Reading the chain.

Three kinds of statement

Everything in a scan is one of three things. Observed: read directly from the chain, such as a transfer, a balance, a swap or a contract event. Inferred: a hypothesis drawn from observations, such as two wallets sharing a funder. Unavailable: a source that failed or does not cover the case.

Keep them apart when you read. An observed transfer is a fact about the ledger. An inferred relationship is a reading of several facts, with other explanations that may also fit. An unavailable check is neither a pass nor a zero; it is a gap.

A scan is a snapshot

A scan reads the chain at the block it runs. Balances, liquidity and contract permissions can change in the next block. Every scan shows when it was read; check that time before acting on anything it says.

The same token scanned an hour apart can give different results because the chain changed, not because the method did. When the method changes, the score version changes with it and the score changelog says what moved.

What the ledger never holds

Names, agreements, intentions and off-chain messages are not on the chain. Two wallets controlled by one person look like strangers if they never touch each other on-chain; two strangers can look connected if they used the same exchange. Read every connection as a link in the ledger, not as an identity.

Top ten, minus the pool

The holder table lists the largest wallets outside the liquidity pool. The pool holds tokens for trading, not for one owner, so it is excluded. The method flags the top ten holding 30% or more of supply, and weighs it more at 50%. One wallet holding 10% or more is flagged on its own.

Clusters change the arithmetic

Ten wallets holding 3% each read very differently if all ten were funded by the same address. A cluster groups wallets linked by a shared funder, by transfers between them, or by buying in the same transaction. A cluster holding 8% or more is flagged; 20% or more weighs more.

A cluster is a set of links. It suggests the wallets may move together; it does not establish who controls them. Exchanges, bridges and routers fund thousands of unrelated wallets, so a shared funder that is one of those never links wallets on its own.

How to read it

Select holders in the lab to see their combined share. Open a cluster to see the links behind it and how strong each one is. Then ask the practical question: if this group sold together, what would the pool return? The exit curve answers that from the pool’s reserves.

Bundles and snipers

Supply bought in the launch block or blocks by wallets that share a funder or a block is counted as bundled. 5% or more is flagged, 15% or more weighs more. Five or more buyers in the first seconds after the pool opened are flagged as snipers.

Timing alone does not prove a private bundle was submitted to a block builder; that needs a relay or transaction source the scan does not have. What the scan shows is the measured share and the wallets behind it.

Fresh wallets and shared gas

A fresh wallet has almost no history before it bought. When 40% or more of the top holders are fresh, the method flags it. When four or more wallets got their first ETH from one source within two blocks, they are flagged as gassed together.

Fresh wallets have ordinary explanations too: new users, privacy habits, a trader rotating addresses. The flag records the pattern; the funding trail is where you test it.

Hop by hop to a hub

The funding trail follows first-funding transactions back, one hop at a time, until it reaches a hub: an exchange, a bridge or a router. Hubs end the trail because everyone uses them. A labelled hub never links two wallets.

Below the hub, a shared funder links wallets. When that link ties a wallet to the token’s creator, the wallet counts as linked to the creator, and the share of supply those wallets hold is reported. 3% or more is flagged; 10% or more weighs more.

Many wallets, one source

Sybil wallets are wallets funded alike by one source or airdropped in bulk. When they make up 25% or more of the wallets checked, the method flags it. The list of wallets and the funding transactions are in the scan.

Testing a link

Open the funding transaction in the block explorer. Check the amount, the time and whether the funder sent to many other wallets. A funder that sent to hundreds of unrelated wallets is more likely a service than a person; the scan’s labels improve as these are found and recorded.

This launch

The creator wallet holding 10% or more of supply is flagged, and so is selling 50% or more of what it had. Tokens the creator sent to other wallets without a trade, 2% or more of supply, are flagged as off-market transfers, and the receiving wallets are listed.

Earlier launches

A creator’s earlier Torch launches are listed with their recorded outcomes. One or more that collapsed or ended with the creator sold out weighs 25, the heaviest single check. Three or more other launches in 14 days are flagged as serial launches, and early buyers who were also first-minute buyers in the creator’s earlier launches are flagged as repeat early buyers.

Other creators tied to this one by a funding source or identical bytecode are shown as the same team, with how each link was made. A first launch has no history to read, which is not the same as a clean one.

How we write about creators

Every statement about a creator restates a measured fact: an amount, a share, a count, an outcome by a fixed rule. The record describes what a wallet did, never why. A creator who thinks the record is wrong can ask for a correction, and the decision is published.

The pool

Pool liquidity under $5,000 is flagged as thin. A pool position not held by a known locker is flagged as unlocked. A lock answers one question, when the position can be withdrawn; it says nothing about minting, taxes, upgrades or other pools.

Quoted liquidity is not executable depth. The exit curve shows what selling a given bag into the pool would return at each step, from the pool’s actual reserves.

The contract

A buy tax over 10% is flagged. Owner powers such as minting, pausing or changing fees are read from the code and reported with the scan. An upgradeable proxy is read from its storage slots: who the admin is, and whether a timelock applies. Bytecode that differs from the launchpad’s standard template is shown as a diff.

The sell simulation tries one sale along one route at the moment of the scan. Passing it once does not mean a sale will go through later. A simulation that could not run is shown as unavailable, never as passed.

Round trips

Volume that went round-trip in repeated buy-and-sell pairs is counted as wash trading; 20% of volume is flagged and 40% weighs more. Two wallets passing tokens back and forth also fire it. Volume is the number most launch boards sort by, which is why it is worth checking.

Borrowed names

An earlier token with the same name or ticker is flagged as a copycat. Names written with characters from another script that read as an earlier token are shown with each lookalike character and the script it came from. A logo or description reused from other creators’ launches is shown with the launches it matches.

Four outcomes, fixed rules

Creator sold out: the price fell 80% or more from its peak while the creator wallet sold. Collapsed: the price fell 90% or more from its peak. Went quiet: no trades for three days. Still trading: none of the above.

The rules use the pool’s price and the creator wallet’s balance, nothing else. An outcome describes what happened on-chain; it is not a finding of intent or wrongdoing.

Why outcomes matter to the score

Outcomes are how the method is held to account. Calibration measures how often launches in each band went on to collapse or sell out; the flag record measures how often each flag preceded one. Both are published and both move as the record grows.

Gaps inside a scan

Funding traces still pending, trades not yet priced, a contract read that failed, a pattern check that did not run: each lowers the scan’s data completeness and widens the range the score could fall in. Missing data mostly hides risk, so the range reaches further up than down.

Gaps outside any scan

Off-chain coordination, people using exchanges between wallets, and the next block are outside what a ledger can show. A scan covers one chain at a time. It cannot predict a price and it is not investment advice.

Checking us

Every scan can carry a signed receipt over its hash, so anyone can check it was not changed after it was published. If a scan, a label or a page is wrong, ask for a correction: requests are acknowledged within 48 hours and decided within 7 days.