time we've known exactly where new coins get discovered.
evidence: strongOne coin, 13 countries, ~20 minutes, zero marketing. The chain tape shows a handful of swaps; telemetry shows 100+ views behind them — the attention iceberg.
evidence: strongA thumbnail fetch landed 2 seconds before a bot's first buy — the cleanest view→buy signal we've captured so far.
evidence: moderateOne address buys ~60 seconds after every large single-tx dump — reproduced twice, loses money both times. It is exit liquidity, not smart money.
evidence: strong30+ wallets sharing one brain, 0.00001 ETH standard entries, bought only 5 of 22 launches. A bot buy means the coin passed their filter — never a take-profit trigger.
evidence: strong$1,491 of "volume" on a $2.95 market cap — the number multiplies tokens by marginal price, not money flow. Measured fees (buys ~0.70%, sells ~0.05%) tell the truth.
evidence: strongThe same watcher sites index Clanker coins as Zora coins — they watch Base itself for new tokens, venue-agnostic. Pump.fun blocks telemetry entirely.
evidence: strongViews and uniques in minutes 0–5 after deploy vs organic buys in the same window — no stimulus. Tweak one attribute per probe batch, watch conversion change. This is the number that decides what ships.