How to Verify a Remote Candidate's Skills Before Hiring

The short answer

Verifying a remote candidate's skills means measuring them under controlled conditions instead of trusting what their CV, portfolio or references claim. In practice that is one proctored online assessment sitting — working English (written and spoken), role-relevant ability, and workplace judgement — taken through a link on any phone or laptop, with an identity-baseline photo captured at the start and the session monitored throughout. The result is an evidence report per candidate: scores, an integrity rating, the proctoring evidence itself, and a ranked position against everyone else you screened.

This matters most in offshore hiring corridors — an employer in the US, UK, EU or Australia hiring in Nigeria, Kenya, South Africa, the Philippines or India — because those are exactly the hires where you will never meet the person, never see them work, and cannot rely on checking references across a border.

Why CVs, portfolios and references fail across borders

A CV is a claim, not evidence, and the further away the candidate is, the harder the claim is to check. Portfolios can be borrowed or bought. References can be arranged. Degree certificates from institutions you have never heard of are effectively unverifiable at screening speed. None of this means offshore candidates are less honest than local ones — it means the informal verification you do without noticing in a local hire (a shared contact, a recognisable employer, an in-person interview) simply is not available.

The fix is not more documents; it is a measurement. When every candidate takes the same assessment under the same monitored conditions, you are comparing what people did, not what they wrote about themselves — and the strongest candidates benefit most, because a measured result finally lets them out-compete a better-written CV.

What to verify: the three layers that predict remote success

Working English comes first, because for a remote hire English is not a nice-to-have — it is the medium the entire job happens in. Verify it in both directions: written accuracy and comprehension, and spoken English scored by AI speech engines from the candidate's own recorded voice. A self-declared 'fluent' on a profile becomes a score you can set a bar against.

Role-relevant ability comes second: numerical, logical, verbal, abstract or attention-to-detail modules matched to the actual role — analytical modules for a data role, attention-to-detail for operations, customer-scenario judgement for support. Generic IQ-style testing bolted onto every role screens out good people for the wrong reasons.

Workplace reliability comes third, and it is the layer most screening skips: situational judgement scenarios measuring ownership, follow-through, escalation and communication under pressure. Remote hires rarely fail because they lacked skill on day one; they fail on the behaviours nobody measured — whether they flag a problem early, how they handle an unclear brief, what they do when a deadline slips.

Proctoring: what makes a remote result trustworthy

An unproctored online test proves only that somebody, somewhere, produced these answers. Proctoring closes that gap without flights or webcam interviews: an identity-baseline photo is captured at the start of the sitting, the camera monitors face presence and additional faces throughout, and the platform records tab switches and screen-capture attempts. Every attempt receives an integrity rating — High, Medium or Low — with the underlying evidence attached, not summarised away.

The evidence-attached part is the point. A rating you cannot inspect is just another claim. Before making an offer you can open the attempt, see the photo, see exactly what the monitoring flagged, and decide for yourself whether a Medium was a flatmate walking past or something worth a follow-up conversation.

The AI problem: why take-home tests stopped working

Since capable AI assistants became universal, an unproctored written exercise measures who has the better model, not who is the better candidate. Take-home coding challenges, written English samples and async written screens are all in the same position: the artefact tells you nothing about the person.

What still measures the person: proctored, timed sessions where the monitoring records tab switches and screen captures; spoken tasks scored from the candidate's own recorded voice, which no chatbot can sit; and situational judgement under time pressure, where there is no time to paste the scenario elsewhere. The integrity rating then tells you when someone was not working alone — with the evidence for you to review.

The workflow, end to end

Step 1 — pick the depth of evidence. A first-filter screen (written English, comprehension, a reliability SJT, proctoring) answers 'is this person worth a conversation?'. A full verification adds AI-scored spoken English, role-relevant aptitude modules and a per-candidate evidence report. Add role-specific modules only where the role needs them.

Step 2 — send one link. Delivery is a share link or QR code: candidates create no account, need no card, and can sit the assessment on any phone. The player is resumable, so an unstable connection in Lagos or Nairobi pauses the attempt rather than destroying it. This is what makes screening practical in weak-network markets — and it is why candidate-paid screening fails there: the employer or platform pays, the candidate just opens the link.

Step 3 — read the ranked results. Every submission is scored on arrival: overall and per-section scores, the integrity rating with its evidence, an AI-written summary of strengths and risks, and a live ranking of everyone in the drive. Shortlist from the ranking, review the integrity evidence on your finalists, and interview the few people the measurement says are worth the hour.

Candidates keep something too: a verified result can become a Skill Passport credential with a public verification link, so someone verified for one employer arrives at the next with evidence instead of a claim. Verification works better when it is done with candidates rather than at them.

What it costs

Verification is priced per candidate, in US dollars, with no platform fee, seat licences or annual contract: credits cost US$0.50 each and a verification assessment set runs 20–40 credits — US$10–20 per candidate — depending on depth. Credit packs (250 / 1,000 / 5,000 at US$125 / 465 / 2,175) discount volume screening by up to 13%. New organisations get 250 free credits, which is enough to verify a first shortlist before paying anything.

