Changelog

Where hoodini is, and what just shipped. 0.x means the copy book is a paper model; 1.0 ships when the live book trades on mainnet.

1. 0.72.2Latitude exits, set by hand2026-09-26
   • Latitude positions now sell half at 2× (was 3×), and sell the rest if they fall 25% under their peak once they have been past 1.5×.
   • The Latitude time stop no longer sells a position that is under its entry — it is left to the stop, the take-profit and the trail.
   • What is left after the first take-profit has its own trail setting, so the remainder can be given more or less room than the position had before.
2. 0.72.1Charts you can scroll back through2026-09-26
   • Scroll a token's chart to the left and older history loads in, for candles and lines alike, until the token's first trade.
   • Every candle interval works in every window, with 4h and 1d added, and each loads a full page of bars rather than the few that fit the window.
   • Candles are drawn the way trading charts draw them: each opens where the last closed, a quiet interval is a flat bar instead of a gap, and a bad price reading no longer stretches the scale.
3. 0.72.0Latitude: later stop, and not after our traders2026-09-26
   • The Latitude book no longer buys a pick three or more of our watched traders already hold — replayed over every paid pick, those lost money on both halves of the record while the rest made it.
   • Its stop now sits at −40% at the tightest, instead of −30%: a stop that close sold many picks at their low just before they came back.
   • Every open Latitude position now records how many watched traders still hold its token on chain, so whether their holding should steer the exit can be measured rather than guessed.
4. 0.71.11Trader records measure the whole trade2026-09-26
   • The book every trader is measured on no longer has 20 slots. Most of its positions used to be pushed out after about an hour to make room for the next buy, so a record said how a trader's buys did in their first hour; now each one runs until a take-profit, trail, stop or the trader's own sell closes it.
5. 0.71.10A quieter channel and a quieter log2026-09-26
   • The Telegram channel posts only FOMO traders' buys and sells while we copy FOMO traders only.
   • Skipped signals no longer fill the activity log.
6. 0.71.9FOMO accounts filter2026-09-26
   • The Traders page has a new filter, FOMO accounts: only the traders with a FOMO account, without the wallets we found and scored ourselves.
   • The Signals and Traders pages now say what is true: we watch and score every wallet on the list, and copy some of them.
7. 0.71.8Market-cap bands can size down too2026-09-26
   • Each market-cap band's share of the ticket can now be set below 100% (down to 10%), not only above it — so a band that loses can be traded smaller rather than at full size.
8. 0.71.7Conviction follows a crowd, and says so2026-09-26
   • A conviction position no longer looks like a copy of one trader. It shows the crowd behind it — the tracked traders who bought and still hold the token — and the buy that opened it moves to the tooltip.
   • The dex book is sized like Latitude now: the hour of the day and the market-cap band, nothing else.
9. 0.71.6Latitude tickets, sized by time and market cap only2026-09-26
   • A Latitude position is now sized only by the hour of the trading day and the token's market-cap band. A rule built for copying traders — halving the ticket on any pair more than two hours old — had been cutting 31 of the last 50 Latitude tickets in half.
10. 0.71.5Cabals, read from how traders actually trade2026-09-26
    • The cabal-linked filter now finds wallets that buy the same tokens in the same minute, again and again, far more often than chance — a crew behind several wallets. Until now it only knew about wallets sniping the same opening blocks, and almost no trader we follow does that.
    • Traders that nearly always buy a minute after another trader now say who they copy. Following a copier puts us third in line.
11. 0.71.4Copying only the proven traders2026-09-26
    • The copy books now follow only the traders proven on the FOMO leaderboards. The wallets we found ourselves are still watched — their buys logged, counted in every crowd and conviction reading, their records kept — but they no longer open positions.
12. 0.71.3A clearer token page2026-09-25
    • The shortlist panel is now two: the watched traders' own buys and sells, and — separately — what our copy book did about them, entry and exit.
    • Buys are green, sells red, and every on-chain trade links straight to its transaction.
    • An info tip next to the momentum score explains what it measures.
13. 0.71.2Fewer paid reads, and who has gone quiet2026-09-25
    • Spam airdrops no longer cost us: a token dropped into fifty or more of the wallets we watch, and never once bought by anyone, is recognised as spam and skipped — about three in five of the tokens those wallets are sent each week. One of its transfers an hour is still checked in full, so a token that turns into a real market is caught within the hour.
    • The conviction check reads every trader's balance in one request instead of one each.
    • Trader pages now say when a wallet was last active and whether it has gone quiet, with its past quiet spells. We keep watching every one of them — a trader who comes back is seen the moment they trade.
14. 0.71.1Runners, found early — not bought blind2026-09-25
    • A Latitude pick that pays for a DexScreener profile after we spotted it now wears a "paid after spotted" badge. Of the week we studied, 37% and 19% of these went on to 5× — against about 8% of everything else — which makes it the clearest early sign of a runner we have measured.
    • The paper book that bought every such pick and held it for a week is switched off: it tied up positions and could crowd out the traders we copy, and its rules can be measured from the price history we now keep without holding anything.
15. 0.71.0The FOMO leaderboard, live again2026-09-25
    • The FOMO leaderboard we found traders on had quietly stopped updating in August. We now read it live — every window FOMO publishes, a few times a day — and a second Robinhood-only board alongside it. New traders who are in profit and trade on Robinhood Chain join the list; nobody is ever removed from it.
    • Trader pages show FOMO's own numbers for that trader across the day, the week, the month and all time, and a wallet we only knew as anonymous now shows the FOMO name behind it when the boards tell us.
    • Conviction positions — the ones several of our traders hold at once — now take a win: all of it at 3×, or once it has been past 2.5×, when it falls a quarter off its best. Below that nothing changes: no stop, no clock.
16. 0.70.3The opening bundle counts buyers again2026-09-25
    • Most tokens were showing an opening bundle of 100%. It was counting the pool's own liquidity, and the launch contracts that hold a token's supply, as if they were buyers. They are not, and they are left out now — so the figure is what wallets actually took in the first half-minute, which on most tokens is a fraction of that.
    • Recent tokens are being recomputed in the background, a few at a time.
17. 0.70.2Every price is the chain's2026-09-25
    • Prices and market caps across the site — token pages, the paid board, Latitude, the market sweep — now come from each token's own pool on the chain, and the market cap is ours: that price times the token's supply. DexScreener is still how we find where a token trades and how deep the pool is; it no longer tells us what anything is worth.
    • The ETH rate every dollar figure goes through is now read from the chain's deepest ETH market as well.
18. 0.70.1Latitude's prices come from the chain2026-09-25
    • Latitude's prices and market caps are now read from each token's own pool on the chain, not from DexScreener. The same goes for the paper books' marks. When a pool cannot be read, the price says unknown rather than borrowing another source's number.
    • It is cheap because it is batched: every pool Latitude watches is read in a handful of calls a minute, not one call per token.
19. 0.70.0Holding Latitude's runners2026-09-25
    • We studied a week of Latitude: 853 picks, 108 of which went on to 5× or more. The runners were not the fast ones. Half peaked more than eight hours after Latitude spotted them, a third more than a day later — and a third first fell below half their price on the way. A book that takes profit early and stops out at a loss sells exactly those.
    • So there is a second, paper-only book on Latitude's picks, built to hold: it takes half off at 2× and lets the rest ride for up to a week with no stop, trailing only once it has run past 5×. It buys two kinds of pick — Pons v2 launches a few hours after they are spotted, and tokens that pay for a DexScreener profile after Latitude found them, at the moment they pay. Every price it uses comes from the chain.
    • Latitude now follows every pick for two weeks instead of one, and scans each one in full as it is spotted.
20. 0.69.0What a token's buyers actually did2026-09-24
    • Every token page now reads the token's whole trade history and says what its buyers did with it: how much the first 10 and first 50 buyers have sold, how many flipped within half an hour, how many sold and came back, how many never sold, and the last 15 minutes of buys and sells.
    • A new Supply & launch box on the left: where the dev's tokens went — locked, held or sold, with the unlock date when it is locked — whether the wallets from the opening bundle are still in, how many of the traders we watch bought and still hold, and how many pools a token has, including the high-fee decoys.
    • A long history is read a few pages at a time, newest first. Until it reaches the launch, the page says so rather than showing a partial number as the whole.
21. 0.68.3Outages end on their own2026-09-23
    • When a service we cannot run without goes down, everything pauses — that has not changed. What has: the pause used to last until someone noticed the service was back and pressed a button. Now it lifts itself once the service has answered three checks in a row, a minute apart.
    • The live trading bot is the exception. It comes back able to sell but not to open anything new, until a person has looked.
22. 0.68.2A holder-data outage no longer stops the lists2026-09-23
    • On 23 September our holder-data provider was down for three hours, and we stopped everything with it — Altitude and Latitude went dark and came back empty. Holder counts are the only thing it feeds, and every page already says "unknown" without them, so an outage there now raises an alert and nothing else.
    • The lists are built from every Fomo buy on the chain. When we fall behind on those, for any reason, the stretch we missed is now read back in the background rather than skipped, so a pause costs a delay instead of the data.
23. 0.68.1The token page gets its footer2026-09-22
    • Every page on the site ends with a footer except the one people land on most. The token page never had one — the only footer in that file belonged to the view you see when a token is not indexed yet, so the fallback had it and the real page did not.
    • It has one now, on all three of its states, and a test walks every route so the next page to go without one says why.
24. 0.68.0$HOODINI in the footer2026-09-22
    • The site's own contract address now sits in the footer, on every page, with one click to copy the whole thing. It links to our page for it rather than anywhere else — which also means it works while the token is still on its curve, and keeps working after it graduates.
25. 0.67.1Shadow mode stops pretending about real positions2026-09-21
    • Putting the bot in shadow mode is meant to stop it sending anything while the paper books keep running. But a position opened with real money is not a paper position, and the bot was treating it like one: told to sell, it would quietly pretend to, leave the position marked as selling, and leave the tokens sitting in the wallet. The book thought it was getting out; nothing was happening.
    • It now refuses outright and says so on the position, and tries again the moment the bot is live. A real position is never pretended at.
26. 0.67.0Conviction has no floor either2026-09-21
    • A conviction position kept one rule that could close it: a stop at half its entry. That is gone. Nothing the book decides ends a conviction position now — not a stop, not a trail, not a clock, and not the trader who bought it walking away. It ends when an operator closes it, or when the token is worth nothing.
    • That is what conviction was always supposed to mean, and it is the whole of what it means now.
27. 0.66.0Your lists, with or without a buy button on every row2026-09-21
    • The feed, Altitude, Latitude, Paid and Signals each put a one-click buy on every row. It spends your buy size, and it sits exactly where a thumb scrolls — which some people want and some people really do not. There is now a switch for it on your account page. It is on by default, so nothing changes unless you change it, and the buy button on a token's own page is never affected: that is a page you opened on purpose.
28. 0.65.3The conviction button actually buys now2026-09-21
    • It reached the worker on the first press and fell over writing the position down: the row was being credited to nobody, and that column will not accept nobody. It is credited to the admin instead — a name that appears on no trader's record, which is what crediting nobody was for.
    • It now looks and behaves like the buy button sitting next to it, in the same place on a row, and asks with the same dialog the Signals actions use — which also fixes it being awkward to press on a board, where the whole card is a link.
29. 0.65.2The site now lets go when it is asked to2026-09-21
    • Every deploy ended with the web server being killed rather than stopping. A live feed is a connection that never ends by itself, so the server sat waiting for one to finish, ran out of its five seconds, and was cut off — along with whatever page someone happened to be loading at that moment.
    • It now hangs up its own live feeds first and leaves cleanly. A feed asked for mid-restart is turned away with a note to come back in two seconds rather than handed a connection that is about to be dropped.
30. 0.65.1Eighteen tokens that could never be read2026-09-21
    • Anyone can deploy a token whose name is a byte a database cannot store. One of them had been jamming the metadata reader since 4 September — it failed once a minute, every minute, and took the rest of its batch down with it each time. Eighteen tokens sat behind it with no decimals recorded at all, which is the number every calculation about them depends on.
