Archalis Radar rating methodology

Version 3, October 17, 2026.

Archalis Radar rates every BNB Chain token from AAA to C, the way credit rating agencies rate bonds. The rating measures how likely it is that a buyer loses their money for structural reasons: a rigged contract, liquidity that can be pulled, fake volume, creators with a track record of rug pulls. It does not measure upside and it is not investment advice.

Two outcomes are not grades:

• SCAM: a confirmed scam pattern. We do not grade these tokens, we label them.
• D (default): the scam already happened: the liquidity was pulled, or insiders sold their tokens into the pool.

All assessments are automated and based only on public blockchain data that anyone can verify.

Scale

Candidate ratings

A token younger than 24 hours receives a candidate rating, for example "candidate CC". It is too young to tell whether it is a scam: many rug pulls happen within the first hour. The rating becomes final once the token passes its first day without incidents. Candidate ratings are shown with a dashed badge.

SCAM label

A token is labeled SCAM, instead of being graded, when at least one of these patterns is confirmed:

• launched with an automated service that we have observed pulling liquidity in nearly every launch;
• rigged balances: a wallet holds more than the whole supply, so it can sell tokens that do not exist;
• honeypot: buying works, selling does not (verified by simulation);
• confiscatory sell tax (50% or more);
• LP tokens handed to a liquidity-removal contract.

What counts as a rug pull

A pool whose liquidity falls by 70% or more from its peak is examined transaction by transaction. For every unit of BNB or stablecoin that left the pool we establish how it left:

• liquidity pulled: whoever holds the LP tokens withdrew the liquidity;
• sold by insiders: the tokens were sold into the pool by the creator, the contract owner, the deployer, the wallets that funded the creator, the creator's volume bots, wallets that received their tokens directly from another wallet instead of buying them, or large sellers funded by any of these;
• sold by market buyers: wallets that bought from the pool and sold back.

It is a rug pull (default, rating D) when liquidity pulled and insider sales together account for at least half of the peak. Insider selling is a rug pull even when the LP tokens are burned or locked: a new token starts entirely in its creator's hands, and whatever is not put into the pool can be passed to other wallets and dumped later.

Otherwise it is a price crash: ordinary market selling. Price crashes are reported separately and are not defaults.

The same analysis covers V2 and V3 pools. When most of a price crash comes from one group of wallets funded by the same wallet, the report shows it, but it is not counted as a default unless that group is linked to the creator.

How grades are computed

1. Six pillars

Each pillar is scored from 0 to 100. The rating score is the weighted average.

2. Rating ceilings

As with the "rating ceilings" used by agencies, some weaknesses cap the maximum rating whatever the average: a great score elsewhere does not offset a structural risk.

3. Young tokens: a ladder calibrated on observed rug pulls

For tokens younger than 30 days, the pillar score alone does not tell which ones will rug: scammers use clean contracts and lock their liquidity for show. What does separate them, in the data, is the creator's network and the time a pool has survived without incidents. Over the first day of data (October 16-17, 2026, BNB Chain pools with at least $1,000 of liquidity):

So a young token's rating cannot go above:

The rating rises on its own as the token survives: every new pool is re-rated at 5, 15, 30 and 60 minutes and at 3, 6, 12 and 24 hours. Grades are meant to match bands of observed rug-pull probability: C above 95%, CC 70 to 95%, CCC 30 to 70%, B 10 to 30%, BB below 10%. The thresholds are reviewed as data grows; the lower bands rest on few cases so far.

4. Outlook

• Negative: dangerous changes were observed after launch (taxes raised, ownership taken back, liquidity falling).
• Stable: otherwise.

Updates

Every new pool is rated at birth and then re-rated:

• after 5, 15, 30 and 60 minutes, with a full analysis;
• after 3, 6, 12 and 24 hours, with a liquidity check and a rating recalculation.

Established tokens are BNB Chain tokens with at least $100,000 of liquidity on the DEXes the radar reads and at least 7 days of history: the top 1,500 by market cap in the BNB Chain category, the most traded of the day, and any established token analyzed on request. The most traded are re-rated every 30 minutes, the others every 6 hours. Watched tokens are re-checked continuously.

Validation

Default rates count only ratings given with the current version of the method (version 3, since October 17, 2026), so that results are never mixed with an older, less accurate scale.

A rating is judged by its results. For each grade the radar publishes the default rate: the share of tokens that ended in a rug pull after receiving that rating at their first analysis. Price crashes are shown next to it but are not counted as defaults. Only the first rating is used, with no hindsight. It is the same kind of evidence traditional rating agencies publish.

Known limits

• Distribution considers the creator, the owner and the supply kept outside the pool at launch, not yet all large holders.
• A group of wallets that is not linked to the creator by funding is treated as market buyers, even when it sells in a coordinated way.
• For V3 pools, ownership of the liquidity positions (NFTs) is not verified yet: they are capped at AA.
• Volume bots and launch services are recognized only once known, by address or by code signature.
• Tokens whose main market is on PancakeSwap Infinity or Uniswap V4 are not rated yet: the radar does not read those markets.
• Off-chain issuer reputation is not considered. A custodial token whose issuer can mint new tokens does not reach AAA even when it is reliable.