Put somebody in the real situation. Score what they actually did.

Storm researches the role before it builds anything: what the work actually is, what good looks like, where it goes wrong, and what a strong performer does that a weak one does not. Domain question banks and industry exemplar sets sit behind that step, so the scenario is built on the trade, not on a generic competency wheel.

The rubric comes out of the same research. That is what lets you defend a score in a field you have never worked in, and what lets a specialist hiring manager read the report and recognise their own job in it.

A simulation is not a quiz and it is not a scripted role play. It is a situation: a difficult customer with a real objection, an incident with incomplete information, a plan that has to be defended to someone who disagrees with it.

The counterpart is a voice agent that responds to what is actually said, so there is nothing to rehearse and nothing to game. The output is the work itself, plus a complete record of how it got made.

Storm collapses that middle. The pile gets screened and ranked automatically. The shortlist gets one link and sits the simulation whenever they like, inside the window you set. The report is written before you open it, the same way for every candidate, with a quote behind every line.

Your judgement is still the product. It just arrives on the first day of the search instead of the ninth, with something underneath it.

Load the CVs you already have, from the ATS, the inbox or a folder. Storm parses each one, normalises the messy parts, and makes the whole pile searchable in plain English: process engineers in the EU with pharma GMP experience and no visa dependency.

Ranking looks at fit against the brief, depth of evidence behind each claimed skill, and how far a CV inflates. Every extracted field carries a confidence and a source, so you can see what was stated, what was inferred, and what a recruiter typed in. Then you invite the top of that list, not the whole pile.

Storm runs people through problem-solving scenarios with AI in the loop and grades the behaviour rather than the tool: how somebody directs a model, what they ask it next, and whether they check the answer before they act on it.

It is a readiness read based on what people did, not on what they say they do. Leadership gets who is ready today, who needs which training, and where AI would land safely first.

The session is recorded as a trajectory: what the person asked the model, what it returned, what they did next. Scores attach to those turns, and a turn nothing supports does not get scored at all.

Directing a model and trusting one are different skills, and they fail separately. Somebody who writes an excellent prompt and ships the output unverified is not AI-ready. They are faster at being wrong.

Everyone sits the situation their own team actually faces, at a time they pick, inside a window you set. The engine grades the work, and the results roll up into one map: where the team is deep, where it is thin, and which gap is worth spending money on.

No rankings, and no verdicts on individuals. Internal evaluation returns capability and growth. Nobody is scored behind their back, and nobody is set against the colleague at the next desk.

Each team is graded on the capabilities that team actually uses, so the grid compares like with like. Every number in it is a set of scored simulations underneath, and every score points at a quote.

One cell is marked, not one row. Operations diagnoses well, decides well and executes well. It falls over at the point where somebody has to check whether the answer in front of them is true, which is a training problem with a name and a price.

Storm turns the module into the situation it was meant to prepare people for. Not finished the responsible AI module, but handled a live scenario where the agent hallucinated, caught it, escalated correctly and scored in band. That is a demonstrated capability, and it has a name and a date attached to it.

The same record redeploys the bench. An internal move gets argued from what somebody did in the target role rather than from keyword overlap between two documents, which is where bench ageing quietly leaks margin.

The same rubric that found the gap grades the re-run, line by line. The grey on each track is where the cohort started. The colour is what the drill added.

Holding the line under pressure barely moved. The drill rehearsed escalating, and the transcripts show people escalating and then accepting the answer anyway. That is a scenario design problem, and it is far cheaper to find here than in a client review next year.

One work simulation of thirty to forty-five minutes, in place of the aptitude test and the panel rounds. Every candidate in an intake faces the same scenario against the same bar, and the output is band-level, which a recruiter can act on without assembling anything first. Panel time is delivery time you pay for twice.

Graduate assessment is tightly regulated, so the compliance pack comes before the volume: adverse impact testing, candidate data handling, accessibility, and a route to appeal a result. That work is in progress, and this is not a use case we will run until it is finished.

The report reads the same way for every candidate, so a student can hold one run against the next and see whether the thing they practised actually moved. One line is marked: the one worth working on before the next application.

Storm never rejects anybody. It produces a score and the evidence under it, and a person makes the call. What the student gets is the same document the recruiter reads, which is the only version of this that is fair.