    • Names are now cleaned of the handful of characters that cannot be written down, and nothing else: emoji, accents and odd spacing are what the contract says its name is, so they stay. A token that still cannot be read is marked as read anyway, with its address in the log, so one bad name can never again cost every token behind it.
31. 0.65.0An admin's close on a conviction position sticks2026-09-21
    • Conviction is the one kind of position the book will hold indefinitely — no take-profit, no clock, only its floor — so closing one by hand is the only way that decision ever gets made. Until now the crowd rule could re-elevate the same token minutes later, which undid the decision rather than acting on it.
    • A token an admin closed out of conviction is now left alone by both automatic routes back in — opening a fresh conviction position, and promoting one already held — for 24 hours by default. Both the switch and the period are settings.
    • It counts across every book, because the judgement is about the token and not about which book happened to hold it. The admin conviction-buy button is deliberately exempt: that is the same person deciding the opposite, and it notes the recent close on the position rather than refusing.
32. 0.64.0Buy it as conviction, from wherever you spotted it2026-09-21
    • An operator can now buy a token as a conviction position straight from its own page or from either board, without waiting for four tracked traders to find it first. It opens at the default ticket on the paper book and, while the bot is live, on the live one.
    • It ignores both ceilings on purpose — the slot cap, which a conviction position never counted against anyway, and the conviction ceiling, which the automatic path still respects. When it goes over that second one it says so, in the log and on the position, rather than going over it quietly.
    • What it does not skip is anything protecting the money: the honeypot check, the sell-tax bar and the route's own refusal to buy what it cannot prove a way out of. Choosing to buy something is not a reason to buy something unsellable.
    • It stays out of the record book, which is the measurement every trader's record is read from — this position has nobody behind it, so it is credited to nobody and cannot move anyone's numbers. After that it is an ordinary conviction position: held on the way up, closed by its floor.
33. 0.63.0Conviction holds2026-09-20
    • A conviction position is one four or more tracked traders are in at once. It is sized up for that reason and it stops following any single trader out — but it had been borrowing the Latitude model's exits to have anything at all that could close it, which meant a token nineteen tracked traders were holding would have been sold at +30%. A multiple fitted against young Radar picks, answering a different question about a different kind of token.
    • It now states its own terms: no take-profit at any multiple, a 50% floor that still closes it if the thing collapses, no trailing stop and no clock. A crowd holding a token is not a reason to sell it.
    • It also reads nothing from the nightly model fit any more, so a retrain can no longer change what a live conviction position does while nobody is looking.
34. 0.62.0A token's timeline stops hiding the trap pools2026-09-20
    • A token page could list half a dozen identical lines reading "Pool created · uniswap-v4" and tell you nothing about any of them. They were never duplicates: anyone can open a Uniswap v4 pool on any token, and on this chain most of them are bait — pools with fees of 80% and 90% that exist to catch anything routing by size instead of by price.
    • Across every pool this site has ever recorded, 104,359 of them charge over 50%. That is a third of all of them.
    • The timeline now prices each pool and says which is which. A run of unusable pools folds into one line with the count and the fee range — four pools at 79-90% is not noise to hide, it is what somebody farming routers looks like. The token's real market keeps its own line, with its fee, and whether a hook takes it.
    • Our router has always stepped around these; the page just never said so.
35. 0.61.0The bot stops paying more for gas than the trade is worth2026-09-20
    • The bot checked it could afford the gas on a trade. It never checked whether the gas was worth paying. On a quiet chain that is the same thing; on a crowded one it is not, and a busy block could have cost more in fees than the position was ever going to make.
    • A buy is now skipped when the round trip's gas — in, out and the approvals between — would cost more than a tenth of the ticket. At the price this chain actually runs at, that is a twelvefold margin: every trade made tonight clears it with room, on any route length.
    • It never applies to a sell. A cap that can block an exit traps the position, and paying over the odds to get out beats not getting out.
36. 0.60.0Latitude gets a gate that works and an exit that can let a winner run2026-09-20
    • Latitude now opens a position only on a token that has paid DexScreener for a profile or boosts. Measured over 654 entries in fourteen days: tokens that had paid reached 2x 42% of the time and 5x 15%; tokens that had not, 14.6% and 6.1%. Nearly three times the strike rate off the same list — and it is work the model's own score was not doing, since among paid tokens the highest-scoring group did slightly worse.
    • Its exits could only ever be 'sell everything at one multiple'. The fitter had no trailing stop in the space it searches, so it could not have chosen one however well it worked, and it settled on +30% — a cap, on a list whose best five entries last week peaked between 35x and 105x. It can now also choose to take half and let the rest run behind a trailing stop. Today's behaviour is still in the grid, so if capping really is best the model keeps it and says so.
    • For the record: the wallet and dex books already do what you would expect — half out at 2x, the rest at 5x, and a trailing stop that arms at 1.5x and closes everything on a 25% fall from the peak. Latitude was the one book running different rules, and nothing said so.
37. 0.59.0The chain prices a position, or nothing does2026-09-20
    • When neither a chain quote nor the pool's own mid could be read, an entry used to be booked at DexScreener's price. Measured on the 1,433 rows last week's repair could reach, that price sits a median 29% BELOW the pool — wrong in the direction that flatters, and worst exactly when the price is moving, which is when the quote fails and the fallback fired. That tier is gone. A position we cannot price from the chain is not opened; the signal is still recorded, with the reason.
    • The same for market cap: a token the chain could not price took DexScreener's figure, built on the same price. It now has no market cap, which the sizing already reads as no opinion and the buy window refuses outright.
    • 918 older positions still carry that price and cannot be repaired — no chain mid was ever read for them. They are flagged rather than rewritten: deflating them by the median correction would be inventing a price for each one, and a median is not these rows' own number. They overstate the two books by about $2,368, roughly a third of each book's reported profit and sixteen points of its win rate.
    • DexScreener keeps the job it is good at — telling us a token paid for a profile or bought boosts, and that a pair exists.
38. 0.58.0One ticket, and a clean slate2026-09-20
    • The paper books sized their positions in dollars and the live bot sized its own in ETH — $10 a paper ticket against 0.003 ETH, about $7.75 — so the simulation was never a simulation of the trade it stood in for. Both now read the same ETH knob, and the paper slot cap reads the live one. A sim that follows different rules than the thing it predicts measures nothing about it.
    • Gone from Admin: the paper ticket, the paper slot cap, and the Latitude and Dex paper tickets. What stays is the switch that turns each book from paper to live, and the switch that turns each book on at all. The live book keeps its own elastic ceiling, because how many positions it can fund is a different question from how many are worth watching.
    • The record book keeps its own fixed ticket on purpose. It copies every signal at one size so the multiples it exists to measure stay comparable — that is what makes it the measurement rather than another opinion.
    • Every open position was closed at its current mark and the books start again under the new sizing, so nothing in them is half-priced under one rule and half under the other.
39. 0.57.1What a stale exit is worth, measured before it is repaired2026-09-20
    • When a tracked trader sells out, the paper and record books close at the previous mark rather than asking the pool. Every other exit — trail, take profit, stop, moonbag, and displacement — prices at a quote zero seconds old. Measured over 14 days, this one path priced at a mark a median of 119 seconds old, and the two minutes a trader spends dumping are not ordinary minutes.
    • That is the same fault as last week's entry repair, on the other side of the trade, and it points the same way: a stale mark in a falling market flatters the book. But that is an inference, so nothing has been repriced. Each of these closes now records what a fresh quote said at the moment of exit, beside the mark it was actually priced at. In a week the two columns answer how big the error is, and any repair is arithmetic instead of a hunch.
    • Fixed alongside: all 562 of these closes claimed DexScreener as their price source whatever had really marked them, because the exit helper defaulted to it and nothing passed a value. It now names what priced the row, and the default that lied is gone.
40. 0.57.0DexScreener stops being able to take the site down2026-09-20
    • DexScreener rate-limited us three times in fifteen hours, and because it counted as a provider we cannot run without, each run suspended every worker and put the site into maintenance — twice for over four hours. It is now an auxiliary provider: when it throttles, the alert still goes out and its readings say unknown, but nothing stops. Marks are quoted from the chain, routes fall back to the on-chain probe, and no reader of ours ever needed it to answer.
    • The reason it throttled at all: about 20,000 calls an hour against a ceiling of roughly 18,000, and 315,000 of them a day were the same handful of tokens asked again every ten seconds. Those answers are now shared for ninety seconds, which cuts the traffic by around 90% and changes no number anywhere — they decide which pool a token trades in, not what it is worth.
    • Every DexScreener call now goes through one paced gate that spaces requests, retries a throttled one twice on its own, and backs off further the more it is refused. A call that succeeds on its second attempt is simply a call that succeeded; the monitor no longer hears about it.
41. 0.56.3The paper book, on the same footing2026-09-19
    • The entry-price repair now covers the paper book too — the one the signals page shows. Its week reads about +$913 where it read +$1,425, and 84 of its wins were losses: the difference was never real, and leaving it would have meant publishing a number we had already fixed everywhere else.
    • Both books are now priced the same way, which is what makes them comparable at all. The gated book still returns about 11% on capital against the ungated record book's 7.5%, so the gates are still earning their place — that conclusion did not depend on the error.
42. 0.56.2The measurement, corrected2026-09-19
    • The record book — the unbiased copy of every signal that trader records and ticket sizes are read from — carried the entry-price fault fixed yesterday on 863 of its closed positions. Those are now priced at the market as it was at the moment, and the book reads +$3,622 where it read +$5,436: a third of it was never real, and 187 of its wins were losses.
    • Positions the books actually traded are untouched. They are a record of what happened, not a measurement, and rewriting them would be rewriting history rather than correcting an instrument.
43. 0.56.1An entry priced at the market, not at a memory of it2026-09-19
    • Fixed: when the chain could not quote an entry, the position was booked at DexScreener's cached price instead — which during a fast move sat a long way under the pool's own price. One position read +11% while the token was down 25%. Measured over a week, entries priced that way sat a median 20% below the market, so a fifth of the gain on those rows was never real.
    • Those entries now take the pool's own price, read at the same instant, plus what a buy actually pays over it. DexScreener is a last resort rather than the fallback, and each row records which of the three it used. Positions already open keep the entry they were opened at.
44. 0.56.0Premium+2026-09-19
    • A second tier. Premium keeps the private Telegram channel and support; Premium+ adds the Signals book and the Traders pages. The membership page compares all three columns side by side.
    • Everyone who was already a Premium member kept Signals and Traders for as long as their membership runs. Nobody lost anything they had paid for.
    • Paying for Premium+ while on Premium upgrades you and keeps the days you have; paying for Premium on a Premium+ membership adds the days and never downgrades it.
    • Vouchers are minted for one tier or the other, and every code already out there is a Premium code. The same choice is on the admin comp.
45. 0.55.0The channel follows the traders, not us2026-09-19
    • The Telegram channel now carries the watched traders' own buys and sells instead of the positions our bot opened. When one of them later sells, it posts as a reply under their own buy, so a token reads as one thread.
    • Not every buy: watched wallets make around 2,400 a day, which nobody can read. A card goes out when a trader whose copies we have measured into profit buys at real size — about 230 a day, against the 160 the channel posted before. Both filters are settings.
    • What our own book does is unchanged and still on the signals page, with every entry, exit and skip and its reason. It simply stopped being what the channel is about.
46. 0.54.1Every venue's trades counted2026-09-19
    • A trader's own on-chain record only counted a buy on a venue whose trade event we had seen before, so buying on the wrong exchange quietly went missing from their numbers. Every exchange on Robinhood Chain was sampled and three more trade events added — Pancakeswap V3 and Giga V3, Orvex V4 and Pancakeswap Infinity, and Ramses DLMM.
    • Copy trading was never affected by this: a buy is proven by who authorised it and whose money moved, not by recognising the exchange, so a trade on a venue we have never seen is still followed.
