Changelog

On 16 October our main chain data provider stopped serving us for about fourteen hours, and with it new launches, scans and holder credits. The site now falls back to Robinhood Chain’s public node for ordinary reads when that happens: new launches, first scans, the exits feed and the looks at the price after a scan carry on, and the launches missed in those hours were read on the way back. Two things still need the main provider and wait for it: a full scan (it finishes as partial without it) and holder credits, whose price is averaged over an hour of chain history that the public node does not keep.

The exits feed now reaches Telegram. In any chat with @OcellusBot, /exits posts each creator selling and each liquidity pull as Ocellus sees it, with how much moved and the level the coin read; /exits all adds the largest wallets selling, which is far more frequent; /exits off stops it. At most 40 posts a chat a day. “Exits” is also in the top menu now.

Every scanned Robinhood Chain token is checked again every few minutes between scans. That check has always noticed when a creator sells, when the largest wallets of the last scan sell, and when the liquidity leaves the launchpad’s locker, but it only told the token’s followers. Now every one of these is kept and shown at ocellus.app/exits, newest first, with how much was sold and the level the token read. The top shows the last 24 hours in four numbers. In our own record, a creator who had sold out was the finding most often followed by a fall of 80% or more within a week. The feed starts today.

About 1,700 launches a day raise enough to be worth a scan, and the day’s first scans cover 400. They used to be picked in an order that amounted to chance, so most went to launches days old. Now coins someone asked for come first, then launches of the last two days, furthest along their curve first.

A report answers “what is this coin”. This answers the question a holder arrives with: what about the ones I already hold. At ocellus.app/bags, paste any Robinhood Chain wallet: Ocellus lists every scanned coin in it with its level, its heaviest finding and what the price has done since the scan, and says how much of the wallet sits in coins reading High or Severe. Coins it holds that nobody has scanned are named, with a button to scan them. No sign-in and nothing to connect: holdings are public on the chain. The page has a link and a picture to share.

Most wallets hold coins nobody has scanned yet. On a bags check, “Scan all” asks for a free Quick Scan of each of them, up to a dozen at a time. They are scanned one at a time alongside new launches, so it can take a while; check the wallet again later and they appear with their levels. “My bags” is in the top menu.

From 10 to 12 October no token was scanned again: one whose scan could not be started stayed first in line and held every other back, so the screener aged and Solana tokens were not refreshed at all. A token that cannot be started is now set aside for a few hours so the next gets its turn, and the site checks for this itself.

With two wallet extensions installed, sign-in opened whichever one had claimed the browser first, usually MetaMask, even if you use Phantom. The Account page now lists the wallets it finds by name and you pick one. It remembers your choice for next time, and buying credits uses the same wallet. With one wallet installed nothing changes.

Signing in needs a wallet the page can talk to, and a phone’s ordinary browser has none. The Account page used to tell you to find ocellus.app again inside your wallet app. On a phone it now shows two buttons that take you there: the same page opens in Phantom’s or MetaMask’s own browser, and you sign in as usual. Another wallet with a built-in browser and Robinhood Chain still works the old way.

A call used to need the coin in the wallet you signed in with. A wallet you have verified on your Account page now counts too: its coins appear under “Coins you can call”, and a call is made from whichever of your wallets holds $10 or more of the coin. It is still one call of a coin for you, whichever wallet makes it, and it still ends when that wallet sells.

A token’s page lists every scan of it. Each scan now says what the price did afterwards, at the furthest moment it was looked at: an hour, six hours, a day or a week later, or that no market priced it any more. It is there so you can judge a scan against what followed, not only take its level on trust. A scan whose later looks were missed, as many were before 6 October, says nothing.

We checked the level against what followed, on about 1,400 tokens with money in them at their first scan. A week later, 46% of those with no finding had fallen 80% or more; with a creator who had sold out it was 62%, while wallets holding together (linked wallets, buy groups, split wallets, bots, a creator still holding) were at 35% to 40%, no worse than nothing found. Those holdings carried the most points, so High had been reading better than Low. Now a creator who sold out counts for three times as much, and a holding on its own moves a token one level, not two. Every finding is still shown. Many tokens read a different level than yesterday: every saved level has been worked out again under the new weights. A level is still not a forecast: about seven in ten of those tokens had halved within the week, at every level.

The chart on the homepage and the statistics page counted every scan, including coins nobody had bought. A coin with nothing in it never changes price, so it could never read as having fallen, and more than half of the tokens were like that: they made Low and Moderate look safe. The record now counts each token once, only if it had at least $1,000 in it when scanned, by the level it reads under the current rules. Because the rules changed the same day, the record starts again from 6 October: a level shows once 30 tokens have a result under the new rules.

After every scan the price is looked at again an hour, six hours, a day and a week later. More than half of those looks were never made: ones that could not be made held up the queue until the rest expired. They are now worked through in full, and coins on a curve paid in another token than ETH are priced too. Looks missed before today stay missed.

On your Account page and above the callers’ board: the scanned Robinhood Chain coins your wallet holds $10 or more of and has not called, each with a Call it button. Until now you had to open a coin’s report to find out a call was open to you. It is read for your own signed-in wallet only, and shows nothing when there is nothing to call.

A wallet you cannot sign in with can now be proved yours: on your Account page, name the wallet and send the exact small amount of OCELLUS shown (a few cents) from it to the burn address. Holder credits are then read from that wallet, under the same rules as any other. It works for accounts that sign in with an email too. Only a wallet can be proved, never a contract, and Ocellus never asks for a key or a signature for it. The wallet has to be able to send OCELLUS: an app that only lets you send USDC, as FOMO does today, cannot complete this yet.

Every holder check asked a public price service for the price of ETH, on every page load. It answered "too many requests", and holders were told their credits were unavailable and could not claim or spend them. The price is now asked for once a minute and reused.

A launch that holders have called says how many calls it has, on its card on the launches page. A holder’s call, not advice.

Launches were only kept from 1 October. Earlier ones are now being read back from the launchpad’s own events, about four hours of history a minute, each credited to the wallet that made it and marked with what became of it. A creator’s count grows as that reaches further back.

The launchpad records as "deployer" whoever called it, and for a launch made through a launcher tool that is the tool’s contract. Ocellus counted one such contract as a single creator: 769 launches in three days, 746 dead, shown in red on every coin launched through it and counted against their level. Traced on chain, those launches came from hundreds of different wallets. A launch made through a contract is now credited to the wallet that sent it, on launch cards, creator pages and in reports; a wallet with a delegation, or a smart account behind a bundler, is still its own creator. 437 saved scans that named the contract are having their level worked out again.

Launches were dropped after three days, so a record only ever covered the last three. They are now kept, from 1 October 2026 on. Counts are of every launch on record, and "dead" means the same on a launch card as on the creator’s page (it did not: the same creator could read 161 dead on one and 402 on the other). A launch paid in a token other than ETH is judged by the share of its own target it raised, not as having raised nothing.

The middle column of the launches page was "Almost bonded": launches 60% or more of the way to graduating. On a quiet day nothing is past 60% for hours and it sat empty, while the "closest to graduating" banner showed a coin at 33% and one at 45% went unseen. The column now lists every launch past 60%, marked "almost bonded", and fills up to ten with the next closest above 10%, each with how far along it is. The banner is chosen from the same list.

