Hire on evidence, not vibes.

Books vs bank for a landscaping company: two charges missing from the ledger, a cents transposition, a duplicated entry, and an outstanding check that ISN'T an error — plus a vague tip about an old double-charge that rewards whoever actually checks.

Twelve departments, one screaming variance that's actually an annual prepayment hitting a single month — and one quietly misfiled software charge that only turns up if a vague tip gets taken seriously.

The literal can-they-use-a-computer trial: follow a written naming SOP, work out correct filenames for a messy shared drive, and keep a log — with one file whose dates quietly disagree and one instruction that only arrives mid-task.

Three refund tickets that each hit a policy edge — the wrong-clock return window, a final-sale exception, an angry escalation — plus a contradiction planted inside the policy document itself that nobody assigns anyone to find.

Transcribe a messy printed packing list into a clean CSV. One line is secretly printed twice — catching it separates careful from fast, and a vague heads-up mid-task tests whether they act on loose information.

Cover a small distributor's inbox for 30 minutes: real customer questions, a supplier email that quietly contradicts the purchase order, one angry customer, and an unowned problem nobody asked them to handle.

One send, five segments — and the biggest list is a trap. Expected-click math, style-guide discipline on three bad subject lines, and a hedged tip about a broken unsubscribe link that a careful marketer treats as a compliance stop.

Messy meeting notes into one investor slide. A membership number is corrected in a buried aside, a key figure only arrives mid-task, and a recurring complaint nobody owns tests judgment about what does NOT belong in front of investors.

Three vendor one-pagers, one recommendation. The advertised-cheapest option hides a setup fee in its fine print, and a secondhand outage rumor tests whether they report like an analyst or a repeater.

A week's shift board with two double-bookings, one quiet 42-hour week, and an understaffed close — plus two swap requests where one quietly violates an availability rule. Detection, judgment, and a clean fix.

A messy 200-row inventory sheet with seven planted errors and one unasked anomaly. Candidates clean it in their own Excel/Sheets, answer three questions, and show whether they notice what nobody pointed at.

Four tables, three business questions, sixty minutes. A status column quietly punishes blind aggregation, and a vague FYI about a years-old form bug tests who actually checks.