chainprint

What is this? A local, offline command-line tool that inspects a source tree and reports the evidence it finds that the code targets Robinhood Chain (chain id 4663), with a deterministic, documented confidence.

Why it exists

Builders, reviewers and research tools often need a quick, explainable answer to "does this repository build for Robinhood Chain, and on what evidence?". chainprint answers it from the files alone: chain ids in toolchain configs, deployment records, the public RPC and explorer hostnames, known public infrastructure addresses, env templates, package metadata and documentation. Every marker comes with its file and line, and the confidence is a sum anyone can recompute by hand.

Why Robinhood Chain only

chainprint looks for one chain: Robinhood Chain, chain id 4663 (0x1237), CAIP-2 eip155:4663. It has no network option and no chain list. Other chain ids are simply not markers (a config for chain 1 scores 0); the Robinhood Chain testnet (46630) is reported in its own labelled section and never counted.

Install

npm i -g @hey-research-lab/chainprint
# or, without installing
npx @hey-research-lab/chainprint .

Node.js 22 or later. One runtime dependency: zod, used to check the structure of the JSON files it parses.

Smallest working example

mkdir demo && cd demo
cat > hardhat.config.js <<'EOF'
module.exports = {
  networks: { robinhood: { url: 'https://rpc.mainnet.chain.robinhood.com', chainId: 4663 } },
};
EOF
npx @hey-research-lab/chainprint .

Robinhood Chain evidence found
Confidence: HIGH (score 60; with a config- or deployment-level marker)

Evidence:
  - chain id 4663 in a toolchain config [chain-id-config, config, +40]
      hardhat.config.js:2  networks: { robinhood: { url: 'https://rpc.mainnet.chain.robinhood.com', chainId: 4663 } },
        matched "chainId: 4663"
  - Robinhood Chain public RPC hostname [rpc-host, code, +20]
      hardhat.config.js:2  networks: { robinhood: { url: 'https://rpc.mainnet.chain.robinhood.com', chainId: 4663 } },
        matched "rpc.mainnet.chain.robinhood.com"

Read 1 file.
A marker is evidence, not identity proof. It shows that code targets Robinhood Chain; it does not show who owns the code or any contract, that the repository is a project, or that a project is legitimate.
Run `chainprint explain` for the rules, weights and confidence bands.

Usage

chainprint [path] [options]   scan a local directory (default: .)
chainprint explain [--json]   print the rules, weights and confidence bands

--json          one JSON document (schema chainprint/v1) on stdout
--quiet         only the confidence and score, e.g. "HIGH 60"
--limit <n>     evidence items listed per rule (default 20, max 1000)
--no-color      no emphasis (NO_COLOR is honoured too)
--help, --version

To scan a directory literally named explain, pass ./explain.

Exit code	Meaning
0	the scan ran, whatever the confidence (NONE included)
1	the tree could not be read
2	usage error: unknown flag, more than one path, bad --limit, a path that does not exist or is not a directory

As a library:

import { scanDirectory, computeConfidence, RULES } from '@hey-research-lab/chainprint';

const report = scanDirectory('./my-repo');
console.log(report.confidence, report.score, report.evidence.length);

The rules

Each rule has a fixed name, weight and tier. chainprint explain prints this table from the code itself.

Rule	Weight	Tier	Reads	Matches
foundry-broadcast	45	deployment	broadcast/<script>/4663/run-*.json	parses as JSON with a transactions array; a chain field, if present, must say 4663. dry-run/ is not a record.
hardhat-deploy	45	deployment	deployments/<network>/.chainId	the file's whole content is 4663
ignition-deployment	45	deployment	ignition/deployments/chain-4663/deployed_addresses.json	parses as a JSON object
chain-id-config	40	config	hardhat.config.*, foundry.toml, truffle-config.js, truffle.js, wagmi.config.*, any JS/TS file calling defineChain(	a chain-id key (chainId, chain_id, chainID, networkId, network_id, chain) set to 4663 or 0x1237; in a defineChain file also id: 4663
chain-id-code	20	code	other source and data files	an identifier or key containing chainid/chain_id compared with or assigned 4663 (CHAIN_ID = 4663, block.chainid == 4663, chainId: '0x1237'), or an ethers provider/network built with 4663
caip2	20	code	source, data, config	eip155:4663 not followed by a digit
rpc-host	20	code	source, data, config	rpc.mainnet.chain.robinhood.com as a whole hostname
known-address	20	code	every file except documentation	an address listed in data/known-addresses.json
env-template	15	environment	.env.example, .env.sample, .env.template, .env.dist, .env.defaults, .env.<name>.example, example.env	a key containing ROBINHOOD; a CHAIN_ID key set to 4663; a value naming the RPC or explorer host or eip155:4663
explorer-host	10	code	source, data, config	robinhoodchain.blockscout.com as a whole hostname
package-metadata	10	package	package.json	a keyword robinhood-chain / robinhoodchain / robinhood chain / eip155:4663; a dependency whose name contains robinhood-chain or robinhoodchain; a name or description containing "Robinhood Chain"
chain-name	5	mention	source, data, config (not package.json)	"Robinhood Chain" in any case, joined by a space, -, _ or nothing (robinhoodChain, ROBINHOOD_CHAIN), but not as a hostname label
doc-mention	5	mention	*.md, *.mdx, *.markdown, *.rst, *.adoc, *.txt	the name, eip155:4663, either hostname, or "chain id" followed by 4663. The weakest marker: anyone can write a README.
chain-listing	0	context	chain lists (chains*.json, *chain-registry* files, chainlist/), vendored chain definitions (chains/definitions/), agent instruction files (.agents/, .cursor/, .codex/, .windsurf/ and similar folders; SKILL.md, AGENTS.md, GEMINI.md and similar)	any marker another rule would count, found in such a file: it says the code knows the chain, not that it targets it. Reported, never counted.
testnet-marker	0	context	every file except documentation	46630 / 0xb626 / eip155:46630 where the mainnet rules look for 4663, and testnet deployment folders. Reported, never counted.

