Contents
01 · The fixed-form era
Reused questions carry a risk.
Consider the fixed form: one set of questions, reused across roles and years. Publishers’ own pages document specific adjacent offerings. SHL runs a public practice-test portal for its own assessments. Korn Ferry has published a validity summary for its self-assessments. Hogan’s own FAQ explains the norms and reliability of its inventories. These pages document those specific offerings; they say nothing about what else these firms build or how common reuse is.
If a form’s questions never change, preparation for that form can become a market of its own. That is not a criticism of anyone’s research. It is a property of the architecture: a question that has been asked for years can be studied, coached and shared.
02 · Generation
The role is the starting point.
A job description should do more than choose a test label. It should shape the situations a candidate has to reason through: the responsibilities, competing priorities, exposure to risk and decisions that define the work.
Theta AI™ uses role-specific situational judgment scenarios with five response options and separate Best and Worst choices. Titan AI™ uses a conversational format to examine the reasoning behind a response. They are different evidence formats, not interchangeable scores.
03 · Measurement · Theta AI™
An estimate. Its uncertainty. The conditions behind both.
A score is an estimate, not an exact reading of a person. Theta AI™ uses an item response model to estimate competency and report the associated standard error. The standard error describes precision under that model; it does not establish predictive validity or eliminate model error.
Calibration status matters. Provisional or synthetic calibration must remain visibly separate from empirically validated norms. A live cohort is not evidence that its own calibration has been independently validated. Interpretation depends on the form, calibration asset, population and intended use.
04 · Adaptive evidence · Titan AI™
Ask for the reasoning a first answer leaves out.
A polished answer can leave important questions unanswered. Titan AI™ can follow a response with a targeted question about missing evidence or a contradiction. The aim is to make the reasoning inspectable, rather than to reward confidence, length or fluent phrasing.
Its four-axis scoring rubric separates Evidence, Communication, Risk awareness and Execution. Observable behavior provides the basis for each rating. A conversational judgment is still a judgment made under a rubric and model version, not a direct observation of future job performance.
05 · Integrity
Integrity signals inform review. They do not decide guilt.
A session event is context, not a misconduct verdict. A tab change, paste event or device signal can have more than one explanation. Any integrity finding should identify the observed event, the limits of the detector and what a human reviewer should check next.
Session controls and response-pattern analysis should be described separately. Preventing a second concurrent session is different from estimating whether a response pattern is unusual. Neither is a guarantee that an assessment is immune to assistance, coaching or misuse.
06 · Provenance
A finding should come with a way to inspect it.
A hiring team should be able to ask where a finding came from. The evidence view should connect an interpretation to the response, the scenario and the scoring method that produced it. A report is more useful when its reasoning can be questioned.
A provenance record and a fingerprint serve different purposes from a validation study. They can help check consistency with a recorded source and version. They do not prove that an original response was truthful, that a record was never compromised, or that a hiring inference is valid.
07 · Architecture
Two evidence formats. A decision that remains human.
Theta AI™ and Titan AI™ answer different assessment needs. A role-specific Best/Worst assessment supports structured cohort measurement. A conversational interview provides a way to examine reasoning in more depth. The hiring team chooses the workflow that fits the decision.
Where a two-stage workflow is used, HR approval separates screening from the next interview stage. This does not mean every customer must run both engines, or that a person has reviewed every generated report before publication. The employment decision remains with the hiring organization.
08 · The contrast
A form against a measurement system.
Assessment providers offer different instruments for different decisions. An adaptive cognitive test, a personality inventory and a conversational SJT do not become comparable because they all produce scores. Compare the construct, intended use, evidence base and limitations first.
SHL documents computer-adaptive ability testing. Korn Ferry publishes reliability and validity evidence for its KFCTD self-assessments. Hogan publishes technical and normative evidence for its inventories. These sources describe specific products and methods. They do not establish a ranking against this platform, and this page does not claim one.
Different questions need different evidence. A method specification describes how a score is produced. A calibration record describes the asset used to produce it. A validation study examines whether the intended interpretation is supported in a defined population and use case.
Keep these categories separate when evaluating an assessment. A synthetic demonstration is not a study of real-world validity. A sample report is a view of the output, not a performance benchmark. Translation does not by itself establish cross-language measurement equivalence.
Generic architecture contrast. It does not rate any named vendor product.
Human review
The hiring team makes the employment decision. A generated result does not imply that anyone reviewed it before publication.
Integrity signals stay prompts
The Response Integrity Review is designed to document the signals a session captures, with their source and limits. They are review prompts, never misconduct findings.
Provisional means provisional
Synthetic or provisional calibration is labeled on the report. It is never presented as a validated real-world norm.
The two-stage option
Where a two-stage workflow is used, HR approval separates screening from the next interview stage. It is not required of every customer.
Version discipline
Pipeline, score contract, model configuration and report provenance are recorded, so a result can be traced to the method that produced it.
Reproducible output
Report verification checks rendering, rasterization and page margins for PDF and HTML outputs.
Public source notes
A source is not a superiority claim.
Sources reviewed 25 October 2026. Vendor documents describe their own products. No independent head-to-head study is presented here.
Method descriptions, sample outputs and empirical validation are different kinds of evidence. This page does not claim universal superiority, validated predictive accuracy or equivalent measurement across languages.
The standard
Measurement discipline,rebuilt for the AI era.
More than 148,000 WISE assessments have been distributed, developed by our founder, Dr. Richie T. Kim. AI Psychometrics applies that measurement discipline to hiring: verifying what people can actually do, at a time when AI can manufacture any first impression.