Updated 1 October 2026.
Block0 reads Robinhood Chain launches from their own transfer history on the public node and reports what happened: who bought first, who bought together, who holds the supply, who is selling, and what the deployer launched before. Every number is reproducible from public data. It does not predict price — and below is the record of the tests that tried to, and failed.
Where it reads from: the free public Robinhood Chain node, the Pons launchpad's own catalogue, the Orbio protocol API, and DexScreener / GeckoTerminal for price context. The Block0 agent re-reads every Pons and Orbio launch under 72 hours old (≥$5k, ≥5 holders) about every 15 minutes.
Every claim that Block0's reads could predict something was tested against history with the rule written before the result. None passed.
| Question | Result |
|---|---|
| Does the "most promising" ranker pick better launches? Daily pick, 53 days, judged on return from entry. | No. It beat a random pick on how often a token later reached $1M, but lost to random on return (median 1.58× vs 2.96×): it picked tokens already part-way up. |
| Do the forensic flags warn of a wipeout? Bundles, top-10 ≥60%, snipers ≥10%, risk ≥50, serial deployer — 627 graduated launches. | No flag passed. Apparent effects were explained by entry size. Only graduated launches were studied, so this does not test rugs before graduation. |
| Do launches in a "markup" state keep running? | The effect reversed out of sample (5 v 2 in discovery, 3 v 6 in validation). Stopped, as pre-committed. |
| Does smart-money convergence lead to better returns? History. | Unanswerable from history: only 13 measurable events, and early prices disagree between sources. Now tested forward — below. |
So the reads on this site are descriptive: they tell you what a launch's holders did. They are not shown to tell you what its price will do.
Whether proven wallets converging on a launch means anything is being measured forward, in public, with the rules fixed before the first event (tools/radar-protocol.mjs). Every convergence is logged when it is seen, priced at that moment, next to up to four launches of similar size and age that no tracked wallet touched. Seven days later each is scored from market candles that start after detection, against its own controls only. No verdict before 30 scored events; the running report is public.
Protocol v3 (2026-10-01 follow-up) excludes any buyer with an observed sale, transfer out or deposit from convergence, including partial outflows. A verdict also needs at least 30 paired seven-day closing outcomes and 80% closing coverage in each chronological half (the early 70% / late 30% split); each paired event needs 80% of its own logged controls covered. Insufficient coverage means no verdict, not a pass or fail. These are completeness safeguards, not proof of statistical power. Earlier v1/v2 events stay in the log, counted and not scored under v3. Until the report says PASS, the radar is unvalidated — and the agent does not post convergence at all.
Cards and dossiers carry a 0–100 risk score: a summary of the facts above, not a forecast — the tests above found it does not predict outcomes. It is a transparent weighted composite of five sub-scores, each computed from the token's own transfer history over its real distributed float (the bonding curve and known venues are excluded so their reserves don't read as a giant holder).
| Sub-score | What it measures | Weight* |
|---|---|---|
| Concentration | Share of supply held by the top 10 wallets, above a healthy baseline | 26–38% |
| Selling now | Early wallets (snipers + bundles) net-selling into a venue in the 30 minutes before the read. Transfers to other wallets don't count. | 20–26% |
| Bundles | Coordinated same-block buying at launch (one actor, many wallets) | 8–20% |
| Coordination | A hidden cluster — wallets linked by hand-to-hand token moves (or a shared funder), holding a real slice. Catches the operator who splits into wallets that avoid the same-block bundle, then shuffles the token between them. Built from the token's own transfers, free. | 12–14% |
| Snipers | Wallets that bought within 3 blocks of the first pool trade, and their held share | 8–12% |
| Deployer | Share of supply still sitting in the deployer's wallet | 8% |
*Weights shift once a token graduates: concentration and live dumping matter more for an established coin, launch-mechanics (bundles) matter less. On top of the weighted mean sit three disqualifying floors — near-total concentration (top-10 ≥ 82–92%), two-plus insiders selling right now, or a coordinated cluster that owns a real slice AND is distributing this minute. Any one forces a high score regardless of everything else, because it single-handedly defines the risk and must not be offset by a clean sub-score elsewhere.
Labels: Looks cleaner <25Mixed 25–44Caution 45–65High risk 66+
Everything in this section describes research that is parked. It is kept because it is the basis for the failed tests above, not because it is in use: nothing on the board or in the agent's posts uses the resemblance score, the corridor or the ladder.
These are two different questions, and we keep them apart. The risk score above is a fact we can check now — how manipulated the launch looks. This second read is a historical resemblance score (0–100): how closely a fresh launch matches the shape the biggest past winners had at the same age. It guides attention — it is not a probability of success. A launch can look clean and still fail; a messy one can still run. Validated discriminators, in order of weight:
It deliberately does not reward low sniper activity — the winner cohort showed heavy early snipers were survivable, while bundles were what killed. A strong resemblance is worth a closer look; it is a pattern, not a promise. The comparison set is every launch that closed above $1M for a day — the full class, including the ones that later faded, not a handful of survivors — but this read is still guidance to focus your attention, never odds; it strengthens as the cohort grows (see the measured limits below).
These come from a study of a selected catalog cohort, labelled by observed outcomes — the pipeline supports Pons and direct DEX listings, but actual venue coverage depends on the committed cohort. Graduation is not a criterion. The study asks two separate questions, because they have very different amounts of evidence behind them:
Each reached token's whole history is replayed and reduced to a trajectory (distribution health blended with real wallet inflow) and a wallets→market-cap curve.
loading cohort definitions…
"Held-peak" is the highest market cap the token stayed at for a full day (reconstructed caps wick on single swaps). Pending and mid tokens are excluded from both sides — a three-day-old token above $1M can't have held it a week yet, so it is never counted as a loser (right-censoring). The launchpad's own token and quote assets are excluded.
The model is a committed, reproducible pipeline — every backtest is cached and re-labelled with live market caps, so the cohort compounds run after run and the model refreshes as pending tokens settle into winners or fades.
The honest question — for Q1, "will it reach $1M?". Below is a separation test: at each age, how well the trajectory tells the tokens that went on to close above $1M apart from the ones that never did. Two controls are graded: all (stalled + dead + faded below the bar) and faded only — tokens that got to $300k and then collapsed without ever holding $1M, the ones that look most like runners early. AUC is the probability a random reacher scores above a random control: 0.50 is a coin flip, 1.00 is perfect. Q2 (will it hold) is measured separately further down.
loading measurement…