What our AI found

Live now: the AI read on every call

Every call the bot posts now comes with an AI read: the facts our models weigh most, read live from our engine at the moment of the call, and a simple tag that counts the red flags. No percentage. Our study found that no AI percentage is accurate enough to show yet, so we show the facts it leans on instead and keep score in public.

Example of the block in a bot call, not a real one. The tag reads High rug risk with 2 red flags or more, 1 red flag with one, and No red flags with none. No red flags is not a safety check: 3 in 4 of our past calls lost half their value within an hour.

THE RED FLAGS IT LOOKS FOR

• Dev sold. The dev already sold 0.5 SOL or more of this coin. 93% of those clones lost half their value within an hour
• Crowded launch. 6 or more wallets bought in the creation slot. 92% lost half within an hour after migration
• Serial dev. The dev's 5th launch or later. 3% made a clean x2, vs 11% for a dev's 1st launch
• Dev with a past rug. One of the dev's earlier coins rugged. 76% lost half within an hour after migration, vs 62%
• Back below migration. Market cap under 0.7x its market cap at migration. 8% made a clean x2, vs 20% when holding 1.5x or more
• Fading. Market cap at 0.7x or less of 15 minutes before. 5% made a clean x2, vs 18% when up 1.5x or more

"Clean x2" = the coin doubled from that moment before it lost half. On our past calls that happened about 14 times in 100. The good signs it also shows (a dev's 1st launch, holding 1.5x its migration market cap, up 1.5x in 15 minutes) never cancel a red flag. The AI read never decides a call: the bot's safety checks do.

Winners and rugs look almost the same

Speed, a burst of new wallets, a recycled X link, bundles, KOL wallets: these are the marks of an organized pump. An organized pump often migrates first, so it "wins" the war. Then it empties within the hour.

• Clones where 40% or more of the wallets were brand new won their war 73% to 74% of the time. But 86% to 95% of those same clones lost half their value within an hour.
• After migration, 64% of the clones still in the race lost at least half their value within an hour.
• On our past bot calls, 74% lost at least half their value within an hour of the call, and 21% hit x2 within 6 hours.
• Winning the war is not the same as going up. 61% of war winners lost more than half their value within an hour.

The one thing that really separates a winner from the other clones is migrating first and leading the market cap by a wide margin. What separates a rug from the rest is mostly speed and age: too fast and too young.

The signs, side by side

Measured on clones that were really in the race (half the curve filled, or migrated), not the hundreds of dead clones at the floor. "Rugged" here means it lost at least half its value within an hour, or our engine flagged it rugged or dumped. A sign is a tendency, never a rule.

Speed, fresh wallets and recycled links show up on both sides: an organized pump can win its war and still lose half its value within the hour. That is why no single sign is enough.

Why picking early is so hard

• When the first clone of a war reaches half its bonding curve, 2 wars out of 3 will end with no winner at all.
• In 64% of the wars that do end with a winner, that winner is not even born yet at that moment. It shows up about 8 minutes later (median).
• After the first migration it gets easier: when the winner is already among the clones alive, it is almost always the migrated clone with the biggest market cap. Our AI did not beat that simple rule, so we did not ship it as a pick.

Live now: the anti-rug filter

Most single signs above also catch real winners, so on their own they cost more than they save. We tested every one of them, and every combination of two or three, on our real bot calls. One pair stood out.

The bot skips a migrated clone when its launch was bought by 6 or more wallets at creation AND its dev has already sold (0.5 SOL or more).

• Backtest on 121 real bot calls: the filter would have skipped 12. All 12 were rugs. None of the calls that hit x2 or more was skipped, and no war winner either.
• 8 of those 12 would not have been caught by any of our other safety checks.
• On the study clones after migration, it skipped 48 of 921, and 98% of those lost half their value within an hour. It held day by day, Nov 6 to Nov 9.
• It only applies after migration. When we do not know the launch wallets or the dev, the filter does not skip anything.

Signs we tested and did not use, because each one would have skipped big real winners: a dev with a past rug, a dev who sold (alone), half the curve in under 3 minutes, a recycled X link, a burst of new buyers, a dev purchase of 2 SOL or more, KOL wallets inside.

Coming next: a public AI score, once it earns it

We will not put an AI percentage next to a coin until it holds up in public. Today it does not, so it stays private:

• Our first score, built on our early calls, did no better than a coin flip on the newer data (test score 0.44 to 0.50, where 0.5 is a coin flip). It is still recorded on every call, never shown.
• A new "who wins the war" model did no better than the simple market cap rule, so it was not shipped.
• A rug risk score did well in the study (0.78) but not on real calls (0.60 or lower). It is frozen on every call and logged so we can measure it live. It is never shown and never decides a call.

The next step is a better question: will this call hit x2 before it loses half? Our best model so far sorts a little (0.72 on days it never saw, 0.60 on real calls), but its top picks were too optimistic, so it shows no %. From Nov 9, 2026, a private "clean x2" score is frozen on every call next to the rug score, so its record starts now. We check it around Nov 20 and again in early December. A % only goes public if it holds up on at least 14 days it never saw and 300 calls, with each band within 5 points of what really happened. Until then you get the facts, not a number.