On 3 October the chain provider refused every call for eight hours and the page went on saying "Live". It now shows "Not updating" with the time of the last read once three minutes pass without one. Automatic scans are spread through the day instead of being spent in a burst, and the owner is told by email when the provider stops answering.

From the scan bar, the search box, a launch card, a wallet’s page or your library: if the token has a public scan (every called token does), you land on its page, with when it was scanned and what changed since. A token never scanned is scanned as before, and the report’s own refresh still takes a new scan.

Three read-only endpoints: a token’s newest scan (level, the findings behind it, holders, creator, Dex paid), a creator’s launches of the last three days, and the live launches columns. Free: 60 requests a minute with a key from your Account page, 10 without one. It serves what is already scanned and never starts a scan. Documented at /developers.

A small picture a project can put on its own site, showing the level at its newest Ocellus scan and linking to the report. It is read each time it is shown, so it follows rescans. It states the level and nothing more; it is not an endorsement.

Read from DexScreener every 20 seconds: its feed of the latest paid profiles, and each coin’s own orders, which give the time it was paid. Only an approved paid profile counts; a community takeover does not. Hover the label for when it was approved. It means someone paid for a listing; it is not a safety check.

Each signed-in account now has 100 new Quick Scans a day instead of 30, starting again at 00:00 UTC. Opening a report someone scanned recently has never counted, and now a scan that fails is given back too. When a limit is reached, the message says which one and when it lifts.

Launch pictures now load: they are fetched as each launch is found, from the IPFS gateway that answers, and one that could not be fetched is tried again within minutes. The page gained a ticker of what just happened (launches, curves crossing 25/50/75/90%, graduations), the launch closest to graduating as king of the hill, a badge on the three curves raising fastest right now, and today’s launch and graduation counts. All measured from the chain; none of it is a tip.

A new page, New launches, shows Robinhood Chain launches as they happen in three live columns: new (the last hour), almost bonded (60% or more of the way to graduating) and bonded (graduated in the last day). Each card shows how much the curve has raised in the token it is paid in, the market cap, the creator’s record of earlier launches, its links, and its Ocellus level or a one-click scan. It refreshes every few seconds; pause it, or point at a column, to stop it moving while you read.

A coin whose scan had aged off the screener list read as unpriced, and a day later the position was written off at zero, whatever its price. Held coins now always stay on the list the accounts read, and the 38 positions it hit were re-priced from the stored price history; each says so on /agent. One had no price on record and stays at zero.

The careful accounts now hold 6 hours instead of 24, sell half at +30% instead of at a double, and cut a position that falls 30% below its entry. Trades opened before ran under v1 and stay in the record as they were. The patient account keeps its week-long hold, so the two settle which is right.

A launch curve paid in another token (DJT and others) was read as paid in ETH when the coin had not been scanned, which produced a false $27k whale buy of TI. The monitor now reads what each curve is paid in, and 279 stored trade values that had been read wrongly were cleared.

The screener lists the newest scanned tokens, and automatic scans of Robinhood Chain launches had pushed every BNB Chain token off it within a day and a half. Each chain now keeps up to 500 places on the list. The market feed is also asked less often as the list grows, so prices no longer go missing when it answers "too many requests". On the callers’ board, coin cards are tidier and a coin’s peak is never shown below where its price is now.

The callers’ board now opens on Coins: one card per called coin with everyone who called it, the change since the first call, the peak, the money held behind the calls and the latest thesis. Callers is a compact leaderboard, one row per caller with their coins, so many more fit on the page. On a report, the Call it button stands out, and a coin nobody has called yet asks for the first call.

The callers’ board carries the offer and its rules, and each call shows the highest multiple its coin has reached since the call. A 10x counts when the price is at least 10 times the call price on two checks 10 minutes apart, in a real market (a pool with $2,000 of liquidity, or a launch curve that raised 1 ETH), checked every 5 minutes. The caller must hold $50 of OCELLUS at the call and at the 10x and still hold the coin; a coin’s creator, wallets it funded, and Ocellus’s own wallets are not eligible. Each caller’s panel says whether they are eligible now.

A caller can add a thesis to their call: why they like the coin, in up to 280 characters, with no links or addresses. They can update it once an hour while they hold, and every version stays visible, so a view cannot be quietly rewritten after the price moves. Theses show with the call on the coin’s report, on the callers’ board (the latest on the card, every version one click away) and on the caller’s wallet page. A holder’s view, not advice.

The Wallets page leads with Moves now: buys and sells by the tracked smart wallets, whales and patient holders, or by the wallets you follow, grouped by coin, for the last hour, 6 hours, day or week. Your followed wallets show their last move, and the ranked wallets are cards you can follow from. A wallet’s page lists its latest moves. A move by someone else is not a reason to buy.

Every Robinhood Chain report has a Call it button with the number of calls. Signed in with a wallet that holds at least $10 of the token (read on the chain, at the market price), you can call it, once. The token’s calls list each caller, when they called, the price then and the change since, and whether they still hold: a call is marked sold once the wallet has sold the whole bag, checked every half hour. A holder’s call, not advice: Ocellus itself calls nothing.

Every account now has a name, given at random (like Quiet Heron 27) and changeable on the Account page. It is what calls, the callers’ board and the live feed show, with the wallet in small print beside it. The top menu now leads to what matters: the screener, investigating a token, callers, the track record, the paper accounts and pricing, with Sign in (or Account) as its one button.

At /callers: holders ranked by how their calls did since, from the moment they call: how many went up, then the median change, and their best call. A call leaves the board when the caller sells the whole bag. The screener shows how many calls a token has, and sorts by Most called.

The callers’ board shows each caller with the coins they called as cards: when they called, what they held then, the price then and now, and the change since. Latest calls shows every call on the board as one stream, newest first, with who called it. A wallet’s page lists its calls too, straight away, without waiting for a scan of the token.

Wallets on record could read 0 for a moment when the database was busy, and On Ocellus now could read 0 before your own visit was counted. A counter that cannot be read now keeps its last value, and On Ocellus now always counts the person looking.

Asked for in the feature requests. The screener’s filters gain Stage: tokens still on their launch curve, or bonded (trading in a pool). A new view, Fewest warning signs, shows tokens that read Low, with no wallets that look like one buyer split across several, a creator without a run of earlier launches, and a price not collapsed from its peak. Fewer warning signs is not a recommendation, so there is no score and no recommended list.

pump.fun now shares a token’s creator fees with its largest holders, paid out by the program from the creator’s fee account. Ocellus read each payout as the creator sending the holder money, so every large holder looked linked to the creator: one token showed a 24% wallet group that was only fee payouts. A transfer made by a program from an account that did not sign the transaction is now treated as a payout, not funding, and is left out of wallet links. The same token now shows one group of 7 wallets holding 10%, linked by tokens they sent each other directly.

A new section on the homepage sets the three scans side by side (how many wallets each examines, how much history it reads, how far back it traces funding, and what it adds) and shows the buy-back and burn every paid order feeds: 5% of each order is set aside to buy OCELLUS on the market and burn it from the project wallet. It also lists the free credits holding OCELLUS earns. The figures are the same live settings the pricing page uses.

Tokens on BNB Chain can now be scanned: holders and where their money came from, linked wallet groups, and trades on PancakeSwap v2 and v3. Paste any 0x address and Ocellus finds which chain the token is on. A token with a long history gets a holder map sampled from recently active wallets, and the report says so. Not yet on BNB Chain: the simulated sale, four.meme launches, the launch and selling watches, smart money alerts and wallet age.

