Machine speed
Alerts calibrated to human order cadence miss sub-second bursts and reversals that agents execute thousands of times an hour.
MiCA Article 92 · Agentic market-abuse surveillance
Prove your surveillance catches machine-speed abuse — with labeled ground truth.
TapeProof pairs a labeled corpus of agentic payment and trading abuse — distilled from the public x402 tape of 165M+ agent transactions, including the documented December 2025 wash-trading boom and 92% bust — with a replay engagement that injects those abuse typologies into your surveillance sandbox and hands you a calibration-gap report your examiner can read.
- 165M+
- agent transactions in the source tape
- 5
- agentic abuse typologies, labeled
- Art. 92
- MiCA surveillance obligation anchored
The problem
Surveillance tuned for human traders can't evidence it catches agents.
Sub-threshold by design
Agents size orders just under per-account limits and structure one economic order into hundreds of children. Every child looks clean.
Many wallets, one controller
Fan-out across dozens of freshly funded wallets defeats per-account rules while the cluster round-trips inventory.
And there is no labeled ground truth. You cannot tune, backtest, or evidence a detector against abuse you have never seen labeled. Examiners increasingly ask CASPs to demonstrate detection — not just assert a policy exists.
The product
Corpus, replay, and a report built for the exam room.
01
Labeled corpus
Agentic abuse typologies extracted and labeled from the public x402 tape, delivered as structured, versioned datasets your data scientists can load directly.
02
Replay engagement
We inject the labeled typologies into your trade-surveillance sandbox at controlled volumes and timings, then measure exactly what your alerts fire on.
03
Calibration-gap report
A regulator-exam-ready readout: per-typology detection rate, threshold gaps, false-negative evidence, and a remediation punch list mapped to MiCA obligations.
Regulatory anchor
MiCA Article 92 made testable.
Under MiCA Article 92, crypto-asset service providers must maintain effective systems and procedures to prevent and detect market abuse. ESMA's supervisory guidelines expect those systems to be calibrated and evidenced, not merely documented. With grandfathering closed since the 1 July 2026 cliff, enforcement is live across the ~283 registered CASPs.
TapeProof turns "we have a surveillance system" into "here is our measured detection rate against labeled agentic abuse, and here is our remediation plan" — the difference between a policy binder and an examinable control.
- Art. 92(1) — systems to detect and report market abuse.
- ESMA guidelines — calibration and effectiveness testing expectations.
- STOR readiness — evidence trails for suspicious transaction reporting.
- Board assurance — a metric leadership can attest to.
Live demo
Typology Explorer
Browse five agentic abuse typologies. Each ships a small synthetic labeled sample, a
visualization of the pattern, and a downloadable JSON corpus served from
/api/corpus/:typology.
—
——
Pattern visualization
What an examiner looks for
All Typology Explorer data is synthetic and generated deterministically for illustration. It is not extracted from any real venue.
Calibration self-check
Five questions your examiner might ask.
Answer yes or no. Runs entirely in your browser — nothing is sent anywhere.
—
Pricing
Scoped engagements, fixed deliverables.
Focused replay
$25K
- Up to 2 abuse typologies
- One surveillance sandbox
- Calibration-gap report
- Remediation punch list
Most common
Full exam-readiness
$40K–60K
- All 5 agentic typologies
- Multi-venue / multi-pair replay
- Exam-ready report + evidence pack
- Threshold tuning workshop
- Re-test after remediation
Corpus license
Talk to us
- For surveillance vendors
- Labeled corpus, versioned
- Redistribution terms
- Update cadence
Engagement process
Four steps to an examinable control.
-
1
Scope & access
We agree typologies, venues, and pairs, and connect to your surveillance sandbox — read-only, isolated from production.
-
2
Inject & replay
Labeled abuse is replayed at controlled volume and timing. We record every alert your system does and does not fire.
-
3
Measure the gap
Per-typology detection rate, threshold analysis, and false-negative evidence, mapped to MiCA Article 92 obligations.
-
4
Report & re-test
You get an exam-ready report and remediation list. Optional re-test confirms the fixes closed the gap.
Get started
Request a replay engagement.
Tell us your venue and which typologies concern you most. We reply with a scoped proposal and timeline, typically within two business days.
Email ebo@testmachine.aiOr write to ebo@testmachine.ai directly.