47. 0.54.0No two tickets alike2026-09-18
    • Every live buy is now nudged by a small random percentage — a fraction of a percent either way, drawn fresh each time, applied after every other multiplier. A ticket of exactly the same size every time is a signature anyone watching the pool can pick out of the flow; no two are alike now. Paper and record buys are untouched, so the measurements stay clean.
    • Market cap is a cohort now, like the hours of the day — but it can only add size, never take it away: a deeper market can buy more, while a thin one simply buys the normal ticket. Admin → Cohorts holds both axes, each beside what the record book says it has returned. All four bands ship at the normal ticket, so nothing changes until they are raised from the table.
    • Fixed: a live dex-book buy ignored the trading-day cohort its own paper row was sized by, and spent the full ticket in an hour meant to be reduced. It has been wrong since the cohorts shipped.
    • Fixed: when the ETH price read failed, the signals page quietly switched every figure from ETH to dollars for a minute and back again. The rate was known — it had been measured a minute earlier — so it is read back instead.
48. 0.53.0Which book is actually paying2026-09-18
    • See more statistics breaks the book down by where a position came from — a watched trader, Latitude, a DEX boost, or a crowd we held on conviction — with what each one made, how often it won, and a share-of-the-book pie beside it.
    • The same cut by hour of the day: the four cohorts the ticket is sized by, so it is visible which parts of the day are carrying the book rather than only which traders are.
    • How positions ended now says what each kind of exit cost or made, not just how many there were.
    • The handbook has chapters for Altitude and Latitude, which never had one, and a pass over every page that had gone out of date. Thresholds came out of it everywhere: outcomes are published, the numbers behind the gates and the exits are not.
49. 0.52.2How a trader does on their own trades2026-09-18
    • The traders board shows each wallet's own on-chain record beside our copy record: the median return of their priced buys and how many buys we have read from the chain. 891 of the 955 wallets have one.
    • Every column sorts — FOMO's 30-day number, the on-chain return, copies, last buy — and clicking the sorted one again returns to the default, which is still what copying them returned for us.
50. 0.52.1Looking a trader up2026-09-18
    • The traders board is paged instead of endlessly scrolling, and has a search box: type a handle or the start of a wallet address and it filters the whole shortlist, not just the rows on screen.
    • A trader whose wallet also turns up opening launches alongside the same other wallets now carries a cabal label, with a filter to show only those.
51. 0.52.0A position remembers its wallet2026-09-18
    • The trading wallet can be changed and changed back. New buys go to whichever wallet is active, and any position already open keeps selling — and topping up — from the wallet that bought it, so switching never strands a bag.
    • Admin → Signals bot lists every wallet ever used, marks the ones still holding open positions, and can make any of them active again.
52. 0.51.1Quiet buying, written down2026-09-18
    • We now record when several watched traders buy the same token within an hour, and whether they were buying into weakness or chasing a rise. Measured over 904 of these: buying weakness with four or more traders peaked 40% higher within six hours and still closed in profit, while chasing a rise spiked further and gave it all back.
53. 0.51.0Positions that follow the crowd2026-09-18
    • When other watched traders buy into a token we already hold, the position stops selling on the rungs built for a token nobody else wants. Three traders in and it takes half at 3× and the rest at 10×; five or more and it takes a third at 5× and rides the rest on a wider trail.
    • The boards carry their numbers on a phone now — market cap, liquidity, buyers an hour, watched traders, holders and age on one line under each token, instead of a symbol and a score. A token page says how many watched traders have bought it, and what that means for a position held there.
    • The crowd is counted as it forms behind the entry, not at the moment we buy — measured over a thousand positions, the crowd beforehand says nothing, because we are usually first in. The crossing is recorded on the position and shown in the activity log.
54. 0.50.0What left the list2026-09-18
    • Altitude and Latitude can show the last day's departures, greyed out under the live rows: tick "include the last 24h". Each says why it left — it climbed from Latitude to Altitude, or it rugged — and what its market cap has done since.
    • Tokens that simply stopped ranking are left out: 274 passed through Latitude in a day and most of them just faded, which is the list working rather than news.
55. 0.49.3Clocks that only look forward2026-09-17
    • A time stop now closes a position that has run its time, not one that was opened long before the rule existed. When a fitted exit changes, open positions are judged on price from there rather than swept out on age.
    • A conviction position is never closed by a clock at all — it is held on a crowd, so it ends when its price says so.
56. 0.49.2Conviction gets an exit2026-09-17
    • A conviction position — one four tracked traders still hold — no longer rides with nothing to close it. It keeps ignoring the seller who left, and now takes the same fitted exits the Latitude book uses: out at the model's multiple, under its floor, or at its time stop.
    • Latitude can enter a token again whenever its list says so. The cool-off after a sale is there to stop the copy book chasing one trader in and out of the same coin; a place on the list is a fresh reading of the whole market, so it no longer applies there.
57. 0.49.1A full book stops meaning a sale2026-09-17
    • When the book is full and a strong signal arrives, what gets sold changed: whatever is going nowhere first, then a position in profit, and one in loss only when there is nothing else. Measured over 931 of these, cutting the loser was the expensive end — it booked the loss and gave up the bounce.
    • The live book can also hold more positions while the wallet can pay for them, up to a ceiling, instead of selling one to fund the next. Everything it holds keeps its stop and its trail.
58. 0.49.0The trading day2026-09-17
    • The hours of the day are now a setting of their own, and they apply to every book — the traders’, the dex book and Latitude. Each cohort spends a share of the normal ticket, and a cohort set to zero is closed.
    • Measured over 2,187 positions that were copied regardless of the clock: Asia returned +19%, Europe +7%, the US hours +40% and +43%. All four make money, so none is switched off — Asia and Europe now trade at a smaller ticket instead of being refused outright, which is what they had been.
    • What a token looks like and when we are awake are now separate rules: the measured market-cap and pair-age cells apply at every hour rather than only in the afternoon.
59. 0.48.1One line per token2026-09-17
    • The closed list groups by token: a token traded five times is one line carrying the whole result, with the individual closes one click underneath.
    • Under the hard stop, keeping a position while the trader still holds is now a switch rather than a rule — with the measurements behind it beside it in Admin → Settings.
60. 0.48.0Vouchers2026-09-17
    • Premium can be redeemed with a voucher code. Enter it on the membership page and the days land on your account straight away — added to any time you already have, never restarting the clock.
    • Each code works once, for one account. An admin mints them one at a time with their own length and expiry.
    • The Latitude book gained a real-money switch, shipped off: it has traded on paper only, and that is where it stays until it is turned on deliberately.
61. 0.47.1The Latitude book gets its own console2026-09-16
    • Admin → Latitude bot: the paper book that trades the Latitude list now has its own page — ticket, open slots, entries an hour, how far down the list an entry may come from, and a minimum liquidity — beside what the record says about each of them.
    • Measured on its first 44 closed positions: +$140 on a $10 ticket, 26 of 44 up, median close 1.28×, median hold 18 minutes. Place 5 on the list was the only losing cohort, so entries now come from the top four.
    • Every token that enters a list is snapshotted at that moment, not just on the quarter-hour, so the next read of what makes a good entry has the numbers it needs.
62. 0.47.0Paying for what only real money needs2026-09-16
    • The live price feed now follows live positions only. It was subscribed to every pool the paper and record books held — the single biggest line on our data bill — to make marks fast for positions that are not real. Paper and record mark every five minutes from the market price and one sell quote, which is what their results were always measured on.
    • The live book is untouched by all of this: it keeps the price feed, its ten-second marks and the paid data provider in every mode, on every kind of position — copied, dex or radar — and a new live position is now heard within ten seconds of opening instead of thirty.
    • Paper positions stay as accurate as they were: their price refreshes every 30 seconds from the market feed, and a move of more than 15% buys a fresh quote on the spot instead of waiting for the next one.
    • We now watch our own monthly data budget and step down before the provider cuts us off: past 80% the background reads (scans, token metadata, viewer pages) move to the public node, past 95% everything but the live book does. Both are slower, not stopped. The state is on Admin → Health and it says so on Telegram when it changes.
63. 0.46.4Resume works2026-09-16
    • Fixed: after an outage suspended the system, "Resume — the cause is fixed" closed the incidents but left the suspension standing, so the site stayed in maintenance. The same empty-value bug that the bot's Resume had, in the switch beside it.
64. 0.46.3No corpses on the lists, no phantom losses2026-09-16
    • Altitude and Latitude drop a token that rugged: more than 90% under its own high, or worth less than half the market cap Latitude first spotted it at.
    • Fixed: a sell quote that came back far under the pool's own price was taken as the price. One paper position was stopped out at −94% on a market that had only halved. Such a quote is now distrusted, the way a nonsensical buy quote already was.
    • Take-profit moonbags are off: a take-profit sells the whole position. Measured over 238 of them, keeping 20% cost money — the bag ends at 0.59× of what it was worth at the take-profit, and only a quarter of them ever rose. A hard stop still parks while the trader holds.
    • On the signals page, the time window now filters open positions too, not only closed ones — the tiles and the statistics panel, which also shows open P&L now.
65. 0.46.2Traders in search, spotted on Latitude2026-09-15
    • The traders page loads as you scroll instead of all at once.
    • Search finds traders too: type a handle (or @handle) or the start of a wallet, and members get the matching traders above the tokens.
    • A token on Altitude that was on Latitude first says so — spotted on Latitude since its market cap then.
    • Buy buttons on Alpha, Altitude, Latitude and Signals keep to their own column instead of covering the one next to it.
66. 0.46.1Altitude and Latitude2026-09-15
    • Radar is now two pages with their own menu entries: Altitude — what Robinhood Chain is buying right now — and Latitude — young tokens gathering buyers early. Old Radar links open Altitude.
    • Fixed: after its first update the lists stopped refreshing, so market caps and the tokens on them stood still. Both lists now update every minute, and each row shows whether it climbed, fell or is new.
67. 0.46.0Radar, rebuilt: Hot and Potential2026-09-15
    • Radar now has two lists. Hot RN on RH ranks what Robinhood Chain is buying most widely right now, weighed by the chance the price holds. Potential finds young tokens gathering buyers early.
    • Both are scored by models trained on our own history of every Fomo buy — which tokens ran and which died — refitted daily and judged on days they never saw. Each row names what lifted it.
    • Every token that reaches the top of a list is tracked: the market cap and time it entered, and what its price has done since. Potential shows its results openly: how many entries reached 2×, and how many halved.
    • Watched traders count on both lists — more when our copies of them have made money.
68. 0.45.1Every named Fomo trader, watched2026-09-15
    • The signals agent now watches every Fomo trader we can name — 266 more wallets, from their Fomo accounts — not only the ones on fomoscan's leaderboard. How copying each one performs decides whether they stay, the same as for everyone else.
    • Trader handles read from a Fomo account are the handle itself ("dioz", not "dioz_on_Fomo"), so their Fomo links work.
    • 330 more active traders are watched again: they had been switched off on a measure of their own buys that we no longer use. Copy performance decides from here.
69. 0.45.0Positions that react on the trade2026-09-15
    • Signal positions are re-priced on every trade in their pool, and a stop or a take-profit is checked the moment the price gets there — not up to 45 seconds later. Every sale is still decided on a real quote.
    • A position's high is tracked trade by trade, so a trailing exit measures its giveback from the real peak.
    • Wallet pages show ETH, WETH and USDG balances with their dollar value, and your account's portfolio values WETH and USDG instead of showing a dash.
70. 0.44.1Feeds that stay right2026-09-15
    • New launches are no longer dropped from market updates in their first minutes just because no data provider had indexed them yet.
    • A quick background re-scan no longer erases what a full scan found — the opening bundle, pool liquidity, who can pull the liquidity and the sell test.
    • Radar scores a token whose own deployer can pull its liquidity the same way the token page does, and always shows its newest board.
    • Alpha and Radar show the same, freshest holder count as the token page, and USDG pools show their real liquidity.
    • Paid: a new boost starts a new entry market cap, "since entry" compares like with like, profiles are no longer pushed off the board by a burst of boosts, and unknown boost counts no longer land in the ≤10 bucket.
    • Tokens first found through Radar or promotions are dated by when their pool was created, not by when we came across them.