A number only matches as a whole number: 46630, 14663 and 4663.5 are not 4663.

One class per file. Each file is given one class, which decides the rules that read it:

Class	Rules
deployment record (the three layouts above, not under test/, tests/, fixtures/, mocks/, examples/, samples/ and similar folders)	its deployment rule (or the testnet marker) and known-address
chain list, vendored chain definitions or agent instruction file (see chain-listing; checked before every class below)	every rule its file type would get, each match reported as chain-listing (testnet markers stay testnet-marker)
env template	env-template, testnet-marker, known-address
documentation	doc-mention only
package.json	package-metadata, and the code rules except chain-name
toolchain config	chain-id-config instead of chain-id-code, plus the other code rules
anything else	chain-id-code, caip2, rpc-host, explorer-host, chain-name, known-address, testnet-marker

A deploy record under a test, fixture, mock, example or sample folder is someone's test data, not this repository's deployment, so it is read as an ordinary file.

The confidence formula

score  = the sum of the weights of the distinct counted rules that matched
         (a rule counts once, however many files or lines it matched)

HIGH   score >= 60 and at least one config- or deployment-tier marker
MEDIUM score >= 25
LOW    score >= 1
NONE   score = 0

The bands are checked from the top; the first whose conditions hold is the confidence. Without a config or deployment marker, even a large score stays MEDIUM: a repository can mention the chain in many places without ever configuring or deploying to it. Testnet markers have weight 0. Changing a weight, band or pattern changes the rules version (now chainprint-rules/2), which every report carries. Version 2 (chainprint 0.2.0) added chain-listing; every other rule, weight and band is unchanged from version 1.

Worked examples (all in test/fixtures/):

Tree	Rules matched	Score	Confidence
Hardhat config with chainId: 4663 and the RPC URL	chain-id-config 40 + rpc-host 20	60	HIGH
Foundry broadcast for 4663, RPC in foundry.toml	foundry-broadcast 45 + rpc-host 20	65	HIGH
hardhat-deploy deployments/robinhood/.chainId only	hardhat-deploy 45	45	MEDIUM
ethers provider with 4663, RPC host, Uniswap V2 factory address, a comment naming the chain	20 + 20 + 20 + 5	65	MEDIUM (no anchor)
.env.example only	env-template 15	15	LOW
README only	doc-mention 5	5	LOW
Config for chain 1	—	0	NONE
Testnet 46630 only	testnet-marker (not counted)	0	NONE
Vendored viem chain definition, a chain registry, AGENTS.md, and docs/ROBINHOOD.md	chain-listing (not counted) + doc-mention 5	5	LOW

Known addresses

data/known-addresses.json lists 50 public Robinhood Chain infrastructure contracts — the Uniswap V2 and V3 factories, the Uniswap v4 PoolManager, Doppler's Airlock, and public launch factories and one launch router. Every entry cites the public page that names it (sources) and the public HEY Research Lab registry that lists it with its on-chain check (listedIn, in hey-research-open). Contracts deployed at the same address on many chains (Multicall3, Uniswap's CREATE2 Liquidity Launcher) are listed under excluded and never match, because finding them says nothing about Robinhood Chain. An address can still exist on more than one chain; a match shows the code refers to that address, nothing more. chainprint compares addresses lowercase and does not validate EIP-55 checksums.

JSON output (chainprint/v1)