The alert now slides in with a sweep of light around its edge and a pulse from its icon, shows the token’s ticker and the dollar amount large, and counts down the ten seconds it stays with a bar along its edge, which pauses while you point at it. The ticker opens the token. Alerts show a ticker, never a contract address, and nothing moves for visitors who ask their system for reduced motion.

A small alert now appears on every page when at least two smart wallets, or whales spending $1,000 or more, buy the same Robinhood Chain token within ten minutes: who bought, how much in dollars, and over how many minutes. The wallets are watched every two minutes and each buy is dated by its own block, so an alert only ever shows buys under ten minutes old. Click it to open the token; “Hide these” turns them off in your browser. It says what the wallets did, not what anyone should do.

Each scan now looks up when the 50 largest holders’ wallets were first active on the chain (pools, curves, burn addresses and known services left out) and shows the median and average age, a bar splitting the supply by wallet age (under a day, a week, a month, older), and the chain’s own age, since every wallet on a young chain is young. The holder tables gain a Wallet age column beside Held for, and the holder list beside the map shows each wallet’s age after its share. On Solana a busy wallet with a long history is marked as at least as old as the oldest activity read. It is shown, not scored: it does not change a token’s risk level.

A paid report now opens with the whole scan on one screen: a headline, then nine areas (holders, wallet groups, trading, liquidity and contract, creator, website, X account, claims, market), each with a colour, one line written by the AI, its key figures and its details opened in place. Below them: when the X account, the domain, the launch and the listing appeared, what the scan read beyond a Quick Scan, and the questions left open. Deep investigations now get the AI as well. Every figure comes from the records, never from the model: a line with a number the scan did not produce is replaced by the rule-built one.

The check that makes every AI finding cite real records was throwing out sound findings, so a Full AI investigation could read much like a Quick Scan. It refused a finding for naming the token’s own address, refused an analyst’s finding when the analyst added a caveat about another finding (only the Skeptic challenges findings; the caveat is now kept as another possible explanation), and refused records the AI had been shown in the baseline findings but that sat outside the capped evidence list. Each finding still has to cite records that exist and name only addresses in the investigation.

The scan page now shows Quick Scan, Deep and Full AI investigations as three choices under the address box, each with what it reads and what it costs, and the button says what it will start. A deeper choice looks the token up and brings its confirmation into view. Buttons on the scans page open it with that depth already chosen.

Every Solana scan now asks Jupiter to quote a 0.1 SOL buy and the sale of exactly what that buys, and shows what the round trip loses and which venues it went through, or that there is no route back out. It is shown as a quote, never as a simulated sale: a quote says a route exists and what it costs, not that the token’s own program will let the transfer through, so it does not count toward anything that needs a sale to have been tried.

Holding OCELLUS now earns free credits by the size of the holding, every 30 days: $25 claims 60 credits, $100 claims 200 and $500 claims 1,000. The claim reads what the wallet holds at that moment and grants the highest level it reaches. The 10% extra on purchases is unchanged.

A Quick Scan now shows the project’s X handle from its listing, free, and no longer pays X to read the profile. The X reads go where they tell more: Deep and Full investigations read the account and its recent posts, for every token, however thin its pool. On a Quick Scan the X row says so instead of reading as a check that failed.

A new page compares Quick Scan, Deep and Full AI investigations: how many holders each examines, how much history it reads and how far back it traces funding, drawn to scale, with what each includes and what it costs in credits and USDG. It also shows the credit pack, what holding OCELLUS gives, and the 5% buy-back. Linked from the menu, the scan page, the account page and the homepage.

100 credits for 20 USDG, paid on Robinhood Chain from the wallet you are signed in with. A Deep investigation takes 10 credits and a Full AI investigation 25; Quick Scans stay free. The payment page asks your wallet to send the exact amount to the project wallet, and Ocellus reads the payment from the chain and adds the credits within a minute. Nothing on the site holds a key or moves your tokens: your wallet shows the transfer and you approve it. A payment from another wallet or of another amount is not lost; it is kept and settled by hand. Holding $25 of OCELLUS adds 10% more credits to a purchase.

5% of every paid order is now set aside, as it is paid, for buying OCELLUS back on the market and burning it. The buying and burning are done by the team from the project wallet, not by the site, and each burn is recorded with its transaction so it can be checked on the chain. Credits earned by holding, and bonus credits, set nothing aside: only money that was paid does.

Under a token’s name, one line now says its level, the finding that weighs most in its own words, how many more there are and how many checks could not run, with a button down to the findings. Nothing in it is new: it puts the heaviest of what the report already says first, for anyone who reads nothing else.

Earlier today these were left unpriced rather than priced wrongly. Now the check between scans reads which token the curve is paid in and how many decimals it counts in, and the screener prices it with that token’s own dollar price from the market feed: a curve paid in a two-cent token is worth two cents a unit, not a dollar and not an ETH. A token the feed cannot price still leaves its curve unpriced.

The early-buyers finding was meant to say how long after the first buy the first sale came, and never did: a scan looks up the time only of the blocks it needs, and the first buy and first sell of a launch are rarely made by the wallets it investigates. It now looks up those two as well, so new scans say it.

The homepage now shows what Ocellus is doing as it happens: launches as the watch sees them, launches queued for a scan, scans as they finish, and tokens whose ten largest wallets have sold at least two points since their scan, each with its age and a link to the token. The eye in the middle is the same instrument as before, and it lights up once for each real event. Beside it, launches and scans today with their last 24 hours, wallets on record and how many people are on Ocellus. Below: how many tokens on the list show split holdings, wallets buying together, liquidity a wallet can pull, and holders selling, each with its newest example; how tokens at each level did a day after their scan; and the six paper accounts. Every number is a count of real rows, and when the chain goes quiet the page says so instead of filling itself with old events. The main button opens the screener; one token can still be looked up from the address bar under it. Shared links to ocellus.app now show the same instrument, at rest as in the logo.

A few launch curves on Robinhood Chain are paid in another token rather than in ETH. Ocellus priced them in ETH anyway, so one curve holding a few billionths of a cent read as an $829M market cap and sorted to the top of lists. The scan had its own version of the mistake, counting the other token as if it were worth a dollar: that one is paid in a token worth two cents, which overstated its market cap 43 times. Those curves are now left unpriced everywhere, including the prices recorded to measure what happened after a scan, until Ocellus can price the token they are paid in.

Until now a pasted picture only found tokens whose logo was a copy of it. Communities post their mascot in a dozen poses and the logo is only one of them, so a full-body picture of a token’s character found nothing, although its logo is that same character. Every stored logo and banner now also has a description of what it shows, from an image model running on Ocellus’s own server, and the search compares those as well: that picture now finds its token first of nearly five hundred, clearly ahead of the next. Only an exact copy is ever called the same picture; a match on what the picture shows is labelled as looking alike, for you to compare.

A picture search now returns the five nearest pictures, or every exact match when there are more, with exact matches first and the largest market cap first after that, each with its price and market cap looked up on the spot rather than only when the token happened to be on the screener page you had open. Every row says how close its picture is: the same picture, one that looks similar, or a different one listed only for comparison, so a large token with an unrelated logo never reads as the match.

