What XenonProbe reads, and where it stops
XenonProbe is a forensic reader, not an oracle. It answers one question — what does the chain record about this address — and it is explicit about every question it cannot answer from that record. This page describes exactly how it works, so that any claim it makes can be checked.
The product principle
Two sentences that decide every design choice in this application.
What it is not
“AI tells you whether a token is safe.”
There is no safety score, no grade, no traffic light, and no verdict anywhere in this product. A score would compress evidence and guesswork into one number and hide which was which.
What it is
“XenonProbe reads observable blockchain activity and explains what the evidence shows.”
Facts, the transactions that prove them, bounded interpretation clearly labelled as interpretation, and a permanent section for everything the chain does not contain.
Where the data comes from
Two upstream sources, both public, both named on every report.
- JSON-RPC node
Live state, read at request time: eth_getCode, eth_getBalance, eth_getStorageAt against the EIP-1967, EIP-1822 and legacy proxy slots, and a batch of read-only eth_calls.
This is the authoritative source for anything happening now: current owner, current implementation, current pause state, current supply.
- Block explorer indexer
History the node cannot serve: transaction lists, event logs, token transfers, holder rankings, creation records, and verified source where it has been published.
Public indexers rate-limit and occasionally fail. Every read retries briefly, then gives up and marks its field DATA UNAVAILABLE with the reason. It is never replaced with a zero.
Network
Robinhood Chain · chain ID 4663
Explorer links on every report are generated from the configured explorer origin, so the transaction you open is the transaction the claim was built from. robinhoodchain.blockscout.com ↗
How capabilities are detected
Without a verified ABI, most tools give up. XenonProbe reads the bytecode instead — and tells you that is what it did.
A Solidity contract dispatches calls through a jump table built from PUSH4 instructions, one per function selector it accepts. Scanning the deployed runtime for that opcode recovers the callable surface of a contract whose source was never published.
The recovered selectors are matched against the catalogue below. Where the contract is a proxy, the implementation’s bytecode is fetched and scanned too — a proxy’s own runtime exposes almost nothing, and reporting it as “no capabilities” would be misleading.
The scan is a heuristic and is labelled as one. A selector found only in bytecode is reported at High confidence confidence; one confirmed against a published ABI is reported at VERIFIED. Where a function is safely readable, a live call confirms the current value.
Detecting a mint function proves the function exists. It does not prove who may call it, whether a modifier blocks it, or whether it has ever been used. Those are separate questions, and the report treats them separately.
Control signatures this build looks for · 67
Events this build decodes · 24
An event outside this list is still shown — reported by its topic hash, and explicitly described as one this build cannot name. It is never labelled with a guess.
The confidence scale
Four levels, deliberately coarse. A finer scale would imply a precision the underlying data does not support.
Directly observable in blockchain data.
Strongly supported by multiple independent on-chain signals.
A reasonable reading of the data, not directly proven.
Insufficient data to answer.
There is no numeric confidence anywhere in the product. “87% confident” would be a number with no derivation behind it, and inventing one is exactly the failure this tool exists to avoid.
How signals are written
Every signal is four fields, and it is not allowed to skip any of them.
- 01
Observation
What was seen, stated flatly, with no adjectives.
- 02
Evidence
The transaction, call, or record it came from — clickable.
- 03
Interpretation
What it plausibly means, bounded and hedged.
- 04
Limitation
What this specific signal still does not establish.
Signals are never labelled “scam”, “rug”, or “safe”. A contract with a mint function and an EOA owner is described exactly that way, because thousands of legitimate contracts look identical and the difference between them is not on chain.
Where the language model sits
Late, optional, and strictly downstream of the facts.
- 1
Blockchain data
RPC node and indexer
- 2
Structured facts
Typed values with confidence and source
- 3
Analysis engine
Deterministic — writes the report
- 4
Explanation layer
Optional, restates the fact sheet
- 5
You
With links to verify every claim
The report you read is written by the engine, deterministically, from typed values. No model output is required to render any part of it.
The explanation layer is a separate button. When you press it, a model receives the same structured fact sheet the report was built from — never the chain, never a search index — and is instructed to restate it. It cannot introduce a number, address, balance, or hash that is not already on the page.
If no API key is configured, the panel says so plainly. It does not fall back to a model-free imitation of model output.
What XenonProbe will never tell you
This list does not shorten as the product improves. It is a property of the chain, not of the reader.
The chain gives us evidence, not omniscience
The real-world identity of a developer or holder
Addresses are not people, and this product will not guess at the mapping.
Anyone’s intentions
Motive is never written to a block.
Off-chain agreements
Vesting promises, custody arrangements, and legal ownership leave no trace.
Whether a team is trustworthy
There is no on-chain measurement of honesty or competence.
Future behaviour
Capabilities describe what the code permits, not what will happen.
Whether a project will succeed
Outside the scope of any chain reader.
Whether funds will be misused
A transfer and a theft are byte-identical on chain.
Whether two addresses are one party
Only knowable when someone has proven it publicly.
Private communications
Not on chain, and not inferable from it.