NeuroFrame and Criteria (Cognify)

Side by side

Cognify and Emotify are products of Criteria Corp (United States); both came to it from Revelian Pty Ltd (Australia), acquired on 3 March 2020. Throughout this page each product is called by the name its rights holder sells it under: Cognify, UCognify, Emotify. All names and trademarks belong to their respective owners and appear here nominatively — to identify the product itself — without logos or brand styling.

What Criteria is, and where Cognify came from

Criteria Corp is a privately held US assessment company, founded in 2006 in Los Angeles by two PhDs, with headquarters in West Hollywood and a second office in Brisbane, Queensland. Cognify and Emotify did not originate inside it. They were built by Revelian Pty Ltd, an Australian company that Criteria acquired on 3 March 2020; the Brisbane office is that acquisition's legacy. Terms of the deal were not disclosed.

The ownership chain matters when you are signing a multi-year contract. Per the acquisition press release of 3 March 2020, that purchase followed "a significant growth investment in Criteria from Sumeru Equity Partners" in 2019; the release gives no figure for the investment and no ownership stake, and we found none elsewhere, so we state only what it says. The same release states that more than 500,000 individuals complete Revelian assessments annually for roles with more than 2,000 employers globally, and that the combined company would have over 4,500 clients in over 60 countries.

Criteria's About page states 4,000+ organizations, 80M+ assessments delivered, 6B+ data points and 60 countries, and lists the Middle East among its regions of presence. Game-based tests are one part of a wide catalogue: alongside Cognify, UCognify, Emotify and GAME sit classic aptitude tests (CCAT, UCAT, CBST), personality instruments (EPP, CPI, Illustrait), skills tests and a structured interviewing block — more than 25 instruments in the 2026 Assessment Portfolio.

That catalogue shape has a practical consequence for buyers: Cognify and Emotify are sold as components of packages — Criteria Assess, Criteria Interview, Criteria Hire — and we found no prices on the vendor's pricing page as of 4 August 2026. The public price signal we did find is the procurement aggregator Vendr, which as of 4 August 2026 reports a median of $13,225 per year and tiers of $5,000–$12,000 (100–500 assessments), $12,000–$35,000 (500–2,000) and $25,000–$100,000+ (2,000–10,000+). Vendr does not disclose its sample methodology, and we did not find a separate price for Cognify or Emotify there.

As for recent direction: the announcements in Criteria's press feed for 2025–2026 concern AI interviewing — Interview Intelligence on 19 May 2025, and the Predictive Interviewing expansion with Real-Time Video Interviewing and an AI Interview Agent on 16 June 2026. We found no press announcements about Cognify or Emotify in that period. We did not find a public changelog as of 4 August 2026, so a press feed cannot tell you whether the products themselves changed; if release history matters to your procurement, ask for it in writing.

How the assessment is built: games, rounds, minutes

Per the vendor's Cognify Information Brief (©2025, checked 4 August 2026), the standard version consists of three mini-games and is stated as "10 minutes, timed + tutorials". The same brief gives the round limits: Grid Lock — "9 rounds of increasing difficulty with a time limit of 3 minutes"; Numbubbles — "10 rounds of increasing difficulty with a total time limit of just under 3 minutes"; Proof It — "5 rounds that the candidate moves through in a 5-minute time limit".

Those three limits add up to roughly eleven minutes of timer; the stated total is ten. The vendor does not explain the relationship between per-game limits and total stated time in the materials we saw as of 4 August 2026, and we do not read anything into it beyond the arithmetic. The same is worth noting for the six-game version, available on request per the Cognify product page (Short Cuts, Grid Lock, Resemble, Tally Up, Numbubbles, Proof It): open documents give different durations for it — "20 minutes + tutorials" in one APAC document (©2023), and both "15 to 30 minutes (depending on version) plus tutorials" and "Estimated Time: 10 to 20 minutes (depending on version) + tutorials" inside a single other APAC file. The 2026 Assessment Portfolio states 10 minutes + tutorials for the three-game version. If the candidate's time budget is a real constraint for you, fix the version and the stated duration in the contract.

The constructs are mapped to the CHC model, and the Information Brief prints the mapping directly: Fluid Reasoning → Problem Solving (Grid Lock), Quantitative Knowledge → Numerical Reasoning (Numbubbles), Reading & Writing → Verbal Knowledge (Proof It). The report gives an overall percentile plus three sub-percentiles. The overall score is interpreted as cognitive aptitude.

Emotify is a separate instrument: "three separate ability-based mini-assessments", stated as "20 minutes, timed". Matching Faces — "30 rounds with a time limit of 3 seconds per round"; Emotional Ties — "20 rounds with a maximum time limit of 1 minute per round"; Emotions in Action — "14 rounds with a maximum time limit of 1 minute per round". The round limits here are stated as maximums rather than fixed durations, so actual session length varies by candidate.