A thin strip above the header shows Bitcoin, Ether, Solana, BNB, XRP and Dogecoin with their change over the last day, refreshed each minute from CoinGecko, or from OKX or Binance.US when CoinGecko does not answer. A small token that looks like it is moving on its own news is often just the whole market moving, and this keeps that in view. On a phone the row slides slowly past.

Screener, Scan, Agents and Account sit along the bottom of the screen, and More opens every other page as a sheet that rises from the bar. The header shrinks to one row with the logo and a search button, and stays in reach as you scroll, frosted over the page under it.

The homepage, the screener and token pages were each a little wider than a phone screen, so the browser zoomed the whole page out and it slid under the thumb. Each one now fits the width exactly. Tapping a search box no longer zooms the page in either, and the buttons and links that were too small to hit reliably are now at least 44 pixels tall.

Page introductions show three lines with Read more under them. The screener’s views scroll sideways as one row of chips, its search sits beside an orange picture button, and Filters and Sort share a row. On a token’s page the picture sits beside its name, the three actions are one row of equal buttons, the sections stay pinned under the header as you scroll, and the chart’s price line wraps instead of being cut off.

Ocellus has its own icon and opens full screen from a phone’s home screen, starting on the screener.

Paste a logo into the screener’s search, drop one on it, or pick one on a phone, and Ocellus lists the tokens that use that picture. It matches by a fingerprint of the picture’s shape rather than its file, so a logo that has been resized, recompressed or screenshotted as a circular avatar on X still finds its token: on 60 real logos put through exactly that, the right token came first 56 to 57 times. When several tokens wear one picture it says so plainly, because a new launch borrowing a popular token’s logo is one of the oldest ways to borrow its trust, and the picture says nothing about which is real. There were two such pairs among the first sixty logos checked. A picture too plain to fingerprint is refused rather than matched on noise, and nothing that is pasted is kept.

Of what the first 10 and first 50 buyers into a launch bought, how much they have since sold, and the same for everyone who bought in its first half hour. Whether the first wallets into a token are still in it says more about who they were than any label can. It is counted at scan time from every trade on the curve and in the pool since the launch, and not from the list of trades a report keeps, because that list holds today’s largest holders and the latest few hundred trades: the early buyers who sold out long ago are exactly the ones it leaves out. Worked out from it, the first tried on this very token said 62%; counted from all 1,866 trades it is 24%. It only appears where the whole history was decoded, and says nothing rather than something wrong everywhere else. Launch buyers selling is how most launches go, and the finding says so.

X bills per read, and the first day of it cost more than it was worth. The reason was not the checks: 73% of the tokens in the screener hold under $1,000 of liquidity, and every one of them was having its X account bought and filed on a page nobody opens. Ocellus’s own careful account will not touch anything under $10,000. The read now happens only above a liquidity floor, and a token below it says so plainly, in money, rather than reading as a check that failed. A token the listing does not price at all is still read, because an unpriced token is usually a new one rather than a dead one, and those are the launches this check is most useful on. Measured against the real distribution, that removes about four fifths of the spend.

Every scan of a token whose launch profile links an X account now reads that account: when it was created, how much it has ever been used, what it says about itself. The reads are billed, and a traded token is rescanned every six hours, so the answers are remembered for as long as they stay true — the profile for a week, the account’s own posts for three days, the search for who is talking for one. Without that, close to half the money would have gone on re-asking when an account was created, about an account whose creation date cannot change. A handle that does not exist is remembered too, because re-asking every six hours is how a typo in somebody’s launch profile becomes a standing bill.

The careful and the reckless accounts differ on eighteen things at once, so between them they can only settle whether obeying the checks beats ignoring them. They cannot say which check did the work, and that is the question worth answering: every rule refuses tokens, and a rule that refuses without earning it is a cost paid for nothing. Four more accounts now trade the same screener, each one the careful account with a single number moved and every one of them looser rather than tighter. Crowded will buy a token whose ten largest wallets hold up to 70%. Unproven does not wait for a simulated sale. Hasty buys the moment a token qualifies instead of letting it age half an hour. Patient buys exactly what the careful account buys and holds for a week instead of a day. What a loosened account buys and the careful one does not is precisely what that rule was refusing, so what happens to those tokens is that rule’s record. Still paper, still $100 each, still no coin held and no order placed.

The holder map, the trades and the holder list sat under every report whether anyone had asked anything of them or not: 566 pixels, near a quarter of the page, showing the same map as the Holders and money tab with less around it. They now appear when a finding is opened and "Show me who" is chosen, which is the moment they mean something, and the button produces them rather than quietly filtering something already on screen. A reader who wants to browse every holder without a finding in mind has the tab, which was always the better place for it. The overview is a fifth shorter and ends on the findings.

The report said everything it needed to and nobody read it: seventeen findings as sentences in a 400px column beside the chart, every one looking exactly like the last whether it said "the ten largest wallets hold 22%" or "this contract cannot mint". They are now cards across the full width, each with its own mark so the page can be swept by shape, and each carrying what a reader can do next: show me the wallets, show the evidence, open the account or the site. Colour is not decoration here. Four states, four hues, and a check that could not run is coloured apart from one that passed, because Ocellus refusing to answer is a different thing from Ocellus finding nothing. Where a finding is a share of the supply it is drawn as a measure with the remainder showing, so a number reads as a quantity; where it is a yes or no there is no measure, and that gap says which kind of finding you are looking at.

The chart used to begin 528 pixels down the page, and on a phone 840. Most of that was not content: the token’s own banner image took 137 pixels above everything, the description was allowed three lines, and the blocks sat 12 to 20 pixels apart with another 18 of padding inside each. The header is now 146 pixels instead of 204, the chart uses a shorter plot that was already built and unused, and on a phone the banner is dropped, the project links sit in one row that scrolls rather than two that stack, and the five market figures are in three columns rather than two. The chart and the first findings are both in the opening view.

The project’s X account was three clicks away under Project. It is now a finding in the overview beside the others: when the account was made against when the token launched, and how much it has ever been used. A four-year-old handle with eleven posts and a real person’s account are told apart by the one profile read every scan already makes, so it costs nothing extra. The rate is always described as an average across the account’s whole life, because that read cannot see an account that posted daily for three years and stopped in March, and saying otherwise would be the sort of confident wrongness this product exists to avoid.

The admin page now shows how visitors split between phone, tablet and desktop. The user agent is not stored: it was already hashed into the daily visitor id and discarded, and a coarse class answers the only question asked of it without making a visit easier to trace to a person. The class is kept from a visitor’s first arrival of the day, so somebody who opens the site on a phone and later a laptop stays one visitor. Visits recorded before this existed are shown as their own share rather than counted as desktop, which would have flattered the desktop figure on the first day and stayed wrong afterwards.

A reader who opens a token after watching it drop is asking about one candle, and the answer was further down the page where they had no reason to look. The stretch the selling actually happened in is now a marker on the chart, saying how much of the supply changed hands there and in how many sales, and clicking it goes straight to who was behind it. It is only drawn where one stretch stands clear of the rest: selling spread evenly across six hours had no such moment, and a marker is a claim that something happened then.