    • Alpha keeps "Load more" after a live update, and live pages no longer reload the moment they open.
71. 0.44.0Your trades, counted from the chain2026-09-15
    • Real trades now count in your P&L only once they confirm on chain from your wallet, with the token amount read from the transaction itself. A swap that reverted no longer shows up as a fill, and the trade box waits for confirmation before saying "Bought" or "Sold".
    • P&L is worked out per position: selling out and buying the same token again starts a fresh cost basis, instead of averaging every buy you ever made.
    • Demo: a double-clicked sell can no longer be credited twice, and a position that can't be priced for a moment is counted at cost rather than dropping out of your totals.
    • The portfolio page follows the same demo/real rule as the header when real trading is switched off, and the header no longer shows your on-chain ETH next to the DEMO badge.
    • Wallet pages list every holding of wallets with more than a hundred tokens, say so when balances couldn't be read instead of showing an empty wallet, and no longer value holdings with prices older than a day.
72. 0.43.0A chart that draws one price2026-09-15
    • The token chart now draws one price per minute — the pool's own whenever we have it — instead of mixing several data providers. Flat markets no longer zigzag, and candle wicks are real moves, not the gap between two APIs.
    • Tokens the signals bot holds get their price recorded from every trade in their pool, whether or not anyone has the page open.
    • Switching window or token no longer leaves the old series on screen, candles keep your zoom while they update, the log scale no longer blanks the chart, and times show in your own timezone.
    • For a token we started watching late, the earlier part of the chart comes from GeckoTerminal, priced in the token itself, and ours takes over from our first reading.
    • Trade markers show the current book's newest trades and sit on the candle the trade happened in.
    • A price that failed to refresh is no longer shown as fresh, and v4 tokens keep their 1h/6h volume and trade counts between refreshes.
    • Market data refreshes the most liquid tokens first: a large pool no longer waits behind tens of thousands of dust tokens for its next price.
73. 0.42.0Trader numbers you can trust2026-09-15
    • Fixed: a trader's realized P&L could count tokens they were simply sent, or moved between their own wallets, as buys and sales. A trade now has to be a real swap in its transaction, and every trader history is rebuilt once under the new rules.
    • Fixed: a sale now books only the proceeds of what was actually held, and a leftover crumb of a token no longer keeps a round trip open.
    • WETH legs are priced at the ETH rate of their own minute, not today's.
    • A trader's copy record reads the same on the leaderboard, their page and the shortlist.
    • Skipped signals are no longer shown publicly.
    • Traders we watch now load their recent history from our own chain feed, which also no longer loses blocks when the service restarts.
74. 0.41.1Re-priced on time again2026-09-15
    • Fixed: for a few hours after 0.41.0, a copy-book position could be re-priced only every five minutes instead of within seconds of its pool trading, so a stop, a trailing stop or a take-profit could fire late. Every position is re-priced on its own schedule again.
75. 0.41.0A pickier copy book2026-09-15
    • The copy book now opens positions only on the kind of token, at the time of day, that has actually paid in its own record — fewer trades, and the ones it takes are the ones worth taking. The Signals channel follows the book, so it is quieter for part of the day.
    • Nothing about the watching changed. Every buy by every trader we follow is still recorded and still counts toward that trader's record on the Traders page, and alerts for the traders you star still arrive for every trade they make.
76. 0.40.1The menu answers again2026-09-15
    • Fixed: on a phone, the menu button on Paid could stop responding while you were signed in. Paid, Radar and Signals were quietly reloading themselves in a loop, about once a second, for as long as the tab stayed open. Now they refresh only when something actually changes.
    • Fixed: the dex book kept re-reading its newest DexScreener boost every 20 seconds and logging it as “seen too late”, which buried every real decision in the activity log. Each boost is now judged once, and the late-event cutoff is 5 minutes, up from 3.
77. 0.40.0The token page, rebuilt as a board2026-09-14
    • A token page is now three columns on a wide screen: the chart in the middle, every reading down the left in one line each, and the things you act on — the trade box, notable holders, the holder list, the deployer — down the right. The trade box stays with you as you scroll. On a phone it is the same page in one column, chart first.
    • New under the chart: the tape. Every transaction in the token, newest first, with the wallets we follow named — shortlist traders, their suspected side wallets, our own bot, and your wallet. Switch it to “wallets we follow” to see only theirs, reaching back past what the page has loaded into everything we recorded at the time.
    • The tape reads transactions, not transfers: a buy routed through a router or an aggregator shows the person who bought, not the contract in the middle, and the launch hook’s fee sweeps are not listed as somebody selling.
    • Values on the tape come from our own price history, so a trade from a stretch nobody was watching reads “—” rather than a number we cannot stand behind.
    • The trade box is a proper one now: Buy and Sell tabs, size presets, a max that knows what gas will cost, and a live quote that says what you would actually receive and how far that lands from the pool’s mid price — price impact and the launch pool’s 2–4% swap fee together, because both are money. Sell by percentage or by number of tokens. It shows what you hold and what it is worth.
    • Nothing about the protections moved into the box: slippage, the moonbag, the rug check and the pre-trade simulation stay in your account settings and are applied server-side, and a trade still asks before it leaves your wallet.
78. 0.39.1The market cap everyone else sees2026-09-12
    • Fixed: market caps on the copy book read a few percent low, and entry market caps a few percent high. Robinhood Chain launch pools run a hook that takes a swap fee on both sides of every trade — measured at 4% on one token, 3% and 2% on others — and it does not appear in the pool's own settings, only in what a quote comes back with. Every market cap was being worked out from a quote, so it carried that fee. MARLIN showed $219.8K against a real $229K.
    • Market caps now come from the pool's own mid price, read straight off the chain. They agree with DexScreener to within a fraction of a percent, and on a fast mover they are fresher. Profit and loss is unchanged and still counted from the prices actually traded at — that fee is real money, paid on the way in and again on the way out, and the figures say so.
    • A watched wallet's buy is now found within a minute even if the live feed misses it. The feed sees a trade in about two seconds; behind it, a poll worked through the watch list four wallets a minute, which at 180 wallets is a 45-minute round trip. Eleven trades in three days were found by that poll instead of the feed, one of them 22 minutes late. One read now covers every watched wallet at once, and says so in the log when it catches something the feed dropped.
    • "Spotted Ns after" is now measured from the block the trade landed in rather than from when the notification reached us, so it says what it claims to.
79. 0.39.0Premium2026-09-12
    • The copy book, the trader shortlist and the Telegram channel are now premium: $49 for 30 days, paid on Robinhood Chain from the wallet you already trade with. USDG to the cent, or the same in ETH at a price quoted before you sign. No card, nothing to cancel — when the days run out it simply stops.
    • Paying again adds 30 days on top rather than restarting them, so renewing early never costs you anything. A payment always asks before it leaves your wallet, every time, whatever you have ticked elsewhere. If the page loses your confirmation after the money moves, nothing is lost: it is credited automatically within a couple of minutes.
    • Free accounts keep Alpha, Radar, Paid, one-click trading, and Telegram alerts for the traders they star. Signals and Traders show the same real headline statistics they always did, with the rows as a preview.
    • The signals channel is now private, and everyone already in it stays in it.
    • Fixed: the live bot counted moonbags, parked rows, dust copies and conviction holds against its open-position limit, so a handful of them could quietly stop it buying while the paper book beside it kept going. The limit now counts what the paper book has always counted — positions it would actually open today.
    • Positions get an actions menu: close, or double — buy again for what the position is worth right now, at today’s market cap. The entry becomes the blended average, so take-profit, the trail and the stop all re-reference it. The expand arrow is bigger.
80. 0.38.0A front door, and the boards behind the door2026-09-11
    • hoodini.sh is now a page that explains what this is: what each board does, how a launch gets read, and the real counts behind it. The feed moved to /alpha.
    • Alpha, Radar and Paid join Signals and Traders behind a free account. Signed out, the headline statistics stay real and the rows are placeholders — nothing behind the blur is a real token, so there is nothing to un-blur.
    • The boards are live over a stream instead of a timer. Alpha used to re-ask the server for the whole table every 5 seconds per open tab; now the numbers arrive when they actually change, and a quiet market costs nothing. With nobody watching a board, the market data behind it stops being fetched at full speed — which leaves more of a very tight GeckoTerminal budget for the feed everyone is actually reading.
81. 0.37.0Every bought token, remembered as it was2026-09-11
    • Each position now leaves three snapshots: what the token looked like within seconds of the entry (market size, liquidity, pair and token age, volumes, holders, safety readings, launchpad, route, the crowd around it and the signal itself, plus the whole token page and DexScreener pair list as raw JSON), what it looked like at the close, and what it did on its own in the 24 hours after. The dataset behind finding lookalikes straight from the alpha scan — see docs/position-snapshots.md.
82. 0.36.4What arrived is read at the block it arrived in2026-09-11
    • A live fill's proceeds are now read from the wallet's balance at the receipt's block and the block before it, never from a read made moments later: a node one block behind returned NEMA's pre-sale balance, so a sale that returned exactly its quote was booked at the gas cost, −99%. A delta below the minimum the chain enforced is treated as a bad read and the quote is booked instead. The NEMA slice is corrected from the chain: +24.3%.
83. 0.36.3The leaf turns2026-09-10
    • The logo is now the neural-network leaf turning about its own tilted axis — rendered from the mark's own 77 nodes and 152 edges, not drawn by a model. A size larger in the header without a taller bar; hover it (or prefer reduced motion) and it settles into the flat mark with the spine and hubs in brand lime. The maintenance page and the handbook carry it too.
84. 0.36.2A position's whole story, and honest top-ups2026-09-10
    • Every position row has a history toggle in its last column: buy-ins with their market cap, gas and slippage, slices sold, parks, the close or the void, the bot's failed attempts — oldest first, each with its transaction on the explorer.
    • A live top-up (a dust upgrade, a moonbag re-open, a conviction add) now replaces the agent's quote-based estimate with the receipt's numbers instead of adding on top of it; STONKFLY had $14.78 booked for $9.85 spent.
    • Rows with several buy-ins say 'first MC → avg MC'. The average is the row's entry — the reference every P&L, take-profit and stop uses — and the Entry MC column shows it.
85. 0.36.1A market must quote before it is the market2026-09-10
    • Market resolution walks DexScreener's pairs deepest first and takes the first one the quoter actually prices, with a short list of known pools as the net. This afternoon a new USDG/WETH pool with $156K of listed liquidity, no swaps and a hook the quoter refuses became USDG's top-ranked ETH pair, and every USDG-quoted route stalled on it (RocketFrog, KIRKINATORX). The native USDG pool still quotes and is the market again.
    • A v4 pool keyed on WETH is now composable: wrapped on the way in, unwrapped on the way out, the router holding the WETH in between. v2 pairs stay on the legacy router path and cannot sit inside a chain.
86. 0.36.0External API health, with a hand brake2026-09-10
    • Every outbound call to Alchemy, the public RPC, Blockscout, DexScreener, the Relay stream, GeckoTerminal, GoPlus, Arkham, fomoscan and Telegram now reports its outcome. A provider that fails for minutes on end with no success in between, or answers credits/auth twice, opens an incident with the provider's own error text and pings Telegram.
    • A provider we cannot run without (Alchemy, Blockscout, DexScreener, the Relay stream) suspends the system: the worker skips every tick but its heartbeat and health probe, the bot sends nothing, the site shows a maintenance page to everyone but admins. Auxiliary providers only alert, and their incidents close themselves when they answer again.
    • Admin → Health lists incidents, shows the suspension state, and has the two switches: Resume once the cause is fixed, Suspend by hand. While suspended, a probe every minute reports when a provider is back. Alerts go to TELEGRAM_STATUS_CHAT_ID and to every admin who linked Telegram on their account page.
87. 0.35.4Details behind an info glyph, a proper close dialog2026-09-10
    • On desktop the Detail column of every position tab and the reason column of the activity log sit behind a small info glyph; hover or focus it for the full text in our own tooltip. Phones keep the text inline. A live row's in-between state (buying…, selling…) stays visible next to the glyph.