{
  "schema": "chainprint/v1",
  "ok": true,
  "tool": { "name": "chainprint", "version": "0.2.0", "rules": "chainprint-rules/2" },
  "chain": { "name": "Robinhood Chain", "chainId": 4663, "caip2": "eip155:4663" },
  "root": ".",
  "result": "evidence_found",
  "confidence": "HIGH",
  "score": 65,
  "anchored": true,
  "rules": [
    {
      "rule": "foundry-broadcast",
      "tier": "deployment",
      "weight": 45,
      "counted": true,
      "items": 1,
      "shown": 1
    },
    { "rule": "rpc-host", "tier": "code", "weight": 20, "counted": true, "items": 1, "shown": 1 }
  ],
  "evidence": [
    {
      "rule": "foundry-broadcast",
      "tier": "deployment",
      "weight": 45,
      "counted": true,
      "file": "broadcast/Deploy.s.sol/4663/run-latest.json",
      "line": 27,
      "excerpt": "\"chain\": 4663,",
      "detail": "Foundry broadcast for chain 4663 with 1 transaction"
    }
  ],
  "scan": {
    "filesRead": 4,
    "complete": true,
    "incomplete": [],
    "skipped": {
      "total": 1,
      "byReason": { "foundry-library": 1 },
      "shown": [{ "path": "lib/", "reason": "foundry-library" }]
    },
    "limits": {
      "maxFiles": 10000,
      "maxFileBytes": 1048576,
      "maxTotalBytes": 268435456,
      "maxDepth": 12,
      "itemsPerRule": 20
    }
  },
  "notice": "A marker is evidence, not identity proof. …"
}

• result is evidence_found when the score is above 0, else no_evidence_found — a reading of the files chainprint read, never a statement about the world.
• rules[].items is the total found; shown is how many are listed in evidence (--limit).
• scan.complete: false means a limit stopped the walk (incomplete says which). Markers in files chainprint did not read are not measured: unknown, not absent.
• On an error, --json prints { "schema": "chainprint/v1", "ok": false, "error": { "code", "message" } }.
• The output carries no timestamp, so the same tree always gives byte-identical output.

Limitations

• A file over 1 MiB is not read, deployment records included; it is listed under scan.skipped and its markers are not measured. A very large Foundry broadcast can fall under this limit.
• Patterns read one line at a time: a chain id split across lines, built at run time, or read from an environment variable is not seen.
• chainprint reads what is in the tree. A marker can be copied, stale or aspirational; a missing marker can live in a private repository, a submodule or a skipped directory.

How it relates to HEY Research Lab

chainprint is an independent, standalone tool from HEY Research Lab, the builder-evidence layer for Robinhood Chain. It uses no HEY API and sends nothing anywhere. Its rules are its own, written from public chain facts; they are not HEY's discovery or qualification logic, and a chainprint result does not put a repository on HEY or change anything there. HEY's own review stays authoritative for what HEY publishes. The known-address list is taken from HEY's public factory registry in hey-research-open. Developer docs: https://heyresearch.xyz/developers.

Compared with HEY's own code-search rules (as of 2026-10-09). HEY's discovery does not count a Robinhood Chain marker found in a lockfile, a chain registry or chain list, a changelog, a vendored tree, an agent's instruction file, or prose other than a README or a document named for Robinhood Chain. chainprint follows the same reading where it changes the result:

• lockfiles and vendored dependency folders are not read at all (since 0.1.0);
• chain lists, vendored chain definitions and agent instruction files are reported as chain-listing and never counted (since 0.2.0, rules version 2) — before that, a copied defineChain for Robinhood Chain alone scored HIGH;
• a README and a document named for the chain (ROBINHOOD.md, docs/robinhood-chain.md) count as doc-mention.

Two differences are deliberate. Every other document (a changelog, a guide, a report) still counts as doc-mention: at 5 points it can never raise a tree above LOW or anchor a HIGH, so it stays visible as the weakest marker instead of disappearing. And chainprint does not treat every file named *registry* as a chain list: a project's own contract registry is often where it records its deployments.

What it does NOT prove

A Robinhood Chain marker shows that a codebase targets the chain. It does not prove who owns the code or any contract, that the repository is a project, or that a project is legitimate. Confidence measures the markers found, not the project.

HEY Research Lab is an independent research project and is not affiliated with, endorsed by or partnered with Robinhood Markets, Inc. or Robinhood Chain.

Security

See SECURITY.md. In short, chainprint:

• makes no network connection of any kind and has no networking code;
• never executes anything from the scanned tree — config files are read as text, never required or imported;
• never follows symbolic links, inside or outside the root, and confines every path under the root;
• never enters .git, node_modules, vendor, vendored, third_party, build, dist, out, artifacts, cache, coverage, target, typechain(-types), framework caches, virtualenvs, or lib/ beside foundry.toml;
• never opens real .env files (only templates such as .env.example), lockfiles, minified files or binaries (a NUL byte in the first 8 KB);
• reads at most 10,000 files, 1 MiB per file, 256 MiB in total, 12 directory levels deep, and opens only regular files (no FIFOs or devices);
• parses JSON with prototype keys (__proto__, constructor, prototype) dropped;
• prints only the matched line, redacted (URL credentials and query strings, long tokens, 64+ hex digit strings) and cut to 120 characters, with control and bidi characters removed. Env template values are never printed unless the value is the public marker itself;
• writes nothing.

Contributing

See CONTRIBUTING.md. pnpm install, then pnpm lint, pnpm typecheck, pnpm test, pnpm build and pnpm scan (leak and attribution scan). Tests run offline against the fixture repositories in test/fixtures/; network access is blocked in the test runner. A new rule, weight or known address needs a public citation, a fixture and a test, and bumps the rules version.

Licence

MIT © 2026 HEY Research Lab