Naming who sold costs a chain call per transaction, so it stops after a fixed number of them. The share of supply was being added up from only those named sales, which meant that on exactly the busy tokens the question gets asked about, the figure left out most of the window: on one read it said 16% where the real figure was 58%. How much moved and when is in the transfer logs in full and needs no such call, so it is now counted from all of it. Only the list of accounts is capped, and the words say so plainly instead of casting the same doubt over the share. The same mistake had put the busiest moment at the start of nearly every busy token, because the calls that get skipped are the later ones.

A button at the top of every report offered "The glance", a cut-down one-screen version of the page already open. Somebody who has opened a token wants the findings; being asked to choose between two depths of the same report is work, not a feature. The report is now one layout.

A token had two pages: its own, at its address, holding the history and what changed between scans, and a separate page per scan, which is where the chart, the findings and the evidence lived and where every link went. Nobody should have to learn that a token has two addresses, and everything added to the standing page was simply never found. The scan now opens on the token’s own page, with what the report cannot know underneath it: who has been selling since, what the creator has done, what the two paper accounts make of it, what changed since the scan before, and every scan on record. A scan keeps its own link so a past one can still be pointed at, and sharing hands out the token’s address, which always shows the newest read rather than freezing at the scan that happened to be open.

A sell-off tells a holder nothing about who is behind it, and the answer changes what it means. Ocellus now reads the recent sales off the chain on request and says who did them: how much of the supply changed hands, across how many accounts, how many bought that same day and kept nothing, how many are contracts rather than wallets, how many have sent tens of thousands of transactions, and how many look like ordinary holders. Each sale is credited to whoever sent the transaction, not to whoever the tokens moved from, so a router or a relayer is not mistaken for a seller. Addresses sharing a long leading prefix are named as likely one operator rather than counted as separate people. It says when the window was busier than it could read in full, because a part of an answer must never read as the whole of one.

Until now "can holders sell?" was answered by watching whether anyone had. That says nothing about a token nobody has sold yet, which is exactly the token you most want to know about. Ocellus now asks the chain instead: it buys the token and sells it straight back in a simulation on the current block, from an address that does not exist, funded by an override. Nothing is signed, nothing is spent, and no wallet is involved. What comes back is whether the sale went through at all, and what a round trip really costs once fees and price impact are paid. A check that could not run says so and is never read as a pass.

The check was built for tokens still on a launchpad curve, and then the records showed that not one token clean enough to interest a careful buyer is still on one: they have all graduated into a Uniswap v4 pool. So the simulated sale now goes through the pool as well, and where a pool can only be reached through one of the chain’s own swap routers, Ocellus reads the way in off the last real trade and copies it. That covers pools priced in a tokenized equity, which is a third of the graduated tokens on Robinhood Chain.

A sortable column for what a round trip cost when Ocellus tried the sale itself, a filter for tokens where the sale went through, and a preset that combines the two under 8%. A blank in that column means nobody has tried, which is shown differently from a sale that failed: an unmeasured token and an unsellable one are not the same thing and must never look alike.

The page opened with a list of what had changed since the previous scan, and never stated the level it had worked out, whether the token could be sold, where the liquidity sat, or who launched it. All of it was being stored and none of it shown. Now the page opens with those facts, each with its own mark, and a check that could not run reads as "not read" rather than being left out, because leaving it out looks identical to passing.

The token page now draws the price path the market feed has recorded, with the high, the low and how far below its peak the price sits, and shows the day’s volume and the split between buys and sells. It is a line rather than a trading terminal: the question here is whether a token is near its high or most of the way down from it, not when to enter.

At /agent, two simulated accounts of $100 read the same rows in the same pass. One obeys every check and will not buy a token Ocellus has not sold in a simulation. The other ignores most of them and does not wait to be told, drawing one line: it will not buy what the chain has already refused to sell. Neither holds a coin, signs anything or places an order. The careful account exists to test the checks; the reckless one exists so the careful one’s record means something, because an account that does well on its own may only have met a rising market. The rules are published in full, every buy stores the facts it was made on, and a position the chain will not sell is written down to nothing rather than marked at a price nobody would pay. What either account is holding stays closed until a position closes, so the page cannot be read as a live signal feed.

Each token page now says whether each set of published rules would buy it right now, and the first few reasons when they would not. It is a fresh reading of the rules against that scan, not a look at what either account holds.

Every check a scan ran is laid out as labelled facts with marks, instead of sentences joined by middots. Each mark is a shape as well as a colour, and nothing in the findings is red or green: a finding is a record, not a price.

In the screener, a long name with several marks beside it could overflow its column and run into the next one. An automatic table ignores a width set on a cell, which is why capping it had no effect; the contents are capped instead.

Burned, locked, bought back, sold, handed out: each as a share of the whole supply, with the acts underneath in order, because a creator who locked and then sold is telling a different story from one who sold and then locked. Every scan has worked this out for a long time and it could only be found by opening the full report, which is the wrong place for the first thing a holder looks for and the first thing an honest creator wants seen. Nothing here is read as good or bad. Selling is shown in the same voice as burning, because a creator selling their own allocation is not by itself wrongdoing and a burn is not by itself a virtue. It is absent where a scan could not read the whole transfer history, since a partial answer here would read as a complete one.

A token that has left its curve had a chart that began the moment it bonded, so a six-day-old token that graduated this morning showed a few hours and the five days before were simply gone. That is not a fault in the price feed: before a token graduates there is no pool for a feed to know about. Ocellus decodes the curve’s own trades to write its reports, so those days were already in the records, reachable only in a mode of the chart nobody found. They are now drawn in front of the pool’s candles on every timeframe, with the existing marker showing where the token left its curve. Two things are said plainly on the chart rather than glossed: the early stretch is built from the trades a scan decoded rather than from a feed of every trade, and it is priced in the chain’s coin at today’s rate, because a minute-by-minute history of that rate is not something this product has. Where the two sources overlap the pool wins, and the gap at the graduation is left visible because the gap is the graduation.

The actions beside a report’s title are a column on a wide screen, which is right there and wrong on a phone: the layout dropped them under the name and left them stacked, so Follow, Look deeper and Share each took a row of their own and pushed the token’s actual figures a further 76 pixels down. They now sit on one line, which they fit on with room to spare. The contract address has been given a row of its own as well, because it was sharing a line with the chain name and at a larger text size the part that fell off the edge was the Copy control, the one thing on that line anybody is trying to reach. It now takes the full width and truncates its own middle, with Copy pinned where it can always be reached.

Switching timeframe would occasionally leave the chart empty with "No candles for this timeframe yet", on a token whose candles the price feed was perfectly willing to hand over. The cause was that a request which failed was written into the cache as though it were an answer, so one refusal from a public API this server asks constantly left an empty list remembered for a full minute. Only the timeframe that happened to be asked at that moment was affected, which is why it looked random rather than broken, and the failure was swallowed silently so nothing in the logs said otherwise. A refusal is no longer cached and no longer silent: the chart keeps whatever it last knew and the next request tries again, while a genuinely empty answer, from a pool minutes old, is still remembered as the real thing it is.

The screener asks the market feed for every token it lists, thirty a request, and it was asking every twenty seconds. Three thousand tokens is a hundred requests, so three hundred a minute: exactly the feed’s published limit, which meant it was being turned away continuously and the list ran on whatever figures it had last. It now asks every thirty seconds, two hundred a minute, which leaves room for the list to grow and for a retry, and costs ten seconds of freshness on a figure that is only shown for context anyway. The reason nobody noticed is worth admitting: a comment in that file still said the refresh was "about 50 a minute of the feed’s 300", written when the screener held five hundred tokens, and it stayed there while the list grew sixfold past the limit.

