Skip to main content
The methodology is the registry of every heuristic and rule Blockgence uses to assert something. For each one it states:
  • what it asserts and what it is based on;
  • its linkage: deterministic, probabilistic, derived, mixed, heuristic or filter;
  • its confidence, its parameters and its known limits;
  • the status of its error rate.
It is versioned, and every document we issue says which version it was built with.

Version and fingerprint

Version

YYYY.MM.DD plus a letter, for example 2026.11.17w. It goes up when a rule that changes results changes. The letter orders the changes made on the same day.

Fingerprint

A hash of the registry’s actual parameters, for example 576c6f78dc064d0c. It changes even when the version does not go up: a registry that grew, a threshold or a count measured again.

Where it is cited

  • In every sealed document, in manifest.methodology: version, fingerprint and entries_total. The evidence package also carries the entries that analysis used and the global limitations, which head every package.
  • In the API: GET /v1/methodology returns the whole current registry, with your API key and at 0 credits. GET /v1/methodology/{entry_id} returns a single entry.
With summary=true it returns only each entry’s index (the full list weighs about 0.5 MB). With family= it filters by family. The numbers in this example are illustrative, and the registry’s text fields are in Spanish.
An entry whose error rate has not been measured says so: error_rate.status is not_measured, with its note. Once measured, it carries the count and its confidence interval.

What changed

From version 2026.09.29g (86 entries, the one the methodology was published with) to the current one when this page was written, 2026.11.17w (167 entries): 81 new entries, 86 changed and none removed. Each entry’s detail is in GET /v1/methodology.
A month-by-month summary of the changes that move results the most:
2026.11.09a → 2026.11.17w
  • Reproducible evidence. The package pins the block each network was read at and records the source responses. What the analyst had on screen in a case is also recorded. All three can be re-run with no network.
  • Measured error rate, with its count and confidence interval, on the entries where there is something to measure it against.
  • Unsigned, unpriced outflows. A movement the address did not sign and that has no price is declared separately: it may come from a spoofed sender.
  • Pass-through infrastructure. Operational wallets, position managers and protocol periphery are named without cutting the trace, unless the transaction does not prove the hop.
  • Smart accounts by method. Safes and ERC-1967 accounts are recognized by their code and storage, not by a list.
  • New bridges: LayerZero (V1 and V2), deBridge, Chainflip and Relay. Also crossings matched by on-chain identifier, and crossings that never reached their destination.
  • New products with their methodology: contract inspector and simulation, token profile, multi-victim scam, Guard rules, attack vectors and laundering pulse of an incident, and clusters by common deposit and by Safe co-signer.
  • Case annotations in the evidence, and the alternatives the analysis considered and ruled out.
2026.10.05a → 2026.10.31o
  • Tron traceable, with live public labels and exchange wallets from their proof of reserves.
  • Monero as a known network. The XMR addresses on the OFAC SDN list come in, validated, along with the payment proof of a transaction, sealed. A possible exit to Monero through no-KYC exchangers is declared.
  • Honest screening. An address that no source identifies no longer comes out as clean.
  • Warnings that tell the truth. “The source failed” only when the source failed. In OSINT, only a result that shows the address counts as a mention, and a blocked engine is not read as “no mentions”.
  • EIP-7702. The native coin swept by a delegation counts as an outflow, with a registry of known delegates.
  • Crossings. The canonical bridges of Arbitrum and the OP Stack (Base, OP Mainnet) are recognized as a crossing between networks.
  • Tornado Cash outside Ethereum is identified and cut.
  • Incidents. Live block monitoring, victim triage, the incident service deliverable and a post-mortem that says where coverage ends. Blind spots of the trace are declared.
2026.09.29g
The version the methodology was published with, on 3 Oct 2026: 86 entries and 7 global limitations.