Merkle Science

The models are the differentiator and the company says so directly, describing a proprietary engine that goes beyond blacklists to score risk in real time from behavioural patterns rather than from membership of a known bad list. Machine learning tracks suspicious wallets by how they behave, graph network analysis deanonymises identities, and the data mining treats the blockchain as a very large public dataset alongside scraped internet material.

Strip the models and what remains is a list of known illicit addresses, which is precisely the product this company positions itself against, so the removal test empties the proposition entirely. That is the exact inverse of Chainalysis, whose attribution database survives the same test.

Screening runs automatically before a business relationship is established and repeats on a regular cycle over existing customers, with real time risk scores produced without human involvement. Meaningful control exists at design time through the behavioural rule engine, which lets an institution define what patterns it treats as risky rather than accepting a fixed model, and that is more configurability than a pure blacklist product offers.

Nothing exists at decision time: no confidence indication accompanies a score, no review threshold is described before a flag reaches a compliance queue, and no account is given of how an analyst distinguishes a behavioural inference from a confirmed identification.

No accuracy, precision or false positive rate is published for the behavioural engine, and the gap weighs more heavily here than for a list based competitor. A blacklist product can be audited by inspecting the list; a predictive engine cannot be evaluated at all without knowing how often its predictions are right, and prediction quality is the entire claim being made.

Nothing describes how behavioural models are validated, how they are monitored as laundering techniques evolve, or how a user distinguishes a high confidence identification from a speculative pattern match.

One customer is named, the largest licensed crypto asset exchange in Indonesia, which selected the platform to strengthen its anti money laundering and counter terrorist financing processes and its reporting to regulators. Beyond that there is no customer count, transaction volume, coverage figure or measured outcome.

Around 25.6 million dollars has been raised across three rounds from 37 investors including a major crypto holding company, a Singapore government backed fund and a quantitative trading affiliate, and leadership is drawn from payments, exchange, banking and federal law enforcement backgrounds. The most recent round closed in August 2022, so the funding record is nearly four years stale, though current independent reviews and profiles confirm the company is actively trading.

No data boundary statement was located. The company maintains what it describes as a vast in house database of crypto crime records, and behavioural models improve with exposure to more illicit activity, so intelligence accumulated through one exchange's monitoring necessarily strengthens the product sold to its competitors.

That may be defensible for a financial crime utility, and nothing states whether customer investigation activity contributes to the shared database, whether a customer can decline, or how enquiry data is separated between institutions that compete directly in the same markets.

No retention schedule, subprocessor list or data processing terms were located, and one disclosed input makes the position more exposed than for peers working from ledgers alone. The data mining approach is described as combining blockchain analysis with internet scraped data, which means material gathered from outside the ledger about the people behind addresses, and that is a materially more intrusive collection than observing public transactions. Forensic deanonymisation through graph analysis is the stated purpose. Nothing describes what scraped sources are used, how the resulting identifications are stored, or how a wrong one is corrected.

No attestation, certification, trust centre or enumerated framework was located, despite the company being classified within the cybersecurity sector and selling to banks, licensed exchanges and government agencies, all of which run vendor security assessments as a condition of onboarding. Those assessments have plainly been passed in private and none of the resulting assurance is published, so a new institutional buyer begins its own review from nothing.

Regulatory engagement is evidenced through a named industry standard rather than asserted. The company is an associate sponsor of a code of practice for digital asset firms that was facilitated by the national monetary authority and developed in consultation with the country's banking association, which is participation in a regulator convened standard setting process.

Screening explicitly names the principal sanctions list alongside other law enforcement lists, so the instrument is identified rather than implied, and the virtual asset service provider framework shapes the onboarding product. Further alliances span digital finance industry bodies in Europe and Australia, and educating regulators and law enforcement is a stated part of the mission.

The two exposures common to this lane apply, attribution error tainting everyone downstream of a mislabelled address and risk propagating to parties several hops from a flagged source, and a third is specific to this vendor and follows directly from its own differentiator.

Predicting risk behaviourally rather than matching against a list means flagging addresses that have done nothing identifiably wrong, because the model infers danger from patterns rather than asserting a fact about a known entity, and an inference about an innocent party is a materially different kind of error from a list match. The property that makes the platform better at catching novel threats is the same property that makes it likelier to flag someone who has done nothing. No false positive rate, confidence measure or appeal mechanism was located.

No guarantee, indemnity or falsifiable accuracy commitment was located, and no correction or appeal process is described. The subject's position is worse here than under a list based system for the reason the product exists: a person flagged by behavioural prediction may never have transacted with anything illicit at all, so they cannot even trace the finding to an identifiable bad counterparty, and they are not told the flag exists. The institution is better served, receiving scores and forensic evidence it can act on, but nothing states what the vendor owes when a prediction proves unfounded after an account has been closed.

The data chain is enumerated by category more openly than most in this lane. Screening draws on the principal sanctions list, other law enforcement sanction lists, general money laundering and terrorist financing information sources and the company's own crypto crime database, and the analytical layer is described as combining public ledger data with scraped internet material. Disclosing the scraped component is notable, since it is the input most vendors would leave unmentioned. What is not named is any individual provider behind those categories, nor any model provider for the behavioural engine, and no subprocessor list or hosting arrangement was located.

Cross chain tracing and real time screening imply interface based consumption, and offices across three continents are presented as supporting round the clock coverage for international customers, but no integration is evidenced. No case management system, exchange platform, core banking system or transaction monitoring suite is named on either side, no blockchain networks are enumerated, and no developer documentation or interface reference was located.

No hosting provider, region selection, residency commitment or private deployment option was located. The question has weight given a footprint spanning Singapore, Japan, Korea, India, the United Kingdom and the United States, several of which impose their own requirements on where financial and investigation data may be processed, and government agency customers in particular operate under handling rules that a published residency position would normally address.

No pricing, packaging or basis of charge was located and the only route in is a booked demonstration with blockchain intelligence specialists. Nothing indicates whether charge falls per address screened, per transaction monitored, per investigation seat or as an enterprise platform fee, and for a buyer set ranging from a regional exchange to a government agency the range is likely to be wide.

Four buyer types are addressed, spanning crypto asset businesses and virtual asset service providers, banks and financial institutions entering digital assets, government agencies and law enforcement, and decentralised finance participants.

The genuinely distinctive dimension is geographic: founded in Singapore with offices there and in Tokyo, Seoul, Bengaluru, London and New York, the company covers Asian markets that the Western incumbents serve less closely, and its named customer and industry body relationships are concentrated there. Independent assessment places it as best fitted to regulated crypto entities wanting on chain compliance depth rather than general purpose security tooling.