Back

FOMO Radar

Which tokens are credible wallets buying right now. Robinhood Chain signals are ranked by trader-credibility conviction — every buyer is weighted by their own realized on-chain track record, never by dollar size. Solana signals run through the same credibility machinery, from per-wallet swap history parsed off a public Solana RPC — and any pool we have not indexed enough swaps for keeps its pool-level buy-pressure score and says so. Every score carries a measured probability of a 3× inside 72 hours — or says plainly that there is not enough labelled history yet.

← Requested on /duplicate — built free in 24 h

Buy signals

Loading live signals…

How the probability is measured

Every signal this radar emits is written down before its outcome is known, then labelled 72 hours later from the prices that actually printed. The percentage next to a score is the share of past signals at that score that went 3× within 72 hours — not a model's opinion of the score.

Current fit

Predicted vs observed

Rolling precision by signal type

Persist. Each emitted signal is stored with its chain, token, timestamp, conviction, signal type and feature block — one row per token per 6-hour window, written before anyone knows what happens next.
Observe. Robinhood Chain is indexed by no public price feed, so its price history is recorded from the chain's own trade prints (ETH per token on every curve trade), every 60 seconds for the first hour and every 15 minutes after that. Solana falls back to public pool candles.
Label. 72 hours later each signal gets: max multiple, whether it hit 3×, whether it rugged inside the first 5 minutes (price or liquidity down 90%), whether it rugged within 24 hours, and — on bonding curves — whether it survived graduation.
Fit. An isotonic regression (monotone, so a higher score can never map to a lower probability) is refitted every 15 minutes over the rolling window of labelled signals. The Brier score shown is out of sample: fitted on the oldest 70% of the window, scored on the newest 30% it never saw, next to what the flat base rate scored on the same tail.
Gate. A market-wide regime measure (median 24-hour return and the share of tracked tokens below their 24-hour open) scales the exposure recommendation. When the whole meme market is bleeding it goes to zero and the signals are suppressed — the answer to a bad tape is no exposure, not a better model.

A signal whose 72-hour window we could not price is marked unmeasurable and excluded from the fit — never counted as a miss. Below the minimum labelled sample the page shows insufficient history (n/N) and no percentage at all. Machine-readable: GET /api/v1/clones/fomo-radar/calibration?chain=robinhood|solana.

Trader leaderboard

Wallets ranked by their realized track record on Robinhood Chain bonding curves — closed round trips only, shrunk toward 50 by sample size. A wallet we have never seen close a trade scores exactly 50 and does not count as qualified.

Entity resolution — sibling wallets count once

A wallet costs nothing to create, so counting addresses counts what a farm can mint. Every qualified buyer on Robinhood Chain is resolved into an entity by union-find over four kinds of evidence, and conviction is summed over entities: wallets that always trade together cast one vote between them.

Same atomic transaction

Two wallets bought the same curve inside one transaction. One transaction has one sender paying for it — independent strangers cannot share one. The same block is a weaker version of the same link and is drawn separately.

Common non-exchange funder

The address that submitted and paid for a wallet's first observed buy. A wallet that submits its own transaction draws no link, exchange and infrastructure addresses are excluded by a maintained list, and a funder with dozens of children is treated as infrastructure rather than one operator.

Shared bundle id

Where the chain publishes a bundle id, wallets sharing one are the same submission. Robinhood Chain does not publish one today, so this rule reports itself unavailable instead of guessing a link.

Portfolio cosine similarity

Token portfolios are turned into vectors weighted by how rare each token is, and two wallets pointing the same way — sharing at least two tokens — are one entity. Sharing the day's hot coin is worth nothing here; sharing two coins nobody else touched is worth everything.

How an entity votes

Once. Its score is its best member's track record, so a farm can neither dilute nor amplify it. Its ETH is the sum of its members' buys on the curve. Its selectivity is computed on the entity's whole portfolio: forty tokens spread over eight wallets is a forty-token buyer, not eight specialists.

When a source is down

That rule draws no link at all and those wallets are counted as separate addresses — the honest, conservative answer. The live state of each rule is listed below and never hidden.

Resolving entities from the live window…

JSON for agents and bots — no key required: GET /api/v1/clones/fomo-radar/signals?chain=robinhood|solana&filter=all|hot|fresh|exits, GET /api/v1/clones/fomo-radar/traders?chain=robinhood, GET /api/v1/clones/fomo-radar/entities (the resolved entities and the evidence that linked them), GET /api/v1/clones/fomo-radar/calibration?chain=robinhood|solana (the measured score→outcome link), GET /api/v1/clones/fomo-radar/docs (the exact scoring formulas). Robinhood Chain data is read straight from the chain-4663 bonding-curve trade logs; Solana pools come from public DEX aggregates and Solana per-wallet track records from swap transactions parsed off a public Solana RPC. Where we have not indexed enough swaps for a pool, that row keeps its pool-level buy-pressure score and says so — a trader score is never invented. Nothing here is financial advice — an on-chain track record is a description of the past, not a prediction.