Emotify's theoretical basis is the ability model of emotional intelligence of Mayer and Salovey (1997), and the vendor explicitly contrasts it with trait models and self-report, on the argument that an ability format is harder to fake. That is a methodological difference from personality questionnaires, and it is worth understanding before you compare Emotify with anything that asks candidates to rate themselves.

What the vendor publishes about validity and reliability

The primary open document for each instrument is an Information Brief. The Emotify Score Report Guide carries the notice "Proprietary and confidential. No part of this document may be disclosed without the prior written consent of Criteria Corp". What follows is drawn from the open briefs, checked on 4 August 2026; we did not test what is available under NDA or on request, and no claim here should be read as a statement about documentation we have not seen.

For Cognify, the brief reports convergent validity against Revelian's Cognitive Ability Test (RCAT), described as similar in complexity and format to the CCAT: correlations "ranging from r = .301 to r = .54", with "approximately 500 people completed each game". Test–retest reliability is given as .81 on a sample of 280 participants over a one-week interval, with the vendor's own caveat that participants completed the assessment on different devices, "potentially attenuating the correlation". A device comparison on 280 randomly assigned participants is reported as showing non-significant differences in performance.

On group differences, the brief states that "small gender differences were observed in some of the games"; for Problem Solving and Numerical Reasoning it reports no difference between first- and second-language English speakers, and "only minor differences" on Verbal Knowledge. The vendor also reports, of its own accord, "a weak to moderate positive relationship between game hours played per week with scores on one of the games within the Cognify suite (Grid Lock)", and characterises the strength of that relationship as minimal and comparable to the effect of experience with conventional psychometric tests.

One peer-reviewed study is Landers, Armstrong, Collmus, Mujcic and Blaik (2022), Journal of Applied Psychology 107(10), 1655–1677. It reports convergence with g at β = .97 on N = 633 (with GPA data), prediction of GPA at r = .16, and prediction of supervisor ratings at r = .29 on N = 49. The article's own disclosure reads: "Participant payments and graduate research hours in this study were funded by Revelian Pty Ltd, and Richard N. Landers became a compensated member of Revelian's Scientific Advisory Board midproject".

For Emotify, the brief gives test–retest r = .74 (n = 322, an average of two weeks apart) and construct validity against the STEU and STEM: r = .40, .54 and .57 against the STEU for Matching Faces, Emotional Ties and the overall score (corrected values .62, .70, .73, over 3000 participants), and r = .38 for Emotions in Action against the STEM (corrected .72, over 4500). The raw and corrected coefficients differ — .38 against .72 in the STEM comparison — and the vendor prints both; worth keeping in view when you compare these numbers with anyone else's. Criterion correlations in the brief are against self-report scales: workplace conflict r = −.19 and self-reported stress r = −.12 (n = 931); DASS-Stress r = −.13 (−.16) and the Brief Resilience Scale r = .22 (.27) (n = 770). Case studies add outcomes without coefficients: 115 employees in South Africa with supervisor-rating differences of +20% to +39% across seven metrics, and an Australian financial institution where the top 20% on Emotify were "58% more likely to be successful in the video interview".

In the Cognify Information Brief (©2025) the vendor states that the instrument has been independently validated by a multinational tech company and by Richard Landers and his team at Old Dominion University; an APAC document (©2023) describes the latter as a 2017 study presented at the SIOP convention in 2018.

In these open documents we did not find internal-consistency coefficients, standard error of measurement, full normative tables, or data on differences by race and ethnicity, as of 4 August 2026. Whether such material exists in client documentation under contract is something we did not check and do not assert either way. If those figures matter to your methodologist, request them directly — that is a normal procurement question, not an accusation.

Regulatory status: who owes the bias audit, and what an audit proves

Start with the fact most vendor comparisons get wrong. Under New York City's Local Law 144, the duty to commission and publish an annual bias audit falls on the employer or employment agency — 6 RCNY § 5-301 and § 5-303 are addressed to them, not to the tool's supplier. A vendor that publishes an audit is volunteering a maturity signal; the rules do not address that duty to the supplier. We give no legal assessment of any vendor's position — we describe the requirement and cite the source. The rules also define independence narrowly: an auditor is not independent if involved in the tool's use, development or distribution, employed by the employer or the vendor at any point during the audit, or holding a direct or material indirect financial interest in either. A vendor's own information brief cannot substitute for an audit under that definition.

On that basis: as of 4 August 2026 we did not find a published bias-audit summary specific to Cognify or Emotify on the vendor's site. We state that as the result of our search and nothing more — it is neither a violation nor a statement about how the instruments behave.