    • The admin close button asks in our own dialog — what closes, how, and whether it is a moonbag, a dust copy or a conviction position — instead of the browser's confirm box.
88. 0.35.3Activity rows link to the chain2026-09-10
    • Every activity row with a transaction behind it carries a small green link to the explorer: the trader's buy on a signal or a paper entry, the bot's own fill on a live entry once it lands, the bot's sell on a live take-profit or close.
89. 0.35.2Live rows are marked faster2026-09-10
    • A live position's pool that just traded is re-quoted within 10 seconds (the paper book keeps its 45s floor), an idle one every minute instead of every five, live moonbags included — stops and take-profits on real money fire on a fresh mark.
90. 0.35.1Live book first, records on paper2026-09-10
    • The signals page shows the live book while the bot is live (admins can still open the paper book with ?book=sim). Trader records — the traders page, the admin shortlist, a wallet's copy history and a token's copy activity — read the paper book whatever the bot mode: it copies every signal, so it is the complete record and it never resets on a switch.
91. 0.35.0Two books, side by side2026-09-10
    • The paper book takes every signal, always — it is the record every trader is judged on. The live book takes the same signal beside it while the bot is live, under its own caps and the wallet's means. Until now the agent ran one book at a time: switching the bot live had moved every new signal, mark, record and board to the live book, and the paper book stopped growing (17:00 UTC today).
    • Trader records, the weekly record verdict and the conviction sweep read the paper book. The pages show the paper book by default; admins reach the live book with ?book=live. The Telegram channel posts the paper book's entries and take-profits, so a signal is one card, not two.
    • Switching the bot on no longer closes or restarts anything; switching it off closes the live book's open rows at their last mark, as before.
92. 0.34.2Sells through v3 pull first2026-09-10
    • A live sell whose first hop is a v2/v3 pool now pulls the token into the router through Permit2 and pays the hop from there: this chain's router refuses a user-paid v3 hop before it reaches the pool (CRUDECAT's moonbag sell, 12 refusals). A v4 first hop keeps settling through Permit2 by itself, as every live v4 sell has.
93. 0.34.1Every live transaction, as the receipt says2026-09-10
    • Each live transaction now records what the receipt and the wallet say: the fee in wei, what actually arrived against what the quote promised, and the slippage between the two. The fill that opens a live position carries its own gas and slippage.
    • Every P&L figure is net of the gas paid — the realized totals, the open book, each trader's record (which sizes the next ticket and can switch a wallet off), the daily-loss halt and the Telegram digest. Paper rows pay no gas, so nothing changes there.
    • Live rows show their real costs in the detail column: gas paid so far and the fill's slippage.
94. 0.34.0One transaction per route2026-09-10
    • The bot sends a multi-hop route (ETH → USDG → NVDA → token) as ONE Universal Router transaction: the hops in between never touch the wallet, one simulation covers the whole path before anything is sent, and a hop that would refuse reverts the lot. Hop by hop, two live buys had filled their first hops and stalled on the last, leaving the quote asset behind.
    • The v4 swap now carries the router's sqrtPriceLimitX96 word. Without it every pool quoted in an ERC-20 (USDG, a tokenised stock) reverted before the swap while ETH-paired pools happened to work — this, not the pools' hooks, was why those last hops refused. Measured against working swaps in the same pool; proven with real sends on testnet and simulated on every recent mainnet route.
    • No position, paper or live, opens on a token the bot cannot actually buy: the exact transaction is simulated before the paper book enters, and a refusal is logged as a skip with the reason.
    • A live sell reads what came back from the chain: ETH received is the wallet's balance delta plus the gas paid, tokens received the balance delta — never the quote.
    • Signals rows show the ticket size as a badge in every tab (boost ×1.5, reduced ×0.5); the header no longer explains the position size.
95. 0.33.1A wider stream, a properly cased wallet2026-09-10
    • The Relay stream reads up to 80 pages a pass instead of 40: the US session produced more than 2,000 requests per two minutes and a third of evening passes were being sampled, so some fomo buys never reached the leaderboard.
    • The bot's wallet in the admin console is printed in EIP-55 mixed case, the standard way to show an address — the chain itself ignores case.
96. 0.33.0Moonbags age out2026-09-10
    • A moonbag now carries the time it became one. Twelve hours later its next mark is stored as a reference; from then on a 50% rise off the reference (the market cap test — supply is fixed), a P&L of +200% or -85% against the blended entry, or an age of a week closes the whole bag. Parked rows count as moonbags. Five knobs.
    • An anonymous account's fomo link goes to fomo itself, with our referral, never to a made-up profile — on the traders page, the signals rows, the token and wallet pages and the Telegram cards.
97. 0.32.1No buy-back for four hours2026-09-10
    • A token the book exited entirely is not bought again for four hours, whoever signals it (a knob). A moonbag is not a full exit: a signal on one still tops it up, now at half a ticket instead of a full one.
98. 0.32.0What the first 111 positions taught the book2026-09-10
    • A review of 24 hours of paper entries (111 positions, 31% won, +$154) found the losses in three places: pairs older than a day (-$280 against +$483 for fresh ones), crowds of four or more shortlist wallets already in (-$136 against +$281 for two or three), and runs that peaked between 1.1× and 1.5× and were given back under a trail that armed at 2×.
    • So: a copy needs a pair under 24 hours old unless the buy is $2,000 or more; four or more wallets in within a day is a crowd we skip; the trail arms at 1.4× with a 25% give-back, never below the entry. All three are admin knobs. Re-run on the same day, the first two alone would have made +$534 on 31 positions.
    • Every mark is now kept for 30 days, so the next review can test exit rules against the price paths instead of peak and close.
99. 0.31.3Moonbags mark on the sweep2026-09-10
    • Moonbags and conviction rows are re-quoted on the five-minute sweep only, never on a trade in their pool: no rule sells them, so a fresher mark bought nothing but chain calls — two thirds of them last night.
100. 0.31.2The channel prints buys and profits2026-09-10
     • The Telegram channel now posts the entry card and the take-profit lines under it, nothing else. Stops, trailing exits, the trader's own exit, parks and admin closes are no longer posted; the signals page keeps the whole record, losses included.
101. 0.31.1Less DexScreener in the signal path2026-09-09
     • A token's resolved market is remembered for fifteen minutes instead of one, and forgotten the moment a quote through it fails — the book's marks alone were asking DexScreener 45,000 times a day. The book asks for its liquidity batch at most once a minute, and only when a position is due for a mark, instead of every ten seconds.
     • What DexScreener does not answer within five seconds, our own index answers: the symbol, the quote side, the graduation time, and a price and liquidity when the market tick refreshed them in the last fifteen minutes. A slow DexScreener no longer holds a signal.
102. 0.31.0A higher bar, a steadier list, a record that can switch a trader off2026-09-09
     • Self-deduced traders: the bar now asks for a median of 5% or better, a positive weighted return and a positive score, not a positive median alone — the first afternoon admitted accounts at 0% medians and -30% weighted returns. The list moves only after two hourly ticks in a row over or under the bar; a third of it had been flipping every hour.
     • Records switch traders off: one that lost more than 80% of its last 20 copies and is in the red is disabled by the hourly record tick, kept for its sells, and gets one probe copy a week; when the last 20 improve it comes back on its own. A losing record still copies at half a ticket, a proven one at one and a half. The separate 25% bench is gone.
     • Never buy a market: a token above $20M of market cap and older than a day is skipped for every signal, every trader, every book. Both numbers are knobs.
     • Arkham labels become handles with letters, digits and underscores only.
103. 0.30.3Side wallets: fairer exclusion2026-09-09
     • A candidate is excluded as a launch sniper for being ahead of strangers only when that count matches or beats the tokens it led its own trader on — a wallet that fronts its trader on twenty tokens and strangers on two is the trader's. An exclusion now holds for two days instead of forever, so fresh evidence gets a fresh look.
     • Evidence that already meets the bar is evaluated every minute, a few pairs at a time, instead of waiting for the candidate to appear at a new launch.
104. 0.30.2Admin: traders, said plainly2026-09-09
     • Our own leaderboard sorts the accounts that clear the bar first and says what each row is: on the shortlist (with source and on/off), over the bar and how long until the 24-hour age rule admits it, or scored but below the bar — which is not followed.
     • A leaderboard candidate whose wallet fomoscan cannot resolve can be added by pasting the wallet right next to the Add button.
     • The signals header shows the paper ticket in ETH as well, and for admins the bot's own ticket beside it — they are two different knobs.
105. 0.30.1Wider pages2026-09-08
     • Every page now uses the full header width, so tables carry more. On the signals book the token and source column got the room and the detail column gave it up.
106. 0.30.0About, security, legal2026-09-08
     • Three plain pages, one hop from the footer: About — what hoodini is, who builds it (CHRONIC QUAL SRL, Romania) and how to check us; Security — the server composes a swap and never signs, your embedded wallet does, and the four contracts a trade may ever touch, with addresses; Legal — terms, privacy and the risk disclosure in plain words.
     • Machine readers get the same story: schema.org Organization and WebApplication data in every page's head, /.well-known/security.txt, /llms.txt, and a sitemap that lists the durable pages.
     • The version badge says what 0.x means: the copy book is a paper model; 1.0 ships with the live book.
107. 0.29.1Buy on fomo2026-09-08
     • The token page's one trade button now opens the token on fomo — where the traders we follow actually trade — with our referral. Axiom is gone from the page and the code.
108. 0.29.0Buying again2026-09-08
     • A token traded and closed before can be traded again: a fresh buy from a tracked trader opens a fresh position, and the entry names the earlier one.
     • A moonbag bought again is a position again — the take-profit remainder or a parked loser gets a full ticket on top, the flags come off, the ladder restarts, and it follows the trader who just bought. The entry becomes the cost-weighted average of everything held: that blended price, and the market cap it implies, is the reference every later take-profit and stop is measured from.
     • A conviction position takes the ticket on top and stays conviction, blended the same way — never above our entry times the crowd cap, never past the conviction size cap.
     • Admin closes read "closed by admin", nothing more.
109. 0.28.3Search by name2026-09-08
     • The search box suggests as you type: tokens whose ticker starts with, or whose name contains, the letters — with market cap, age and the address tail, because copycats share a name. Click one to open it.
     • Enter on a unique ticker goes straight to the token; anything with several matches lists them, biggest first. Addresses resolve as before.
     • Under the hood, a trigram index over 320,000 tickers and names: a lookup went from a 96 ms table scan to a few milliseconds.
110. 0.28.2Token page header, reorganised2026-09-08
     • Identity first: symbol, name and badges; the address, when it was spotted and graduated, decimals; then one quiet row of places to look — explorer, Dexscreener, fomo, the launchpad, the project's own links. One primary action on the right: Buy on Axiom for a pool, the launchpad while on the curve.
     • The scan button moved to the verdict strip, now directly under the market numbers, next to momentum and the time of the last scan. Scanning is a measurement, not a link.
111. 0.28.1A wider paper book2026-09-08
     • The paper book may now hold 20 positions; the live book keeps its own cap. Each is an admin knob. At ten, the paper book was displacing its 23rd-best position at 0.76× to make room — the cap was deciding exits, not the trader.
     • Signals activity log: filter by kind — entry, take profit, exit, park, skip and the rest — with counts, and pages instead of a 60-row cut.
     • Mobile menu sits on a solid background.
112. 0.28.0Our own leaderboard2026-09-08
     • fomoscan's leaderboard froze on 3 September, so the traders now come from the source itself: Relay's public list of every fomo request touching Robinhood Chain. A two-minute tick keeps every fomo buy of $50 or more that lands on the chain, folded per account, token and day, and every record we ever fetch is kept so no transaction is looked up twice.
     • Every hour each account with recent buys is scored at today's prices — what copying its buys of the last seven days would have returned: median and weighted return, wins, breadth. Ten buys of $200 or more with a positive median admits an account to the shortlist as self-deduced, up to 150 watched wallets. Arkham names the ones it knows; the rest are anonymous, marked as such everywhere a handle appears. Admin shows the whole board.