The paper accounts re-judge what they hold on every pass, but a token that stopped being priced, or dropped off the screener entirely, skipped every exit check and sat open indefinitely at whatever it was last worth. That would have quietly credited an account with money it could never have taken, which is the same flaw already fixed for tokens the chain refuses to sell. A position that goes twenty-four hours without a price is now closed at nothing, with its own reason. The rest of this site already counts a token no market prices as a total loss when it measures what happened after a scan, and the accounts follow the same convention rather than a kinder one for themselves.

The paper accounts’ list of closed positions showed how long ago each was bought, under a heading that said how long it was held. Five positions opened in the same pass and sold minutes apart therefore all read "after 29 min", which made the record look like it had sold everything at once when the real holds ran from two minutes to twenty-eight. It now measures from opening to closing, and says separately when the sale happened.

"Launches watched" read as a running total and sat directly above "11,105 in the last day", the same figure again. Both counted the same table, and that table is deliberately pruned to twenty-four hours, so counting all of it and counting the last day of it could never differ. It said the same thing in four more places: the funnel below it, the page’s social preview image, the daily Telegram digest, and the post the page offers you to share, which is the worst of them because it invited people to quote a rolling day as a lifetime figure. All of them now say which one it is, and the field that made the mistake possible has been removed rather than corrected, so it cannot be reached for again.

The screener now marks a token "often looked at" when it is among the most opened and searched-for of the week. The line is a percentile rather than a fixed number, so it keeps meaning "unusually looked at" however much traffic the site has, instead of quietly turning into "looked at". The counting keeps a promise the rest of the site already makes: the event log deliberately strips anything that looks like an address, so which token a person opened is never recorded, and this counts the token rather than the person, the way a library counts loans of a book without recording who borrowed it. There is no visitor column in it and there is not meant to be one. A busy token can be busy because it is about to run or because it is about to go to zero, and the label says only that people are looking.

The roadmap never mentioned the off-chain checks, although they run on every scan: the project’s website read for HTTPS, for whether the homepage shows the contract address at all, for whether there is any content or only JavaScript, and for every link it offers; the domain looked up in the registry for its age, registrar and whether the details are hidden; the order of events, which is often the whole story when a domain is younger than the token it belongs to; and the claims a project makes about itself, checked against the chain where they can be. Those are now listed where they belong. A stage of its own has been added for the part that is genuinely missing, reading the accounts a project points at, and it says plainly what it is waiting on.

Every one of the careful account’s first nine exits was the same signal, "the ten largest wallets sold", with positions abandoned after as little as two minutes against a rule that allows twenty-four hours. The light check between scans reports any drop of a single percentage point of supply, which is a sensible share and often a meaningless amount: across the screener the flagged tokens had given up about one point on market caps as low as $3,174, so roughly thirty-three dollars of selling. Leaving now requires the sale to be meaningful both ways at once, a real share of the supply and a real amount of money, because a fixed share is nothing on a small token and a fixed amount is nothing on a large one. On the same data that cuts the signal from 277 tokens to 37, and the ones left are sales of $2,000 to $73,000.

The careful account refused any token where the ten largest wallets selling would halve the price. That reads as a concentration check and was not one: it compared what those wallets hold against the pool’s depth, which is a market-cap-to-liquidity ratio under another name. Worked backwards, it demanded the pool hold about half the supply, which no token does once it has graduated, so it was quietly rejecting almost the whole chain. Concentration was already covered by a separate rule capping the ten largest at 35%. The test has been replaced by the thing it was accidentally measuring, said plainly: the pool must hold at least 5% of what the token is said to be worth. The token page shows that figure too.

Where a scan could not establish where the liquidity sits, the rules printed the word "undefined" at the reader. It now says the check did not run, and such a token is refused rather than quietly treated as fine.

Add @OcellusBot to a group, or message it: /scan an address for the level, the findings and the report; /created for when a token was made, to the minute; /follow a token to hear when its level changes, its largest wallets move, or its creator claims it; /launches low, moderate or all for the first report on each new Robinhood Chain launch, posted once the token is an hour old and still trading; /claims for every creator claim; /digest for the day’s numbers each morning. In a group only its admins change settings; a channel’s admins post the commands in the channel. The bot reads nothing but commands. The Ocellus channel is @ocellusapp.

The highest-scoring smart wallets and every whale are watched between scans: every ten minutes their trades on Robinhood Chain are read from the chain and their records move without a rescan. Any Telegram chat can ask for their buys with /smart on: who bought (by its record, never by name), what, for how much, and how long after the launch. Followed tokens hear their buys and sells. One post per token an hour, thirty a day per chat.

Website, X, Telegram, Discord and GitHub icons beside each token, from the launch profile the creator wrote on the contract and from the market feed’s listing. Every Robinhood Chain launch on record gets its profile links read, so tokens scanned before this keep up. A link is what the project says about itself, not a check.

A public leaderboard at /wallets/top: smart wallets by score, with their tokens, winners, profit, typical buy, holdings and last trade, and whales by what they hold. Traded in the last 30 days, or all time.

Every wallet with a record in three or more scanned tokens gets a score from 0 to 100: its win rate discounted for how few tokens were seen, its profit, how often it was early, and how big its typical buy is, fading for positions last traded months ago. Smart from 40, Top from 60; a wallet that trades like a program scores 0. Whales are now measured on what they hold in the chain’s coin plus their scanned tokens at today’s market, with a Large whale tier from $100,000. The wallets page ranks by score and shows holdings and typical buy; each wallet page explains its score in words.

Every full-history report now lists the creator’s own acts, oldest first with dates: tokens burned, locked in a locker, bought back, sold, or handed out to other wallets, each as a share of the supply. A row in the report, a line under it, and a “Creator burned · locked” figure at a glance.

The holders table can be filtered to wallets that got the token on day one (the first 24 hours after launch) and still hold it, or to wallets holding for 1, 2, 5, 7, 14 or 30 days and more, with the share of supply each group holds; day-one wallets are marked on the map. Known for scans that replayed every transfer from token creation.

Sign in with the wallet that launched a token and you can claim it. A claimed token says so on its report and in the screener (with a filter and a “Claimed” view), and the creator can note, under any finding, where they believe the report is wrong. What they write is shown as their word, next to the finding, with Ocellus’s answer once there is one. A claim changes nothing in the findings.

Every report says how many tokens its creator has launched on Robinhood Chain, over what span, and what became of them; a creator whose launches keep dying counts against the level, and each creator has a page listing every launch. The screener can filter to first launches, or leave out serial creators.

Tokens that go into a contract and out to another wallet in one transaction now read as one move between the two wallets, and link them, the way a direct transfer does.

A contract holding tokens under a lock with an unlock date now reads as a locker: out of the ten largest wallets, with the date in its name and a “Locked tokens” row in the report.

A public page of what Ocellus has looked at and what happened next: tokens scanned, launches watched, wallets on record, the levels, and the median price change after a scan by level. Counted every minute, nothing estimated.

The report’s chart is now a trading chart: candles with wicks and volume, 1 minute to 1 day from the token’s deepest pool (its launch curve included), or the trades the scan itself read, with the all-time high and the last price drawn as lines. Price or market cap, dollars or the chain’s coin, linear or log. The findings stay pinned on it by number, and clicking one still lights the wallets behind it.

