How the BlackOps Market trust score is calculated, and what it cannot tell you

Design principles

Before the mechanics, the four commitments that produce them. These are the reasons a number here should be treated differently from a number on a directory page.

1. Everything is derived, not typed. The published score is computed from published dimension values and published weights by the same code that renders the page. There is no spreadsheet step, no manual override and no marketing override.
2. Unverifiable claims earn no credit. A dimension value is an assessment of verified evidence. A market's own marketing does not raise its score, and where marketing is the only evidence the claim is listed as unverified and the dimension is scored accordingly.
3. Uncertainty is published as an interval. Where confidence is medium we report a range, and two scores with overlapping ranges are reported as a tie. We would rather publish an uninteresting result than manufacture a ranking.
4. The subject does not see the assessment. No market, vendor or directory is contacted for comment before publication, and none pays us anything. A pre-publication review right is a veto, and a veto is incompatible with a critical assessment.

The eight dimensions

Each dimension is scored from 0 to 10 as an integer, carries a fixed weight, and the score is the weighted mean scaled to 0–100. Weights sum to exactly 100 and are frozen for the duration of an edition; changing them mid-edition would make scores incomparable across dates.

Why these weights

The weights encode a single thesis: the operator's ability to reach your money dominates, and the surrounding harms are what actually land on people. Custody is the largest single dimension because every documented catastrophic loss in this ecosystem since 2014 has been an escrow event, and because the escrow claim is the one architectural property a buyer cannot verify from outside. Brand integrity is weighted at 12% — unusually high for a platform assessment — because the observed harm rate in this ecosystem is dominated by phishing clones rather than by platform failures, and a market's name is a direct cause of that harm.

Survival is deliberately not weighted heavily. A four-year record is evidence of a working business, not of a working design, and weighting it heavily would mean that every incumbent scores well by default. A market that is new, well-designed and honest should be able to out-score a four-year-old one that is well-designed and dishonest, and this model allows exactly that.

Evidence hierarchy

Every claim we assess is placed in one of five tiers. The tier determines the label on the claim table and, indirectly, how much the claim can move a dimension value.

One rule deserves emphasis because it is the most common failure in this market of analysis: corroboration is not independence. Forty sites repeating one operator statement is one source, not forty. When we see a claim propagated across the directory layer, we identify the origin and score the claim on the origin's tier, not on the number of copies.

Confidence and intervals

Each market carries a confidence level, derived from the mix of evidence available about it — not from how confident we feel about the platform.

• High — the market's status and material attributes are established by at least two independent T1/T2 sources. A ±3 interval applies.
• Medium — status is established, but material attributes rely substantially on operator self-report or a single independent source. A ±8 interval applies.
• Low — the evidence base is thin or contested. A ±14 interval applies, and the score should be treated as provisional.

When two intervals overlap, the result is reported as a tie. Torzon (67.2, interval 59.2–75.2) and BlackOps (66.6, interval 58.6–74.6) overlap almost entirely. We therefore report them as statistically indistinguishable, and we say so in the headline rather than presenting a 0.6-point difference as a ranking. The comparison engine's re-weighting controls exist so that readers who weight the dimensions differently can see whether the result is robust or arbitrary. Under our published weights it is not robust: it is a tie that happens to be close to the middle of the range.

Score bands

What the score is not

• It is not a probability of failure. A score of 67 does not mean a 33% chance of collapse. We do not have the base rate, the exposure data or the resolution time to produce a calibrated probability, and we decline to publish a number that looks like one.
• It is not a recommendation to transact. There is no score at which we would say a person should buy prohibited goods from a marketplace. The score describes a platform's risk shape, and the risks it describes are reasons to reduce exposure, not reasons to proceed.
• It is not a measure of the people involved. It describes architecture and evidence quality. Most participants in this ecosystem are not the adversary, and the model deliberately contains no dimension that would characterise them.
• It is not current to the second. Market status changes within hours. Every score carries a date, and the date is displayed next to it, always.
• It is not a substitute for legal advice. Nothing we publish changes what is lawful where you are.

Refresh cadence and event triggers

• Fortnightly scheduled re-audit of every active market, with dimension values and weights re-checked against new evidence.
• Event-driven re-publication within 24 hours of a shutdown, seizure, exit-scam allegation, major architecture change, or a confirmed phishing campaign against an audited brand. The Nexus case was published as an event-driven update on 6 October 2026, three days after the first report.
• Quarterly review of the model itself: whether the dimensions, weights, tiers and intervals still describe the ecosystem. Model changes are versioned and never applied retroactively to a previous edition without a note.

Known limitations

We would rather publish these than let a reader discover them.

• We do not verify claims by testing. We do not create accounts, and we do not interact with these platforms. Every architectural claim about a market is therefore assessed from documentation and reporting, not from observation. We regard this as a non-negotiable editorial line, and it is also the largest source of error in the model.
• Operator self-report is unfalsifiable from outside. A market that lies about its escrow configuration looks identical to one that does not. The Nexus case demonstrates the failure mode of any assessment based on published claims.
• Scores are not comparable across editions when weights change. Each score is labelled with its model version.
• Selection is not random. We audit markets that are prominent enough to find and that enough people care about to be worth auditing. Markets we have never heard of are unassessed, not safe.
• Absence of evidence is not evidence of absence. A quiet period is not stability, and a market that is not on our list is not necessarily defunct.
• Our sources have interests. Blockchain-intelligence firms sell the product of this analysis. Their methodology is published and their findings are usually rigorous, but the incentives are real and we note them.

Corrections policy

Corrections are published, dated, and permanent. We do not silently edit a published score, and we do not delete the original claim.

1. A correction notice states what was claimed, what is now known, when the change was made, and which scores it affects.
2. Where a correction changes a dimension value, the score is recomputed and the previous score is shown alongside it, with the model version of each.
3. Where a correction changes a market's status, the status change is published as an event-driven update, in the same format as any other status change.
4. Material corrections are also recorded in the changelog below and pushed to the RSS feed.

To submit a correction, write to the desk with the specific claim, the specific evidence, and the source of that evidence. We will publish it if we can verify it, and we will say so if we cannot. We do not publish anonymous corrections and we do not publish corrections that merely restate a market's marketing.

Conflict of interest

We do not accept payment, advertising, sponsorship, affiliate revenue, referral fees, hardware or travel from any marketplace, vendor, directory, forum, blockchain-analytics vendor or law-enforcement body. We do not operate an access directory, we do not broker, and we do not hold funds. Where an author has a professional, financial or personal interest in a subject, it is disclosed on the report page and the affected dimension values are re-reviewed by the other author. Any actual or potential conflict discovered after publication is disclosed in the changelog.

Changelog

Bibliography

Every source behind the current edition, with its tier. The same list is printed at the foot of each report.

Open the comparison engine →