     • Side wallets: two accounts whose fomo buys are paid by the same wallet are one person. A candidate that shares its trader's payer is confirmed on sight, no timing threshold.
113. 0.27.1One trader, several chains2026-09-08
     • Every shortlist row now carries its chain. A trader's other wallets — the Solana wallet on their fomoscan profile, and the Base, BNB or Solana wallets that paid their fomo buys — are linked under the trader as their own rows, with a Chain column and an "other chains" filter in admin. Nothing watches those chains yet, so they cannot be enabled, but the payer checks that tell a trader's own Relay request from a stranger's already use them.
     • Cheaper marks: a position whose pool trades every block was re-quoted on the chain every ten seconds — 126,000 calls a day, 88% of our Alchemy spend. It now re-quotes at most every 45 seconds while trading and every five minutes otherwise, and every copy-book read is labelled in the spend accounting.
     • Cheaper scans: the first automatic look at a launch, thirty seconds in, no longer asks Blockscout for holders and source — the explorer has not indexed them yet and the reads came back unknown. The twenty-minute re-scan makes them once the token is still alive. That was 33,000 Blockscout calls a day, half our credit spend.
114. 0.27.0A buy is a buy on any venue2026-09-08
     • The buy rule no longer depends on recognising the venue. A buy is three facts: the wallet authorized the action — it signed the transaction, or its UserOperation is in the receipt, or its account executed a call inside the transaction (read from the call trace), or its own WETH, USDG or the pair's quote token left it, or fomo created the Relay request, or the wallet verifiably funded a Relay request on the origin chain; the wallet's money moved; and the token really arrived, at most two final holders. Swap events are evidence for the cost, not a gate, so Pancake, Ramses, Giga, Ekubo, Orvex, Bankr, o1 and the launchpads count the same as Uniswap.
     • An authorized action without payment — a claim, a mint, a self-pulled airdrop — is filed as a claim, never copied. A token's own logs prove nothing: a Swap it emits about itself and zero-value transfers are ignored, which also removes address-poisoning noise from the book.
     • The same rule now decides sells: an outbound transfer counts as the trader selling only when the trader authorized it. It runs identically for traders and side wallets, on the live feed, the poll, and the side-wallet evidence. Relay requests that a trader funds from Base, BNB or Solana are verified on those chains through Alchemy once the networks are enabled on the app.
115. 0.26.2Side wallets earn it on receipts2026-09-08
     • A wallet that keeps buying just before a trader was recorded from the token's transfer log alone. Before it is confirmed as a side wallet, each of its matched buys is now re-read with the same rule a trader's buy must pass — signed by the wallet, or its own UserOperation, or a Relay request fomo created — and only verified buys count; a delivery that merely looked like a buy is dropped from the evidence.
116. 0.26.1The whole book counts2026-09-08
     • Open P&L, the realized figures, the statistics and the charts now count every position — slot positions, dust copies, moonbags and conviction alike. The open tile still breaks the book down beside the total. Only the per-trader records stay slot-and-dust only: a conviction position belongs to the crowd, not to the trader who happened to buy last.
117. 0.26.0The stop parks, the trader closes2026-09-07
     • A position that falls under the stop is no longer sold while the trader still holds it. It moves to the Moonbags tab as a parked position — its slot is freed, no rule sells it — and it closes the moment the trader exits. The channel gets a line when a position parks.
118. 0.25.8Proven means a ticket, not a dollar2026-09-07
     • A trader's record counts as proven — and earns the bigger ticket — only with at least one ticket of realized profit behind it; three wins in five closes for a dollar no longer qualifies.
     • The favicon and the precomposed touch icon are served as files, so browsers asking for them no longer log router errors.
119. 0.25.7Only fomo's own requests2026-09-07
     • A Relay fill is the trader's buy only when fomo created the request. fomo tags its requests with its name and its 0.5% fee; a request made by anyone else carries no such tag, even when the trader's address is written into it as the requester. That is exactly what a Base pool wallet had been doing for days — funding buys for top traders with their own addresses filled in — and it passed every ownership check until now. Those fills are deposits again.
     • The hour-of-history rule is gone: fomo's tag is the proof, so a trader's first buy from a new funding wallet counts at once.
     • The handbook at docs.hoodini.sh caught up: a Copy trading section — the Signals book, Traders, the Telegram channel and bot, and a chapter on how a buy is told from a delivery, fakes included — plus the five boards, the routing, sellability and the providers as they are today.
120. 0.25.6A payer earns trust first2026-09-07
     • A Relay fill counts as the trader's own buy only when its payer has been on record with that wallet for at least an hour, besides paying for nobody else. This afternoon a spray payer's very first fill passed, opened a position on a ten-minute-old token, and then paid five more traders within twelve minutes; the token went to zero. Of the 48 genuine buys that day, 47 had a payer already known for an hour, so the rule costs nothing in practice.
     • Payers are matched case-insensitively — a Solana address is case-sensitive text, and the seeded week of history and the live rows now meet in one index.
121. 0.25.5Two bad entries, two fixes2026-09-07
     • SZR opened at a chain quote of $1.48 billion per token and stopped out 22 seconds later. Its Uniswap v4 pool is keyed on WETH, and the quote mapped the ETH side to native ETH, which flipped the swap direction. A v4 leg now uses the pool's own ETH currency, and a chain quote more than 20% away from DexScreener's price is not trusted: a ten-dollar quote moves no pool that far, so it is the wrong pool or the wrong direction, and the paper entry takes the market price instead.
     • The side-wallet hunt now windows only fresh launches: a trader's buy is examined for wallets that bought just before it only when the pair was younger than eight hours at the time — a knob, Side wallet: pair age limit. On an older pair the buyers ahead of a trader are just the market.
     • Several positions opened on buys that were four to six hours old: the Blockscout poll trails a wallet showered with sprays and surfaced its backlog as fresh buys. A buy first seen more than five minutes after the fact is recorded but never copied, and the poll now reads up to five pages per wallet per pass so a cursor catches up.
122. 0.25.4Relay fills are the buys2026-09-07
     • fomo's in-app buys reach Robinhood Chain as Relay solver transactions: the trader pays from a Solana or Base wallet, a solver delivers the token. Relay's public request record names the payer, and a buy now counts when that fill is addressed to the wallet by a payer who pays for nobody else — alongside the wallet's own signature and its UserOperation. Measured over seven days: 1,570 such buys across 58 traders, some 200 to 700 a day, all of which the previous two releases had filed as deposits.
     • A payer who buys the same token for several tracked traders within minutes is a spray, not a trader — 16 such payers, 813 fills last week — and stays a deposit, with the reason spelled out. The payer is kept on every row so the check reads our own history.
     • The position size is what the trader actually paid, from Relay's record, instead of an estimate from the token amount.
     • A fill passes the token through Relay's router and proxy before it reaches the trader, which the old distribution guard counted as three or four recipients and rejected in silence. It now counts final holders only, every declined candidate leaves a deposit row with its reason, and the live path retries once when an RPC or Relay lookup does not answer.
123. 0.25.3Benched means smaller, not off2026-09-07
     • A trader whose copies keep losing is no longer switched off. They stay watched and copied at a quarter of a ticket — a knob, Benched trader ticket, under Signals bot; 0 restores the weekly probe.
124. 0.25.2Smart accounts, and a clean slate2026-09-06
     • The tracked wallets are smart accounts: their own actions arrive as UserOperations bundled by a third party, not as transactions they sign. A buy now counts when the wallet signed the transaction or the transaction carries the wallet's UserOperation — either EntryPoint. Without the second half, the previous release saw no buys at all.
     • Measured over the last seven days across every tracked wallet: 7,865 UserOperations, of which about 6,900 sold tokens and none bought a memecoin — the one buy-shaped operation was an airdrop claim. Of the 2,385 buys recorded that week, 2,044 were deliveries from a spray farm of some thirty rotating signers; those are now deposits. The remaining 341 came from wallets the side-wallet hunt had promoted on the strength of the same sprays.
     • The paper book was reset and starts fresh; side wallets are unconfirmed again and must earn their place from genuine buys, which the hunt re-confirms in place.
125. 0.25.1Only the trader's own buys2026-09-06
     • A buy now counts only when the trader signed the transaction or paid from their own wallet. Tokens delivered by someone else's transaction — a relayer buying through a pool and sending the output to a well-known wallet — are logged as deposits and never open a position, count toward a crowd, or reach the channel. Measured today: 1,374 of the last 1,508 recorded buys were such deliveries; fomo lists them as deposits and flags the tokens.
     • Open positions that had been opened on deliveries were removed from the book.
126. 0.25.0Conviction, redefined2026-09-06
     • A crowd now means tracked traders whose first buy is at least six hours old and who still hold on chain, with at least two of them carrying a winning copy record here. Ten wallets piling in over ten minutes is a launch, and no longer counts.
     • We never buy above the move: a fresh conviction entry is skipped when the price is above 1.5× the crowd's own median entry, and an elevation or add is skipped when the price is above 1.5× our own entry. The pool must be a day old with $25,000 of liquidity, and the size is one ticket — the ladder to 2× is off.
     • All of it is knobs, and conviction is switched back on under them.
127. 0.24.5Conviction is not P&L2026-09-06
     • Conviction positions are listed but never counted: every P&L tile, the realized statistics and charts, the traders' copy P&L and the records that size tickets ignore them. Their rows keep their own P&L for the record.
     • Conviction is paused for now (the holders threshold is set out of reach). The two rows it held were pulled to zero within an hour of being sized up; both had been elevated at several times the price of our own entry minutes earlier, which is the wrong end of a crowd. The rule needs a different definition of a crowd before it comes back.
128. 0.24.4Admin, one page per section2026-09-06
     • The admin console is eight pages under one nav — Support, Settings, Signals bot, Traders, Telegram, Users, Statistics, Jobs — each loading only what it shows. Every form and switch saves exactly as before; the two settings forms share one handler.
     • The wallet manager paginates: 25, 50 or 100 rows a page, remembered per browser, filters reset to page one.
     • The size of a conviction position at the threshold is a knob (1× the ticket by default), next to the per-holder step and the cap. On phones the Signals tabs fit one row and admins see the close button.
129. 0.24.3Conviction, graduated only2026-09-06
     • A conviction entry or elevation now needs a graduated token — past any launch curve — with at least $10,000 of liquidity on both DexScreener's figure and our own, and a route that actually quotes the buy. Six of the first seven conviction positions had opened on pools that drained to a few hundred dollars within the hour; they were removed from the book, fomodoge stays.
130. 0.24.2Curves count2026-09-06
     • Buys on launchpad curves are recognised: the Pons v2 curve's buy and sell events and the Flap portal's, decoded from live transactions. A trader buying before graduation now shows up as a buy and counts toward a token's crowd. pools.trade already traded on Uniswap v4, so it was covered; Bags and hood.fun had no trades to sample yet.
     • The conviction book may hold ten positions, up from five.
131. 0.24.1Buys, not handouts2026-09-06
     • A buy is now recognised only when the transaction carries a swap AND delivers the token to at most two addresses. A relayed buy (the relayer signs, the router swaps, the tokens hop to the trader) still counts; a batch that hands the token to a crowd — an airdrop, a sniper's dispersal, a launchpad push — does not, swap or no swap.
     • Conviction entries get their own liquidity floor, $2,000 by default, instead of the regular $5,000; elevating a position we already hold needs none.
132. 0.24.0Conviction2026-09-06
     • A fourth open book. When six or more tracked traders bought a token in the last seven days and still hold it on chain, we hold it too: sized at one ticket at the threshold and a quarter more per further holder, up to twice the ticket. A regular, dust or moonbag position is elevated the moment its token gets there.
     • Conviction positions hold no slot and are never sold by a rule — no take-profit, no trail, no stop; only an admin closes them. The holders are re-read on chain every half hour and shown on the row; if they leave, the row says so and waits for you. The same half-hourly pass sweeps every token with enough tracked buyers, so a crowd that formed without a fresh buy is caught too.
     • Every buy-in of a position is now recorded — entry, dust upgrade, each conviction add, the bot's real fill — with the price, market cap and ETH rate of the moment. A multi-buy row shows its buy-ins and median entry market cap; its P&L runs off the blended cost basis.
