Who called it, before it pumped.

Step 1

Resolve the token, anchor the move

We resolve the ticker or contract through Birdeye / GeckoTerminal, then run a move-detection pass over the OHLCV history to find the first liquid breakout. That breakout's timestamp is the move-start anchor — every "pre-move" claim is measured against it. No claim sneaks in because it sounds early; it has to be earlier than this point.

Step 2

5-axis Twitter search

One query is never enough — callers reference the same coin five different ways. We run them all:

• Contract address — the only unambiguous identifier.
• Project handle — mentions of the official @account, excluding self-promo from that account.
• Cashtag + chain — $TICKER solana / $TICKER base etc.
• Name + context — token name plus chain + memecoin / gem qualifiers.
• Address prefix — first 8 chars of the contract for partial-paste callers.

For collision-prone tickers — $AURA, $ON, $BANK, and others that can name unrelated assets — the cashtag alone is never sufficient. A candidate must also carry target-specific evidence such as the project name, contract, official handle, or chain context resolved for the selected token. This deliberately trades recall for protection against silent wrong-asset reports.

We also ingest Farcaster casts via four parallel axes (contract, project handle, ticker + chain, name + context) through Neynar. Casts go through the same disambiguation gate and the same Tier A/B/C scoring as tweets, and a caller posting on both networks is deduped by handle so they surface once with the union of axes that found them.

Step 3

Semantic review, then Tier A / B scoring

Matching a keyword only creates a candidate. Before public promotion, the evidence is reviewed for the resolved asset and for an actual investment thesis rather than a news headline, service alert, giveaway, or unrelated company. Wrong-asset, rejected, service-account, and unjudged rows stay in the audit trail but receive no public caller credit.

Identity-confirmed candidates are also scored on account age, follower-to-following ratio, thesis specificity, one-token concentration, and feed noise. Only semantically verified pre-move Tier A or B candidates enter rankings, caller totals, profiles, activity, and reputation.

• Tier A — established account, healthy ratio, specific thesis, and low noise.
• Tier B — credible evidence with some weaker account-quality signals.
• Tier C — retained for internal/audit context only; never promoted as a verified caller.

Step 4

Gain since the call, from the candle

We fetch the 1-minute OHLCV close at the exact minute of each caller's earliest tweet from Birdeye (Solana) or GeckoTerminal (EVM chains). That's the price we measure their gain against. If a caller also typed a price into the tweet, we use it as a sanity check — disagreement greater than 5x flips a flag and the candle wins. No candle data, no claimed gain — we never invent the number.

Step 5

Cross-lookup reputation

A caller's reputation carries across reports. Only identity-confirmed, semantically verified Tier A or B pre-move calls are recorded. Rejected or pending evidence can never increase a profile's prior-hit count.

Step 6

Onchain skin-in-the-game

A caller who actually bought their own thesis is real signal; one who only tweeted is talk. Where wallet identity is publicly verifiable (ENS, Farcaster), we resolve the X handle to its onchain wallet and check the early-buyer set for that token. If the caller bought within ±48h of the tweet, the full report carries a "Bought their own thesis" badge with the tx hash for verification. We do not scrape X bio links — too noisy, too easy to spoof.