What to check instead of 'top 1%' talent claims

Every agency in this category says 'rigorous vetting' and then shows you nothing. 'Filters out 99.7%', 'top 1%', 'hand-selected' — these are numbers you cannot check, drawn from a denominator nobody will define. Top 1% of what applicant pool? Screened by whom, on what criteria, weighted how? The claim is designed to end your questions, not answer them.

That's a problem, because the screening step is the only part of the process that actually predicts whether someone will still be working out in month four. A recruiter who won't show you the rubric is asking you to buy a black box. Here's what to ask for instead — from any recruiter, marketplace, or screening tool you're evaluating.

Ask for the criteria in writing, with weights

A real screening process has components and each component has a weight. If a recruiter can't hand you a one-page breakdown of what they're measuring and how much each piece counts, the 'vetting' is a judgment call — which means it's as consistent as whoever happened to run the call that day.

What a weighted rubric looks like in practice: graded work sample, scenario-based judgment questions, specificity of experience (not just years, but named tools and outcomes), written English quality, and salary-band alignment. Each of those should have a number attached. If you ask and get a paragraph of adjectives instead of percentages, that's your answer.

Demand a graded work sample, not a resume screen

Resumes are backward-looking and easy to inflate. A graded work sample — a real task at roughly the difficulty of the actual job, scored against a fixed rubric, tells you what a candidate can do right now. For a bookkeeper, that might be a categorization exercise on a messy transaction set. For a customer support specialist, a written response to a hostile ticket. For an SEO specialist, a brief audit of a page with obvious problems.

The grading matters as much as the task. A work sample scored on a pass/fail basis by whoever happens to review it is only marginally better than no work sample. You want a rubric with defined criteria, applied the same way to every candidate, so you can compare scores across finalists rather than impressions.

If an agency tells you they do work samples but won't show you a sample of the sample, the actual task and the scoring guide, ask why. A legitimate process has nothing to hide there.

Check whether scenario judgment is tested separately

Skills and judgment are different things. A bookkeeper who can categorize transactions correctly may still make a poor call about whether to flag an anomaly to you or just fix it quietly. A customer support specialist who writes well may handle a straightforward ticket fine and a complaint about shipping delays badly.

Scenario questions, 'here's a situation, what do you do?', are the standard way to separate those. They should be graded, not just collected. If a recruiter describes their scenario questions as 'a conversation' or 'a chat to see how they think', that's not a scored input; it's an interview, and interviews have well-documented consistency problems.

Ask what happens if nobody clears your bar

This question surfaces the recruiter's actual confidence in their process faster than almost any other. A process that's genuinely well-calibrated produces finalists who meet the brief most of the time. A recruiter who knows that will offer a free re-run if no one clears your bar, because they expect it to rarely happen.

A recruiter who hedges, says 're-runs are available for a fee', or pivots to convincing you that the finalists are actually good enough, that's a different signal. The re-run policy isn't just about fairness; it tells you how much the recruiter is willing to bet on their own screening.

Look at whether the salary band is part of the screen

Presenting a finalist who is clearly out of your budget range wastes everyone's time and suggests the recruiter isn't reading your brief carefully. Salary-band fit should be an explicit screening criterion, small weight, but nonzero.

For reference: an executive assistant in the Philippines runs $800-1,400 a month at full-time rates; in Colombia (closer time zone, US Central year-round), that band scales by about 1.25. A bookkeeper in the Philippines is $800-1,500 a month. These are published ranges, not guesses. If a recruiter is presenting candidates without disclosing where they sit in the range, ask before you meet anyone.

The English-quality piece is often the most skipped

For roles with significant async written communication, executive assistants, content marketers, customer support, project managers, written English quality is a real predictor of day-to-day friction. A candidate who writes clearly in a screened scenario is almost certainly going to write clearly in Slack. A candidate who doesn't will cost you editing time you weren't expecting to spend.

Ask whether English is scored in the screening and how. 'We only consider fluent English speakers' is not a score. A rubric that treats grammar, clarity, and tone as distinct sub-criteria and applies them to actual written output from the screening process, that's a score.

One tool that makes its rubric public

Rolemote publishes its screening rubric on its how-we-screen page: graded work sample 35%, scenario judgment 25%, experience specificity 20%, written English 15%, salary-band fit 5%. The weights are fixed and visible before you start. The whole search, describe the role, get an AI hiring brief, read your scored finalists, runs free. You pay a flat fee only when you ask to meet finalists, not a percentage of salary. Typical offshore agencies in this category charge 25-35% of first-year salary, or about $4,500-6,300 on common roles; Rolemote's flat fee is $1,995. If no finalist clears your bar, a re-run is free.

The point isn't to use Rolemote specifically, it's that a published, weighted rubric is a checkable thing. Any recruiter or tool you evaluate should be able to hand you something equivalent. If they can't, the 'top 1%' claim is doing the work that the process should be doing.