Reports now read who holds a Raydium or PumpSwap pool’s liquidity tokens: burned, locked in Raydium LaunchLab’s vault, or held by a wallet that could withdraw them. A Solana token no longer stays at Moderate for a check that could not run.

Every token in the screener is scanned again every six hours, or every day when it is Severe or very small, oldest first, so a report is never much older than that. The two-minute check still queues a scan the moment the largest wallets move.

Report headers and screener rows show the whole contract address; one click copies it.

Every scanned token in one list: its level and the findings behind it, holders, the ten largest wallets’ share, liquidity, how far the price would fall if the top ten sold, split wallets, and how far it sits below its peak. Filter by level, chain, size, age and kind; search by name or address; switch between a table and cards. Four panels at the top rotate through what just got scanned, what is moving, where holders are selling and where smart money holds. Severe levels and very small tokens are hidden until asked for. Not picks: a place here cannot be bought, and a Low level does not mean safe.

A token that has not left its launch curve is not listed by any market feed. Its price, market cap and liquidity are now read from the curve itself every two minutes, so the newest launches have figures too.

The market feed misprices pools quoted in an unusual asset: one token was shown at 1.7 times what a swap actually paid. Every Solana price is now checked against Jupiter’s executable price and corrected where the two disagree. Tokens the feed does not list are priced from Jupiter alone.

The address that holds every Raydium pool’s tokens has no account of its own, so it looked like a wallet holding a third of the supply. It is now a pool: left out of the ten largest wallets, drawn as a pool in the holder map, and older reports are corrected when opened.

Follow a token and Ocellus tells you, in the bell and by email, when its creator sells, when the largest wallets of its last scan sell, or when its liquidity leaves the locker. It is checked every couple of minutes, and you hear about each kind of move once per scan, not over and over.

Between full scans, every token is looked at again every couple of minutes: what the largest wallets of the last scan hold now, what the creator holds, whether the pool’s liquidity is still in the locker and how deep the pool is. All tokens are read together in a handful of calls. A material move (the top wallets shedding five points of the supply, the creator selling half, liquidity leaving the locker) queues a fresh scan by itself.

A new check finds wallets that look like one buyer split across several: identical balances to the last unit, or first receiving the token in the same block in equal amounts. It reads the holder list alone, so it works even on venues whose trades cannot be decoded, and it ignores the round amounts a max-wallet rule produces.

Lists and reports show a token’s own logo: the one its creator set on the token contract at launch, which exists before any market lists the token, or else the one on its market profile. Where a project has published a banner, it sits dimmed behind the report’s heading as the project’s own image. Ocellus keeps a small copy of each (a megabyte logo becomes two kilobytes) and serves it itself, so pages stay fast and your browser never talks to a third party. A token without a picture keeps the glyph drawn from its address. A token in any list now opens its newest report; its scan history is one click from there.

You no longer need a wallet to have an account: create one on the Account page with an email and a password, confirm the email, and you can follow tokens, get alerts, and post and vote on feature requests. Holder credits and buying credits still belong to a wallet. Passwords are stored only as a one-way hash, and a forgotten one is reset by an emailed link.

Lite is now one screen: the market numbers, the chart with the findings pinned to it, what needs a look by name, what is clear, what could not be checked, and three figures (the ten largest wallets’ share, holders, what the creator holds). Pro is a terminal where everything follows one selection: click a marker on the chart, a finding or a wallet, and the chart, the findings, the holder map, the trades and the holders all focus on it, and the link remembers it. Findings are grouped under three questions (who holds it, can they rug it, is the buying real), each with its other reading, and no number is shown twice.

Reports now open on a level of risk signals (Low, Moderate, High or Severe) with a gauge showing how near the next level is, and a tally of serious findings, warnings and checks that could not be read. Findings are weighed, not counted: wallets that bought together and have sold out count for almost nothing, the same wallets still holding a fifth of the supply are serious, and so are minting after launch, unlocked liquidity, a large creator or linked holding, and a claim the chain contradicts. A level can never read Low while a check that matters could not run. It is read from records and cannot see intentions: it is not advice, and low does not mean safe.

Reports now say what a share of the supply cannot: how far the price would fall if the largest wallet, the ten largest, the creator, the linked wallets, the wallets that bought together or the bots sold everything they hold. It is worked out from the reserves that actually set the price, read from the bonding curve or the pool at the snapshot, so it is arithmetic, not a guess. It is the depth of the market, not a forecast.

A new check reads the sells the scan decoded: how many different wallets other than the creator sold and were paid, and what a sale really cost in fees and tax. Plenty of buying with nobody able to sell is flagged as serious; a sale costing a tenth or more is flagged too. It is what happened on chain, not a simulation.

The ten largest wallets’ share is now of the whole supply, as market sites show it, with the share of circulating supply as small print; token pages were recomputed to match. Bots that bought together are counted once, as bots. Linked wallets count in full only where the links were checked (a transfer, or a shared funder the scan measured); wallets tied in only through a funder of unknown size, which may be an exchange or a bridge, are shown as unverified and weigh little.

A link to a report or to a token’s page now unfolds on X, Telegram and Discord into a card with the level of risk signals, its gauge, the findings behind it and three numbers. Share on a public report shows that card, copies the link, opens a post on X or saves the image. The example report uses the new design too.

When two holders were funded by the same address, a scan that ran out of budget used to leave that address unsized, and an exchange or a bridge could then read as a link between strangers. Sizing it is one read, so it now happens even when the budget is spent, and fewer wallets end up in groups they do not belong to.

When a launchpad token leaves its bonding curve, its pool’s liquidity becomes one Uniswap position. Scans now find that position in the graduation transaction and ask the chain who owns it: in the launchpad’s locker it reads as clear, and held by any other address it is a serious finding, because that address can withdraw the money behind trading. Until now this was "not checked", which kept every graduated token from reading as low risk. Ocellus checks who holds the position, not the locker’s code, and says so.

Snipers and trading bots buy most launches, so meeting them again says nothing about a token. "Seen before" now counts only wallets that do not trade like a program, says how many bots it left out, and no longer raises the risk level for them.

Pages now use the full width of the window. The holder map keeps a height that fits the screen, and whales, smart wallets and bots are badged on their bubbles and tagged in the list.

Under the chart a line says what each shape means, and picking a marker says what it is, which finding it belongs to and what the map and tables are now showing; findings drawn on the same spot are listed so none is hidden. "Seen before" now names the other launches those wallets were in, with how many wallets each shares, linked to their pages.

A project that trades or lists meme coins was being categorised as one because its site uses the words. The category now needs the token to say it about itself, and reports already made show the corrected category without a new scan.

Pro’s eight sections are now four: Overview, Holders & money, Project and Evidence. Look deeper and Share are buttons at the top, and the row of status labels and the yellow freshness banner are one line under the token’s name, with a fresh scan offered there when the report is old. Links into the old sections still open the right place, and the previous layout stays one click away at the foot of the report for now.

Reports now open on a bubble map: one bubble per holder sized by its share, linked wallets coloured and drawn together with the funders they share, the creator ringed, and the holder list beside it. Hover either side to find a wallet on the other; zoom and drag to look closer. The earlier funding graph is one click away.