     • The live bot adds top-ups to a row instead of overwriting its first fill, which also lets dust upgrades run on the live book. Conviction entries and adds post to the channel as their own card and line.
133. 0.23.2In ETH2026-09-06
     • Every position row shows its size in ETH before the entry market cap, and its P&L in ETH beside the percentage. Totals, the realized and open tiles, the statistics modal and the traders' copy P&L are stated in ETH, with the USD figure beside them.
     • The ETH/USD rate is recorded at entry and at exit, so the ETH figures are exact from the USD legs, both books. Positions older than this release use the current rate for both legs.
134. 0.23.1Windows, fees, and the admin's red button2026-09-06
     • Signals has a time window — 4h, 24h, 7 days, all time — and the closed list, the activity log, the realized tiles and the statistics modal all follow it. Open positions are always now.
     • Open P&L counts the slot book only; dust copies and moonbags are reported beside it.
     • The gas the bot pays to buy and sell a live position is recorded on the position and subtracted from its P&L. Paper rows already carry pool fees and price impact inside their chain quotes, plus 2% assumed exit slippage.
     • Admins get a close button on every open row — position, dust or moonbag — that closes it at what its tokens fetch on the chain right now.
135. 0.23.0Dust, moonbags, and the shortlist in your hands2026-09-05
     • Dust copies: a watched trader's small buy (under the dust threshold, $50 by default) is no longer ignored — it is copied at half the ticket, tagged dust, and holds no slot. A real-size buy of the same token later upgrades it to a full position at the blended entry, slot and all.
     • Moonbags: every position that exits in profit — trail, second take-profit, trader exit, time exit, displacement — keeps 10% of its original tokens open for good. No slot, no rule ever sells it, only marks. The ladder is now 50% at 2×, everything but the moonbag at 5×.
     • Signals shows three open books — positions, dust, moonbags — above the closed list, and the traders page is for signed-in users like the book. Page copy and Telegram lines describe outcomes, not rules.
     • Admin: a Traders & wallets section lists every followed wallet and fomo account with its type (manual or automatic), source, status and record; wallets can be switched off and on (a switched-off wallet is still followed for its sells while it has open copies), added by hand from a fomo handle, and picked from the leaderboard names not yet tracked. The leaderboard job now adds only traders in profit.
136. 0.22.1Cleaner cards2026-09-05
     • A buy card is never rewritten: exits and take-profits reply under it as their own line. A position that is already closed when its card comes due — the free-channel delay, now 60 seconds by default, is the filter — is not announced at all, and outcomes without a card are not posted. The plan line is gone; a card shows how many shortlist wallets bought the token in the last day when more than one did.
137. 0.22.0What 199 closes taught us2026-09-05
     • Take-profit now sells half at 2× by default and the trail arms right there, firing 40% off the peak. 94 of the first 199 copies reached 1.5×, and the ones that peaked between 1.5× and 3× still closed red — that was the money left on the table.
     • Buys under $500 are ignored (they returned +9% against +82% for $500–2K) and pairs older than a week are skipped unless the buy is $2K or more.
     • At the cap, a strong signal — a proven trader, three shortlist wallets on one token inside an hour, or a $2K+ buy — displaces the biggest loser, else dead money, else banks the biggest winner. Any signal may take a dead-money slot: a position open 6h+ that never saw 1.5× and sits within 20% of entry.
     • Traders with under 20% won after 8 copies and a negative record are benched: one probe copy a week keeps their record alive.
     • Telegram: the entry card states the plan, take-profits post as their own line, and a "hold the queue" switch in admin pauses the dispatcher without losing the queue.
138. 0.21.9Under the bar2026-09-05
     • The Buy buttons of the launch list no longer float over the header while scrolling.
139. 0.21.8Credit where due2026-09-05
     • A copy made on a side wallet's buy is credited to the trader it shadows: the record that sizes tickets (proven / losing) and the traders board count it for the trader, never for the side wallet. Positions remember which trader they belong to.
     • Side wallets are assigned to the trader they lead by the least — a wallet seconds ahead of its own trader and minutes ahead of a follower belongs to the first — and are no longer mistaken for snipers for being ahead of that follower. The hunt for a trader's side wallets pauses for two days once more than ten are confirmed.
     • A position whose pool drained no longer sits at -1% forever: when the chain quote keeps failing for 15 minutes the mark falls back to DexScreener so the stop can close it.
     • The displacement rule now has its own test. The homepage links the Telegram channel.
140. 0.21.7The channel is live2026-09-05
     • t.me/hoodinisignals is open: every entry of the book lands there as a card the moment the agent buys, and its exit edits the card. Linked from the Signals page.
141. 0.21.6Room for the proven, and a channel2026-09-05
     • When the book is full and a proven trader buys, a position makes room: the biggest loser first, and when nothing is red, the biggest winner is banked. Positions under ten minutes old and live positions mid-flight are never displaced. Both books.
     • Telegram signals, phase one: every entry of the current book becomes a card in the public channel — trader or suspected side wallet, market cap, liquidity, the contract address, and buttons to hoodini, the one-click buy and fomo. Exits edit the card in place and reply under it. Every post is stored, and a webhook serves the bot's commands.
142. 0.21.5Statistics2026-09-05
     • “See more statistics” on Signals opens the book's charts: how positions ended, return per position, cumulative realized P&L, hold times, and the best and worst traders to copy. Everything re-computes live with the page.
     • Exit reasons are grouped by kind again (the repricing notes had split them into dozens of buckets).
143. 0.21.4Real money rules2026-09-05
     • A live position is real only once its buy has filled, and closed only once its sell has confirmed. A buy that is skipped or fails voids the row — it never counts. A failed sell leaves the position open, flagged, and retried on a backoff; nothing is ever closed on paper while the tokens are still in the wallet.
     • A failed second leg names exactly what is stranded in the wallet and how much; a partly-converted sell closes the position and says what it left behind.
     • Two guards halt new buys while sells keep running: a gas reserve the buys may never touch, and a daily realized-loss limit on the live book. The halt shows at the top of the admin's bot section with a Resume button; you can also halt by hand.
     • Every ten minutes the live book is audited against the wallet: a position whose tokens left outside the bot closes, one holding less than recorded is corrected.
     • The live book sizes tickets from the paper record until a trader has five live closes. Admins can view either book with ?book=sim or ?book=live.
144. 0.21.3Two books2026-09-05
     • Every position and every book event now carries the mode it was made in — sim (paper) or live (the bot sends) — and Signals, Traders, token and wallet pages show only the current mode's book, statistics included. The page says which book it is.
     • Switching the bot's mode in the admin closes the old book's open positions at their last mark, restarts the worker into the new book, and never sells real holdings by itself — it tells you which tokens are still in the wallet.
     • The live book's ticket is what the bot really spends; one entry per token applies per book.
     • The admin's background jobs show the signals bot: book, open positions, wallets watched, live feed, last buy seen, last entry, last chain mark, the bot's last action and the signer.
     • Ticket size follows the trader's record: with at least five closed copies in the current book, a trader in profit who won half or more gets 50% more per ticket, one in loss who won under 40% gets 50% less — paper and the bot's ETH alike. The percentage is a live knob; rows show their multiplier.
145. 0.21.2Custody2026-09-05
     • The signals bot's private key is sealed in the database — AES-256-GCM per key, bound to the wallet address, unlocked in memory with a key-encryption key that lives only in the server's root-only environment. A database dump reveals nothing without the box; the box reveals nothing without the dump.
     • The admin panel shows the full bot address with a copy button, its balances on both networks, where the key lives, and which wallet the worker is actually signing with. Revealing the key needs a typed phrase, is rate-limited and audited, travels in a response no cache may keep, and disappears from the screen after ten seconds.
     • A new wallet can be generated from the panel (typed confirmation). The previous wallet is retired, never deleted — its key stays revealable so funds can be swept — and the worker restarts so nothing of the old key remains in memory. A restart button covers a stuck worker; settings themselves apply live within five seconds.
146. 0.21.1Prices from the chain2026-09-05
     • Paper entries are priced on the chain: the ticket is quoted through the real route the moment the signal passes the gates. DexScreener's quote at detection lagged the pool in fast moves and produced fake multiples (a 15× stale entry read as a 29× win); legacy entries carry a small “ds” tag.
     • Marks come from the chain too — the position's own tokens quoted through the real route, re-quoted when its pool trades (live Swap events) and every five minutes regardless. A drained pool had printed a $112 mark on a $0.0002 token; that cannot recur, and every take-profit, stop and trail now fires on a chain-confirmed price.
     • Real bot fills overwrite the paper prices: tokens received for ETH spent on the buy, ETH received on the sell.
     • One entry per token, ever: whoever buys first, the first signal that passes the gates opens the position; later buys are conviction while it is open and skipped after it closed. A skipped first signal opens nothing, so a later clean signal on the same token can still be the entry. Two same-tick entries on one token are now impossible.
     • Positions show market cap at entry and now (price × circulating supply, read from the chain at entry), and both tables paginate — 15, 30 or 50 a page.
147. 0.21.0Buy what the signals see2026-09-05
     • One-click buys and sells now route the way the signals bot does: the token's real market on any pool version, any hook, quoted in whatever it is paired with. A token quoted in USDG or a tokenised stock is bought in two or three transactions — ETH to the quote asset, then on to the token — sent one after the other with the amount of each step read from the previous receipt. The token page says so up front, and Signals open rows have a Buy button.
     • Fixed: when a token's quote asset had no ETH pool, the old route resolver settled for a Pons fee-ladder decoy and would have bought through an 80–90% fee. Pools above 3% are no longer routes at all.
     • Signals and Traders fold properly on phones — two-line rows, nothing squeezed off-screen.
     • Crawlers now get the indexed page without triggering chain reads: they were 97% of token-page traffic and about half of all provider spend.
148. 0.20.0The wallets before the wallet2026-09-04
     • Side-wallet detection: for every verified buy by a shortlist trader, the agent records who bought the same token right before them. The same wallet running ahead of the same trader on three different tokens (a live knob) is a suspected side wallet — it joins the shortlist under that trader, is watched live and copied with the same ticket. A launch sniper is not a side wallet and is excluded: the full pattern against three or more traders, running ahead of other traders on tokens its own trader never bought, twenty tokens in an hour, or a recent buy list that is mostly not its trader's — the launch crowd.
     • Suspected side wallets are marked everywhere they appear — nested under their trader on Traders, tagged on Signals and token pages, explained on their wallet page — as an inference from timing, never a confirmed link.
     • The wallet poll runs at half the rate: the live WebSocket path now sees buys in about two seconds, so Blockscout is the safety net, not the eyes.
     • Fixed: the bot's ETH-per-buy field would not save (its minimum did not sit on the input step).
149. 0.19.1The book behind the door2026-09-04
     • Signals is for signed-in users: the statistics stay public, the live book sits behind a blur for visitors and what renders there is placeholder data, not the real positions — nothing to unblur. The live stream is gated the same way.
     • Traders is sorted by copy P&L — the traders our signals have made money on first — and shows FOMO's own 30-day P&L and board rank next to ours.
     • The shortlist re-scrapes FOMO's full leaderboard (about 55 traders, not the public top 5) every six hours and only ever adds: a trader who drops off the board keeps their place and their measured record.
150. 0.19.0The bot gets a wallet2026-09-04
     • A dedicated Signals bot section in /admin: paper ticket, take-profit rules, ETH per real buy, daily spend cap, max open positions, slippage bound, network (testnet/mainnet) and the live switch — plus the bot wallet's address and balances, a testnet self-test transaction, an audited key export, and the bot's own trade log.
     • Shadow mode by default: every paper entry and exit is resolved, quoted, composed and simulated against the chain exactly as a real trade would be, and the outcome — or the reason it couldn't trade — is recorded. Nothing is sent until the live switch is on and the network is mainnet.
     • The open book is now marked every 10 seconds, so a take-profit or stop fires within seconds of the price getting there.
