Legitimate first-time buyer moving BTC to their own hardware wallet. Near-identical on paper to S1 (new account, large amount, fast outbound) — the discriminator is behavioral: QR entry, zero idle gaps, no tab switching, fluent short session.

A business account paying its regular USDT supplier — same amount range, same destination, every month for over a year. Looks like a large fast outbound on paper, but it's routine recurring business activity.

A stable, recurring monthly remittance to the same family member's wallet for 8 months — a known destination, a known pattern, nothing anomalous.

Near-identical to S1 on paper — brand-new account, 9,000 EUR, card, first purchase. The discriminator: zero idle gaps, a manually typed address, long deliberate time on detail screens, destination is a known exchange. THE critical false-positive test: if this comes back INTERVENE, the prompt is too aggressive. Run this fixture after every prompt change. Verdict is PASS, not SOFT — its signals are exactly as clean as L1's (zero coercion signals, if anything more deliberate), so forcing friction here would also have to catch L1, which must stay frictionless.

A routine purchase from a small, properly onboarded online retailer's checkout — same surface shape as an unfamiliar small seller, but the merchant is a verified Mercuryo integration with an established settlement flow, not a self-generated link.

7 weeks of daily conversation with someone met on a dating app, who steered her toward an 'investment platform' showing 22% profit. A small successful withdrawal was the bait. Secrecy coaching ('don't tell family', 'say you're buying for yourself') is baked into the script.

A pop-up claims the account is compromised, followed by a live phone call from a fake 'support agent' who stays on the line while the victim moves funds to a 'secure vault' address and is told not to use the in-app chat.

A Telegram 'product review job' pays real commission on the first two small tasks, then a 'premium' task demands a deposit to unlock accumulated commission, under a 40-minute deadline.

A caller claiming to be police says the victim's identity was used in a money-laundering case and that funds must be moved into a government 'custody account' — framed as an active investigation the victim cannot discuss with anyone.

Contacted by a fake 'asset recovery firm' claiming to have traced 18,400 EUR lost to a scam three months earlier; charges an upfront 'processing fee' under a 72-hour deadline. She was already a victim once, and this is the same playbook wearing a rescuer's mask.

A voice message that sounds exactly like her son, followed by texts from an unfamiliar number: an accident, urgent money needed, can't talk right now, don't tell dad yet. Breaks a two-year stable pattern of small occasional transactions.

A medium-experience trader in a Telegram 'signals' group, shown profit screenshots and pressured by FOMO and a 2-hour entry window. An informed but risky decision, made alone, with no real-time coaching — the calibration test for SOFT vs INTERVENE.

A 'payment processing agent' job: receives funds from multiple different senders, immediately converts to crypto, and forwards it, keeping a 5% margin. The user isn't losing money — they're unknowingly committing a crime. Needs a different message than the other fixtures: explain the legal exposure, not the financial risk.

Adversarial test: the probe answers read exactly like a confident, self-directed buyer (similar tone to L4) — deliberate research, address copied from the official page, prior successful use, explicit denial of any outside influence. But the checkout session itself shows the single strongest coordination-signal combination in the system (idle_gaps_over_60s + tab_switches + pasted address), suggesting the denial may itself be coached — whoever is directing someone in real time will often also prepare them for questions like these. Tests whether the pattern screen notices the contradiction between the story and the session instead of taking a clean denial at face value.

A 'seller' in a local marketplace group offers a heavily discounted graphics card and insists on payment through a self-generated Mercuryo-branded checkout link instead of the marketplace's own payment flow, claiming it's safer. The link looks like a normal widget, but the seller was never onboarded as a registered Mercuryo merchant — the settlement wallet was created specifically for this sale.