A swap routed on into another asset was being priced from only part of what was paid, which drew prices thousands of times too low and made the market cap jump. Those trades are now recorded without a price, and charts of earlier reports leave them out and say how many.

A holder list that accounts for all but a sliver of the supply (the accounts and the supply are read a moment apart) is now a footnote, not an unfinished report.

A token’s page now says what the whales and smart wallets of the scan before did since: who holds less, who is out entirely and who added, with the share before and after. It is worked out from the whole holder list when a scan finishes, so a wallet that merely slipped down the list is not called gone, and small shifts from dilution are ignored.

For Pons launches, reports now read the description and link the creator stored on the token contract itself, so a token needs no market listing for Ocellus to find its X account and to tell what kind of project it says it is. It is shown on the Project claims tab as the creator’s own words.

Token pages now show a market cap for launchpad tokens too: the live price times the supply the last scan read.

A checklist of what traders ask about (minting, contract owner, liquidity, holder concentration, linked wallets, the website), each marked clear, worth a closer look, or not checked. It counts what was found; it is not a safety rating.

Every scan now records what each wallet did in the token: when it got in, what it spent, what it took out and what it still holds. A wallet that was early and mostly profitable across at least three scanned tokens, without trading like a bot, is called a smart wallet; one holding 20,000 USD or more in ETH or SOL is called a whale. Reports say when either is in a token, and the rules are printed on the Wallets page, which is open to signed-in members.

Every investigated token has a standing page with each public scan of it and what changed from one to the next: holders, the ten largest wallets’ share, linked groups, new buy groups, the creator’s share, whales arriving and mints after launch. A figure is left out when either scan could not read what it rests on. Reach it from any report, the homepage strip or a shared summary.

A wallet that is fully out of most tokens within half an hour of buying, token after token, is marked “trades like a program” on its page, on the Wallets board and in reports, with the reason in words. That now applies to whales too: a large balance no longer hides a bot. Such wallets are left out of “whales arriving” on token pages, and a report’s conclusion now has a “Snipers and bots” line: how many are in the token, how many still hold, and how much of the supply between them.

Open any wallet from a report to see each scanned token it traded, how early it was, its profit, its balance, and the groups it was seen in.

Reports now look for three or more wallets buying within a second or two of each other, at the launch or later, and say what else they share: a funding source, identical buy sizes, brand-new wallets or a link to the creator. A crowded launch on its own is not flagged.

Ocellus now remembers the wallets it finds in buy groups. When they turn up in another token, that token’s report says so, including reports made earlier.

Wallets with a public attribution (KOLs, funds, teams) are named wherever they hold, fund or trade a token, with whether they sit inside a linked group and a link to the source of the name.

Every report opens as Lite: the conclusion, the price, four key figures, what stands out and what could not be checked, on one short page. Pro is the full report with every holder, wallet link, trade and record. Switch whenever you like; your choice is remembered.

The chart’s price, market cap and right-hand edge stay current while a report is open, refreshed every 20 seconds. Pons tokens are read from their own market on Robinhood Chain; others come from DexScreener, and only from a pool deep enough to mean something. The stretch between the snapshot and now is drawn dashed, because it was not read.

On the Summary and under Trading history, drawn from the trades the scan itself decoded, with what was bought and sold beneath it. It is a record up to the snapshot, not a live market feed, and can be switched to ETH or SOL.

It opens on the wallets that matter, draws look-alike counterparties as one group, labels every wallet, shows a summary on hover, and has zoom buttons and a wallet finder. The page scrolls past it instead of getting stuck.

Each finding shows its evidence strength and any dispute as tags. Open it for the records, the other reading and what could not be checked.

A real full AI investigation of Ocellus’s own token, run once and saved as it came out: every specialist’s findings, the Skeptic’s challenges and the Lead’s summary are the model’s. The homepage’s figures come from it too.

A deployer’s drip airdrops (the same amount sent to several wallets over time) are recognised like its single-block ones, and spam or promotional tokens landing in many wallets no longer make the deployer and the top holders look funded by the same address. Fair-launch checks now state the airdrop in the sentence itself.

Without a name, site or description that says what a token is, its category stays unclassified, and sites describing a product now count towards utility.

Before, the change could happen below the screen.

Paste an address and press Quick Scan. There is no chain to choose: Ocellus reads it from the address and shows which it recognised.

Deeper investigations can have each specialist’s findings written and challenged by an AI model, kept only when every cited record exists.

They also look at whether the examined holders keep buying the same other launches, and get more time so funding tracing finishes on busy tokens.

The ring turns, a sweep circles it and the iris opens and closes. Its eight long marks are the eight stages: one lights as each finishes.

A token that anyone follows gets a fresh free Quick Scan about every 24 hours, run only while nobody else is waiting in the queue.

How many investigations are ahead of it, and roughly how long that takes.

At the top, with a copy button and one click to investigate it.

Hold about 25 USD of OCELLUS for an hour and claim 60 investigation credits every 30 days, from your Account page: 6 deep investigations or 2 with full AI review. This is a beta rate, more generous while Ocellus is in public beta. Valuation keeps working after a token leaves its bonding curve.

Follow up to ten wallets from their pages. When one buys a token on Robinhood Chain, an alert arrives in the bell at the top of the page within about five minutes, with one click to scan that token. Tokens that simply arrive in the wallet (airdrops, transfers) are not alerts.

When a token you follow gets a new public scan, you are told what changed since the scan before, and the alert opens the token’s page.

Add an email on your Account page and confirm it to get alerts there too, at most one email every 15 minutes. It is optional, used for nothing else, and removable.

Signing up for alerts sends one email, with a one-click way off the list.

Signed in, your Account page keeps every scan you start and a list of tokens you follow, each with its latest report one click away. There is a Follow button on every report.

Your credits first, then how far your OCELLUS holding is from the free-credit threshold, with the pricing detail one click away.

Ask for what Ocellus should do next and vote on what others asked for, on the Feature requests page (linked in the footer and on the roadmap). Voting needs a signed-in wallet, so it is one vote per person, and the team answers each request on the page: planned, being built, shipped, or not planned and why.

Accounts, credits, reports, follows and alerts are copied every six hours to storage kept apart from the servers Ocellus runs on.

Press / or ⌘K (or use the box in the header) to jump to any page, or paste a token or wallet address to scan it or open it, from wherever you are.

Panels sit above the page with depth, buttons and fields share one look and one focus ring, tables keep their headers and highlight the row you are on, pages arrive instead of appearing, loading shows the shape of what is coming, actions confirm with a brief note, and every token and wallet has its own small aperture glyph.

The homepage shows the latest public Quick Scans, one click from each report, and a link to a public report now unfolds into a card with the token and its key figures on X, Telegram and Discord.

Every page now sits on the same width and edges as the header, and the changelog, roadmap and account pages share one dashboard layout with their key numbers up top.

A proper menu on every page, a shorter report header and a tidier scan page.

The scan page and live reports share the homepage’s scan bar, headline and aperture.

For Robinhood Chain and Solana tokens, with no wallet needed.

Every finding links to the on-chain records behind it, states what it could not check, and carries the Skeptic’s challenges.

A full investigation is open without scanning anything.

Reports can be shared with a link, and only by their owner.

Ask to be told when social-signal alerts open. Only the email address and your consent are stored.

Terms, a privacy notice and a risk notice are published.