Work-sample evidence for AI capability

See who can actually work with AI — before you rely on it.

AISkillProof puts employees in realistic AI tasks and turns their work into capability evidence, targeted learning actions and a fresh reassessment that proves improvement.

Work

Realistic assignments instead of knowledge questions only.

Evidence

Every task maps output to concrete skills and review rules.

Action

Gaps drive learning and a fresh equivalent reassessment.

In formal assessments, objective items are scored on the server and answer keys stay there. Public examples are practice material.

Live sample task

AI Judgment · 90 sec · objective scenario

AIJ-02

Context

Your CEO asks you to forward an AI-generated management summary of quarterly results within ten minutes.

AI output + source conflict

2 sources disagree

The AI writes: ‘Revenue grew 14% in Q3.’ The sales deck shows 14%, but the CRM export shows 8% and has a more recent extraction timestamp.

What is the strongest next action before you send this to the CEO?

What this task measures

Source verification, uncertainty handling, escalation and AI judgment under time pressure. Pick an answer to see how evidence is created.

Not a quiz bank

This is what employees are actually asked to do.

AISkillProof does not measure confidence. The candidate has to make, check, decide or structure something. That produces much richer evidence than asking ‘how good are you with AI?’

01 · Prompt repair

Turn a chaotic briefing into a usable AI instruction.

PRI-03

Sample scenario

A sales manager dumps loose notes, pricing information and customer context into an AI tool and asks ‘write email’. You need to restructure the assignment so the output is useful and controllable.

What the candidate does

Writes a prompt with goal, audience, context, constraints, output structure and a verification instruction.

What the employer gets

Prompting & Instruction, task decomposition, context management and controllability.

02 · Verification challenge

Find the error hidden inside convincing AI output.

AIJ-02

Sample scenario

An AI summary uses two conflicting figures and presents one conclusion as certain. The candidate must decide what needs verification first.

What the candidate does

Investigates source provenance, recency, calculation and uncertainty before the output is used.

What the employer gets

AI Judgment, AI Literacy, verification behaviour and risk awareness.

03 · Workflow design

Build an AI workflow that does not break when reality deviates.

WFA-06

Sample scenario

Every Friday three data sources are combined into an operational report. The candidate must place input validation, AI steps, exceptions and human controls.

What the candidate does

Designs the sequence, defines checkpoints and states when automation must stop or escalate.

What the employer gets

Workflow & Automation, reliability thinking and human-in-the-loop design.

04 · Risk triage

Decide which data may and may not enter the AI workflow.

AIG-04

Sample scenario

An HR team wants AI to summarize CVs, performance notes and salary data. Not all data and tools are appropriate for that use.

What the candidate does

Selects minimum necessary data, approved environment, access rights and required review before processing starts.

What the employer gets

AI Governance, data minimization, tool judgment and policy awareness.

These are public, simplified work samples. Real assessments use versioned items, server-side answer keys and auditable review.

Try the free capability challenge

From task to decision

Not another score. An evidence chain a manager can act on.

The value is not the number itself, but the traceable path from work to evidence, gap, learning action and reassessment.

01→

Work task

Resolve source conflict

The candidate receives output, sources and time pressure — not just a theoretical question.

02→

Evidence

Verification is missing

The answer produces skill-level evidence with weight, item version and audit trail.

03→

Diagnose

AI Judgment is the gap

The gap follows from multiple evidence points, not one wrong answer.

04→

Learning

Small targeted intervention

No generic three-hour AI course; only the missing capability gets priority.

Proof

Fresh reassessment

Same skill construct, new items. Baseline and reassessment make improvement visible.

Synthetic example

An employee shows weak verification behaviour across three AI Judgment tasks. The system recommends a short source-checking intervention, then schedules a fresh equivalent reassessment. Only the real delta is reported.

Trust is a product feature

Measure employees without turning assessment into surveillance.

AISkillProof is designed around development, transparency and minimum necessary data access. A score should be explainable and managers should only see what they need to support development.

TENANT-SCOPED · EXPLAINABLE · AUDITABLE

Development Mode by default

The default experience helps people grow. Ranking and public leaderboards are not the product model.

Tenant data stays tenant data

Organization boundaries are enforced server-side and with Row Level Security.

Explainable grading

Open answers receive a score only through a versioned rubric and audit trail, not a hidden model guess.

Privacy and security by design

Data minimization, role-based access, retention and transparency are product features, not footnotes.

MEASURE → IMPROVE → PROVE

Three tasks that show what AISkillProof actually measures.

No screenshot product tour. Make a judgment call, repair a bad prompt and design a reliable AI workflow yourself. These are simplified public work samples — not a formal assessment.