Token due diligence, with receipts
How Lyris reads a Robinhood Chain token from the chain, scores it by public rules, sets the project's claims next to the evidence, and signs every run.
Abstract
Lyris is a due-diligence tool for tokens on Robinhood Chain. Given a contract address, it reads the contract, the full transfer history and the token's pools, scores eleven checks by fixed public rules, compares the project's website with what the chain shows, and writes a verdict. Each run is recorded in a SHA-256 hash chain and signed with Lyris's Ed25519 key, so a report can be shared and checked later. A general agent with live tools covers wider research. Lyris's token launches soon on Robinhood Chain.
Motivation
New tokens on Robinhood Chain launch every day, and the facts that matter most before buying are already public: who controls the contract, what it can do, who holds the supply, and how deep and how locked the liquidity is. Lyris reads those facts in one pass and puts them in plain words.
A project's website tells its own story. Lyris sets that story next to the chain, claim by claim, so the reader sees both at once.
How a check runs
- 1Confirm the contract
Lyris checks there is code at the address and that it answers decimals() and totalSupply() like an ERC-20.
- 2Read the contract
owner(), the EIP-1967 implementation slot, and the bytecode, scanned for the selectors of admin functions.
- 3Rebuild the holders
Every Transfer event from genesis, in 10M-block windows, up to 80,000 transfers. Balances are summed per address, and the largest 15 holders are labelled.
- 4Read the pools
DexScreener's pools for the token, plus the LP burn share read from each v2 pair contract.
- 5Score
Eleven checks, each with fixed thresholds and a fixed maximum, combined into a 0–100 score.
- 6Compare and write
The agent reads the project's website, builds the site claims vs chain table and writes the verdict.
- 7Sign
The run is hashed step by step and the receipt is signed.
Scoring model
Checks fall into three groups: Contract (ownership, upgradeability, admin powers), Holders (top 10 wallets, deployer and linked wallets, holder count, supply in other contracts) and Liquidity (USD liquidity, liquidity vs market cap, LP burn, pool age).
| Check | What it reads | Points |
|---|---|---|
| Ownership | owner() on the token contract | Renounced to the zero or dead address: 10 of 10. No owner() function: 8 of 10. Owned by a wallet: 3 of 10. |
| Upgradeable | The EIP-1967 implementation storage slot | Empty: 10 of 10. Set, so the token is a proxy whose code can be replaced: 0 of 10. |
| Admin powers | The token's bytecode (and its implementation's, for a proxy), scanned for mint, blacklist, pause, fee, max-transaction and trading-switch functions | None found: 10 of 10. Found with ownership renounced: 6 of 10. Fee, pause, cap or trading-switch only: 4 of 10. Mint or blacklist: 0 of 10. |
| Top 10 wallets | Balances rebuilt from every Transfer event; pools, other contracts and burn addresses are set aside | Under 30% of circulating supply: 15 of 15. Under 50%: 8 of 15. 50% or more: 0 of 15. |
| Deployer & linked wallets | The wallet that sent the first mint transaction, plus every wallet that received tokens straight from it or from the first mint's recipient | Under 10%: 15 of 15. Under 30%: 7 of 15. 30% or more: 0 of 15. |
| Holders | Every address with a balance, from the full transfer history | 100 or more: 5 of 5. 20 or more: 2 of 5. Fewer: 0 of 5. |
| Held by other contracts | Supply among the largest holders that sits in contracts other than the token's pools (launchpads, lockers, vesting or team vaults) | Under 20%: 10 of 10. Under 50%: 4 of 10. 50% or more: 0 of 10. |
| Liquidity | Every Robinhood Chain pool DexScreener lists for the token, in USD | $50K or more: 10 of 10. $10K or more: 5 of 10. Less, or no pool: 0 of 10. |
| Liquidity vs market cap | Pool liquidity as a share of DexScreener's market cap | 10% or more: 5 of 5. 3% or more: 2 of 5. Less: 0 of 5. |
| LP tokens burned | For the main v2 pool: the LP supply held at the zero and dead addresses, read from the pair contract | 90% or more: 10 of 10. 50% or more: 5 of 10. Less: 0 of 10. |
| Pool age | When the first pool opened, from DexScreener | 7 days or more: 5 of 5. 1 day or more: 2 of 5. Under a day: 0 of 5. |
The score is points earned over points possible for the measured checks, scaled to 0–100: 75 and up reads Looks solid, 50 to 74 Mixed, under 50 Red flags. A check that can't be measured, such as an LP lock on a v3/v4 pool, is marked unknown and left out of both sides of that ratio.
The agent's read
The scored report is the evidence; the agent turns it into a read a person can act on. It takes 2 to 4 concrete claims from the project's site (team, utility, supply, locked liquidity, partners, audits) and marks each one Yes, No or Can't tell against the report. It then writes a verdict, red and green flags, and three checks to do before buying, using only the report and the pages it read.
Proof of run
Each run produces a receipt. The chain starts at the hash of the task, adds each tool step as the hash of the previous link, the tool name, and the hashes of the step's input and output, then closes with the answer hash, the model and the time. The final hash is the receipt id, signed with Ed25519. The public key and a signature check are served at /api/receipt.
A valid receipt shows the report is the one Lyris produced and that it is unchanged since. The data it records is what the chain, DexScreener and the project's site said at the time of the run.
The general agent
Beyond token checks, Lyris runs a general agent with six tools: token_check, read_page, chain_read on Robinhood Chain, token_price from CoinGecko, defi_data from DefiLlama, and calculate. An agent run plans, uses up to 14 tool calls across up to 8 model turns, and closes with a Result, Sources, Next actions and a receipt.
Model routing
Lyris answers with free models on OpenRouter, chosen per task. Nemotron 3 Ultra (NVIDIA) leads for agent runs, token checks and analytical questions; Gemma 4 31B (Google) leads for quick questions; Nemotron 3.5 Super (NVIDIA) and Gemma 4 26B (Google) step in when the first choice is busy. The model that wrote each answer is shown with it and named in its receipt.
To share the free capacity fairly, each visitor can start 10 agent runs an hour (token checks included) and send 60 chat messages an hour.
Token
Lyris's token launches soon on Robinhood Chain. Today the holder dashboard reads any wallet's token balance live from the token contract. Holder programs are on the roadmap below.
| Item | Value |
|---|---|
| Ticker | Announced at launch |
| Chain | Robinhood Chain |
| Contract | Added at launch |
Roadmap
Every item in this section is Planned. Each one moves to Live on the transparency page the day it ships.
- 1Planned · Holder credit vouchers
Every 1M tokens held earns one voucher of agent credit a day, from a 100K minimum, starting at $0.40 a voucher. Claimed from the holder dashboard with one wallet signature.
- 2Planned · Token payment bonus
Top up agent credit with the token and get 20–25% more credit, with extra holder discounts from 100K, 1M and 10M tokens.
- 3Planned · Buyback and burn
Tokens paid for credit and a share of revenue go to burns. Every burn will be listed on the holder dashboard with a link to its transaction.
- 4Planned · Sealed runs
Runs executed inside a hardware enclave, with the receipt signed by a key that lives in the enclave.
- 5Planned · More models
Larger paid models alongside the free ones, picked per task the same way.
Conclusion
Lyris reads a token the way a careful buyer would, all at once: the contract, the holders, the liquidity and the project's own words, scored by public rules and signed so the report can travel. The token grows with it as the planned holder programs ship.