     • The copy agent listens live: a WebSocket carries every token transfer into or out of a shortlist wallet the block it lands (the 2-minute poll stays as the safety net), and every swap on a pool we hold triggers a book pass within a second.
151. 0.18.9Take-profit, and the knobs in admin2026-09-04
     • The paper agent takes profit: when a position's mark reaches the take-profit multiple (default 10×) it sells the slice (default 50%) once, books it as a closed position at the mark minus 2% slippage, and lets the rest ride under the usual exits.
     • Paper ticket size, take-profit multiple and slice are live knobs in /admin with the exact line of code each one drives; the Signals page prints the rules it is running.
152. 0.18.8Sellability, measured where the token trades2026-09-04
     • The sell round-trip now runs in the pool DexScreener names as the token's market, in that pool's own quote token — USDG, GME, NVDA, RBLX, whatever the creator paired with — instead of an ETH-only route that landed in Pons v2's 91%-fee anti-snipe decoys and called good tokens unsellable.
     • Fee-aware verdicts: a fee-ladder pool reads as sellable with its fee shown on the token page; a quote far below the market price means a decoy and reads unknown, never a verdict. The headline no longer says 'honeypot' unless GoPlus flagged one.
153. 0.18.7Curve stage on Alpha2026-09-04
     • Tokens from curve launchpads (Pons v2, Flap, hood.fun, Bags, pools.trade) wear their stage on the Alpha list: graduated into a DEX pool, or still on the bonding curve. Graduation is the event record — the launchpad's migration event or the first pool creation — not an inference.
     • Two new filters next to DEX paid: Graduated (curve tokens that made it) and On the curve (still raising — named, no pool, so no market numbers yet).
154. 0.18.6Sign in where you click2026-09-04
     • Scan again, the watchlist star and the unknown-token scan open the sign-in dialog directly when you're not signed in — no error line first, same as Buy.
     • Notable holders list one row per Arkham entity (several labelled wallets of the same name were reading as the same holder twice), summed, with the wallet count; dust holdings are dropped.
     • GoPlus authenticates with the app key as a header; the per-IP throttle no longer blanks the honeypot and tax readings.
155. 0.18.5Scans that measure2026-09-04
     • Sellability is checked on the token's real route — the v3 quoter, the v4 quoter or the v2 router, resolved from the chain — instead of only the v2 router, which read nine in ten tokens (all the v4 launches) as unknown.
     • Top-10 and opening-bundle reads no longer depend on a metadata pass that usually hadn't happened yet: the scanner reads total supply itself. And a token gets a second and third look 20+ minutes after launch, once Blockscout has indexed its holders, instead of one scan 30 seconds in.
     • GoPlus: a rate-limit answer is now a back-off, not a silent unknown, and the client can authenticate with an app key — the free key lifts the per-IP limit the server was hitting on every call. Liquidity shows the indexer's number for v4 pools until they are read on-chain.
     • Scanning is two-speed now. The background pass is light — holders, top-10, source, GoPlus; no Alchemy calls — and runs at half the old rate. The full read (opening bundle, pool depth, sellability round-trip) runs when someone opens the token page or hits Scan, at most once per 15 minutes per token.
156. 0.18.4Smart money on the Alpha list2026-09-04
     • Alpha rows show when shortlist FOMO traders bought a token — a receipt-verified swap, counted per distinct wallet — and light up lime while the paper agent is holding it. Hover for the detail; the token page has the full activity.
     • Alpha's column headers sit on their columns again (the Signals track had a per-row width).
157. 0.18.3Wallets board retired2026-09-04
     • The Wallets board and its background discovery are gone. It ranked wallets by inferred early entries; Traders tracks real, verified traders with a measured copy record, which is the smart-money surface from here on. Wallet detail pages are unchanged.
     • Trader names link to their FOMO profile now, referral attached, everywhere a handle appears.
158. 0.18.2Buys, not transfers2026-09-04
     • The copy agent copies buys only. A buy is a swap: a same-transaction WETH/USDG leg out of the wallet, or a Uniswap v2/v3/v4 Swap event in the transaction's receipt — FOMO executes trades through Relay's proxy, so the wallet itself never pays on this chain. Airdrops, batch pushes and claims, which had produced most of the 'signals', no longer count as anything.
     • The pair-age gate is gone: measured on the shortlist's real buys, these traders buy month-old pairs, not launches. The liquidity floor drops to $5K; honeypot and sell-tax gates are unchanged. Pair age and how many shortlist wallets bought inside 24h are recorded on each entry instead of deciding it.
     • Every event now carries the token ticker and the price at signal time — skipped signals included, so the gates can later be judged by what those tokens did next.
159. 0.18.1Honest P&L2026-09-04
     • Signals P&L rewritten. Realized losses were rendering as a dash — the USD formatter treated negatives as dust — so a -$253 closed book looked like nothing. Every P&L is signed now and never hidden, on Signals, Traders and wallet pages.
     • Open positions are marked net of the same 2% exit slippage the closed book pays, so open + realized = all-time, shown with return on capital deployed and how fresh the marks are.
     • Realized statistics — wins of closed, median / best / worst return, median hold, exits by reason — are computed over every closed position, not just the ones listed.
160. 0.18.0Copy trading, in the open2026-09-04
     • New Signals board: when one of FOMO's proven 30-day traders buys a young token on Robinhood Chain, the paper agent vets it (pair age, liquidity, GoPlus) and paper-holds $100 until the trader exits or a stop does. Every entry, exit and skip is shown with its reason, live. Paper only — nothing signs.
     • New Traders board: the copy shortlist with each trader's measured copy P&L — what following them actually returned here, not what they claim. Signed-in users star traders into a watchlist.
     • Token pages show which shortlist traders touched a token and what the agent did about it; a shortlist trader's wallet page shows their copy record. Every token has a fomo.family button.
     • Telegram alerts: link your Telegram from your account and get a message when a starred trader's copy enters or exits, plus the book once a day.
161. 0.17.0The ladder & what happens after2026-08-13
     • The Paid board tracks every promo's market cap from the moment the event landed: a minute-sampled ladder sparkline per token with the % since entry — hover for the entry MC itself.
     • Below each list, compounding outcome statistics: how many dex-paid tokens dump within the hour, how many boosted tokens are rug-shaped by 4h, median returns and peak multiples — split by boost size, every number carrying its sample size and its thresholds.
     • Boosted tokens wear an amber ⚡ badge on their token page, next to DEX paid.
     • Radar is public; portfolios load instantly and show USD captured at fill time.
162. 0.16.1Radar for everyone2026-08-13
     • Radar is public now — the momentum-ranked board no longer needs an account. Signing in still unlocks trading, scanning and the Wallets board.
     • Portfolio loads instantly (positions price concurrently in the background) and shows USD alongside ETH — captured at the moment of each fill.
     • The wallet leaderboard ranks by actual ETH held, with exchange hot wallets identified by their transaction shape and excluded.
163. 0.16.0Smart money2026-08-13
     • New Wallets board (signed-in): the wallets that got into opening windows of tokens that later reached scale, ranked by proven money — realized PnL measured exactly from their swaps' WETH legs, then actual ETH held. CEXes, routers and infra excluded; Arkham identities where known.
     • Every top wallet's buys are tracked in the background — the latest ones stream live on the board, each linking to its token.
     • Section nav in the header — Alpha, Radar, Paid and Wallets with their own icons, on desktop and in the mobile menu; search moved to its own line.
     • Buy buttons everywhere now show what the click will spend (your configured size), and the token page takes any custom amount — float-validated and checked against your live balance minus gas before anything is signed. Sells check the gas tank first too.
     • The holder map says what it shows: how many top holders are drawn and when they were scanned — with an Update button right there for signed-in users.
164. 0.15.0Live pages & real charts2026-08-13
     • Token pages are live now: open one and its market cap, price, liquidity and volume track the market every ~10 seconds — no more numbers frozen at the last background pass.
     • Real trading charts on every pooled token, embedded straight from DexScreener.
     • Holder history: every tracked token's holder count and top-50 wallets are snapshotted continuously — accumulation and distribution become analyzable, not just guessable.
     • Dead tokens (no liquidity, under 10 holders) stop costing background work — but open one's page and it revives on the spot if the numbers came back.
     • Market data now falls back to DexScreener for everything GeckoTerminal can't price, and the DEX-paid badge lands within seconds of opening a token we hadn't checked yet.
     • The scan button moved up top, got an icon, and its spinner now stops when the scan actually lands.
165. 0.14.0The Paid board & a softer face2026-08-12
     • New Paid tab: two live columns of tokens that just paid for their DexScreener profile or just bought boosts — polled every 15 seconds, newest on top, with market caps, copy-CA, and one-click buy. Paying is a signal of intent to market, not of quality.
     • Tokens discovered through promos or the market-wide radar now enter the full pipeline automatically — scanned, priced, imaged and tradeable like any launch we watched from birth.
     • In-app trading covers Uniswap v2 pools too — the last pool kind we didn't route. Every graduated token now buys and sells in-app.
     • The holder bubble map went interactive: it settles live, pans, zooms, names wallets on hover and opens them on click — with the opening bundle clustered and linked, Bubblemaps-style.
     • Token images now also come from DexScreener when chain metadata has none — the whole backlog gets swept.
     • Dexscreener links on token pages, copy-CA buttons in the feed, and a fresh look: Inter & Manrope, rounder corners, soft hover shadows.
     • Fixed: 'Buy on Axiom' links now key by what Axiom actually resolves — the pair address or v4 poolId, never the token CA.
166. 0.13.0One-click trading, everywhere it trades2026-08-11
     • One-click trading now works on Uniswap v4 pools too — the pools.trade launches, not just v3. Buy and sell with native ETH straight from the token page.
     • Fixed the trade box going missing on tokens whose real liquidity is a different pool than the one we first saw — we now resolve the deepest pool live from the chain, so deep v3 names (and v4 launches) all trade in-app.
     • Prices now read like prices: $0.0₇591 instead of 5.91e-8.
     • Tidied the header; index counts moved to the footer.
167. 0.12.0pools.trade2026-08-10
     • New launchpad integrated: pools.trade — Uniswap's official token launchpad for Robinhood Chain. New launches are now detected and labelled, and flow through every signal like any other pad.
168. 0.11.0Radar2026-08-10
     • New Radar board (signed-in): the hottest tokens chain-wide — established names and fresh launches alike — ranked live by a momentum score (volume acceleration, buy pressure, holders, smart money, MC fit, liquidity), gated by safety.
     • The public feed is now “Alpha” — switch between Alpha and Radar from the tabs.
     • Every token page now shows its live momentum score.
     • Operators can flip real-ETH trading off chain-wide (everyone drops to demo) from the admin panel.
169. 0.9.0Intelligence & support2026-08-10
     • Smart-money & notable-holder callouts on token pages — when a fund or known entity holds a launch, you see it (powered by Arkham).
     • Deployer reputation: entity labels and wallet freshness surface right in the feed and on the token page.
     • Funding-source sybil detection — deployers funded from the same wallet get tied into the ring graph.
     • In-app support: open a ticket and we answer in the thread.
170. 0.8.0Trade from hoodini2026-08-10
     • One-click buy & sell straight from your built-in non-custodial wallet — size, slippage and moonbag from your settings.
     • Demo (paper) trading with live P&L, a portfolio page and a public leaderboard.
     • A confirmation step on real trades, until you tick “don’t ask again”.
     • An operator admin panel: live settings, users, stats and background-job health.
171. 0.7.0See the crews2026-08-09
     • Cross-launch bundler/sybil ring graph — spot wallets that keep launching together.
     • DEX-paid badges, minute-candle price charts, token socials and images.
     • Sharper sellability and holder-concentration reads (pool & infra excluded).
172. 0.6.0The console2026-08-08
     • A live feed of every Robinhood Chain launch, read the moment it graduates.
     • Multi-launchpad filters, time windows and sorting.
     • The trust ribbon — sellable, honeypot, mint, owner, liquidity — with unknown shown, never hidden.
     • Scan any token on demand.

Follow @hoodiniapp for release notes, or read the handbook. Something off? Open a ticket.