For comparison, the cost of not verifying is one bad remote hire: two to three months of salary, the opportunity cost of the seat, and the re-hiring cycle. At US$10–20 per candidate, verifying an entire 25-person shortlist costs less than one day of the salary you are hiring for.

Engineering roles: the half a code test does not reach

For a technical hire the workflow above needs one honest amendment. AssessAll does not execute candidate code — the platform's validation layer refuses to store a coding question at all, returning "Coding questions are not currently supported — the grading pipeline does not execute code." There is no IDE and no test-case runner in the candidate player. If you need to see code run against hidden test cases, buy a code-test platform; nothing here replaces one, and a proxy for coding ability is not coding ability.

What is worth measuring alongside it is the layer that decides whether a cross-border engineering hire works out, and it is not syntax. Tracing program logic with no language to hide behind — a 25-minute pseudo-code module covering loops, accumulators, nested iteration, integer division, value-versus-reference and a bug-spotting trace. Reading precisely: a 24-minute technical-reading module built on a product manual, an API changelog and an SOP, probed for conditions, exceptions, deadlines and precedence, where every answer is decidable strictly from the text. Executing a specified process exactly, deductive logic including the discipline of refusing an unwarranted conclusion, and judgement about when to escalate a slipping deadline.

That is the split most teams hiring engineers across a border end up running: something that executes code for the finalists, and something that measures reading, logic, judgement and English across everyone. The full version of this decision — including the arithmetic for when a code-test subscription becomes the cheaper instrument, and the South African statutory frame that changes how a screen may be built there — is on the remote developer verification page.

What verification looks like on the page — and the two panels that carry it

Verification is only as good as the artefact you can forward. When a hiring manager in London or Austin asks how you know the person in Manila, Lagos or Nairobi can actually do the work, the answer is a document they can read in four minutes — so it is worth knowing which parts of that document are load-bearing and which are decoration.

Two panels carry almost all of it. The first is quoted evidence from the candidate's own responses: a verbatim transcript of what the person said or wrote, with the item score beside it. Interpretation text can be assembled from a lookup table for any score; a transcript cannot. On AssessAll's voice report that panel appears only where the sitting captured responses longer than thirty characters, which means a screen built entirely from multiple-choice items produces a report with no evidence panel at all — and a remote-verification report with no quoted work is worth a question before it is worth a shortlist.

The second is the proctoring summary, and it needs reading more carefully than it usually gets. AssessAll's prints a band under an Integrity Band heading that in fact reports violations, so LOW in green is the good outcome; nine counters sit beside it — photos captured, tab switches, no-face events, multiple faces, paste blocked, print screen, window blur, resumes, looked away. The number that matters most is the photo count. With proctoring enabled but no band recorded the badge prints its most favourable label, so all-zero counters and zero photos is an absence of evidence, not evidence of a clean sitting. Ask any vendor what their equivalent badge shows when capture was off.

Both panels, and the six other questions worth putting to any vendor's sample, are annotated on the published Voice & Service Suitability report — the complete PDF with no signup — and on the report gallery.

The pass mark you did not set is 70 (added 16 September 2026)

Every discussion of remote verification is about whether the score is real. Almost none is about the bar the score is compared to, and the bar is where the quiet defaults live.

Counted across AssessAll's own shipped assessment definitions on 16 September 2026: 256 of 304 carry an explicit pass mark of 0, 40 carry 70 and 8 carry 60. The zero is deliberate, and the platform treats it as a first-class state rather than as a bar everyone clears — the submission handler suppresses all pass and fail framing when it sees an explicit 0, and issues no credential, because certifying a pass is meaningless on an instrument with no pass concept.

The number that matters to a buyer is the other one. Where the pass mark is never set at all, the platform applies a legacy default of 70. If you add a module of your own to a verification pack and leave that field alone, your candidates are being measured against a threshold nobody chose for the role, derived from nothing about the job, and it will look on the report exactly like a decision somebody made.

It is worth knowing how that default behaves, because the failure mode is a common one and it will exist in other systems too. An earlier version of the handler used a falsy check rather than a null check, which silently coerced a deliberate 0 into 70 — turning three shipped families of profile instruments, which have no pass concept at all, into tests with a 70% bar. It is fixed, and the fix is documented in the code beside the line. The general shape is the point: a threshold is the setting most likely to be inherited rather than chosen, and the least likely to be printed on a report next to the score it decided.

So add one question to every vendor conversation about remote verification, and it is one almost nobody in this market publishes an answer to: what is your default pass mark, and what happens to a candidate when nobody sets one? If the answer is a number, ask where it came from. If the answer is that there is no default, ask what the report prints in the pass column. A bar that arrived by inheritance is not a standard, and a documented standard-setting method is what turns it into one — there is a free panel calculator for that at the Angoff cut-score helper, and a published example of a screen that deliberately sets no bar at all at the national talent development cohort blueprint.