It is equally important to know what an audit does not establish. The FAccT 2025 study "Auditing the Audits" collected 44 published reports covering 116 audits and found audits for roughly 2% of Fortune 500 companies. Within that corpus, 53% of audits contain at least one impact ratio below 0.8, 54% contain at least one ratio above 1.0 (impossible under the law's own definition, indicating inconsistent choice of comparator group), and 83% report missing demographic data. In one published game-assessment audit, 331,177 candidates with unknown demographics fall outside the gender table — more than the 283,256 included in it. And the four-fifths rule itself, 29 CFR § 1607.4(D), says a ratio above 0.8 "generally" will not be regarded as evidence of adverse impact, then immediately adds that smaller differences may still constitute it where they are significant in both statistical and practical terms. It is an evidentiary heuristic, not a safe harbour.

Separately, and independently of Criteria, here is what does get published in this market. Harver (the soft-skills platform formerly known as pymetrics) publishes a BABL AI audit dated 17 July 2025, with all disclosed impact ratios at or above 0.914; HireVue publishes DCI Consulting audits including a separate table for its game modules; the audit covering Arctic Shores was published by an employer-client, FDM Group.

On complaints and proceedings: in the sources we checked as of 4 August 2026, we did not encounter reports of complaints filed or proceedings brought concerning Criteria, Cognify or Emotify. That is a statement about our search, not a finding about the company. Where such filings exist for other vendors in this market, the only responsible way to report them is as the fact of filing, with no regulator's decision attached.

The 2026 landscape, briefly and with dates, because a great deal of published material on this is out of date. The EU AI Act names recruitment and candidate evaluation as high-risk in Annex III, point 4(a) — but the obligations of Chapter III, Sections 1–3 were deferred to 2 December 2027 for Annex III systems by Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026. Transparency duties under Article 50 remain at 2 August 2026. The Chapter II prohibitions have applied since 2 February 2025 and were not softened. Which of them cover which specific products is deliberately outside the scope of this page: that is a legal qualification, not a vendor comparison, and it is a question for your counsel. In the US, Colorado repealed and reenacted its AI statute through SB 26-189, signed 14 May 2026, and federal EEOC guidance on AI in hiring was withdrawn while Title VII and the UGESP remain in force. In the UAE, the binding rule is Article 18 of the federal PDPL — the right to object to decisions made by automated processing, including profiling — and, in the DIFC, Regulation 10 on autonomous systems.

Languages and the Gulf: what the vendor's own language sheet says

If you are hiring in the UAE or the wider Gulf, this is the section that decides most of it. Criteria's Test Languages sheet (©2026, "Last Updated: 03/18/26") lists Cognify for Turkish; the Cognify product page gives English (US) as the default. UCognify — the language-independent alternative the vendor points to for international use — is listed for twelve languages: Simplified Chinese, Canadian French, French, German, Italian, Japanese, Korean, Brazilian and European Portuguese, Castilian and Latin American Spanish, and Turkish. Arabic appears in that sheet for the Employee Personality Profile and the Universal Cognitive Aptitude Test. Russian appears for the UCAT.

One caveat we insist on, because it cuts against a conclusion that would be convenient for us. Emotify is not listed under any language in that sheet — but we found no emotional-intelligence category in the sheet's "By Test" section, and several other instruments are also absent from it. Absence from the sheet therefore does not establish absence of localisation. We draw no conclusion about Emotify's language availability; ask the vendor directly.

A separate and practically important distinction: the Candidate Summary Report is available, per the same sheet, in 20 languages including Arabic. So the report an employer reads and the interface a candidate sits in front of are two different questions. The vendor also notes that additional fees may apply for languages and, for languages it does not cover, points to third-party translation without warranting its accuracy. For a Gulf employer that means one concrete procurement question: in which language will the candidate actually take the assessment, and in which language will the hiring manager read the result?

There is an academic literature on the cross-national applicability of game-based cognitive assessment, but the studies most often cited on this point sit behind paywalls and we could obtain only abstracts. We therefore do not carry any of their conclusions onto any named product. What we can say is narrower and safer: cross-cultural invariance is an open empirical question for this genre generally, and a buyer comparing candidate pools across countries should ask any vendor — including us — what evidence exists that scores mean the same thing in each of them.

For symmetry: our own language position is not stronger. The NeuroFrame client handbook exists in Russian only. If your methodologists, your works council or your legal team need to read the underlying methodology in Arabic or English, that is a real gap on our side today, and it belongs in this comparison as much as anything above.

Where the norm comes from: an external reference or the client's own staff

Two instruments can produce identical-looking percentiles and still answer different questions, because the comparison group differs. Broadly there are two designs. In the first, a candidate is compared with a large general norming population — the question answered is "how does this person rank among people in general?". In the second, the requirement is defined externally, per role, and the candidate is measured against that requirement rather than against a crowd. The second design has an important sub-case worth naming separately: a model trained on the client's own successful employees, which answers "how similar is this person to the people we already have?".

Criteria's reports compare a candidate against what the Emotify Score Report Guide calls "our global norming group", with five classification bands. For Emotify, the brief describes that group as "thousands of individuals from a wide range of industries and job types", aged 18 to 64, gender-balanced and ethnically diverse. The exact size and geography of the norming sample are not disclosed in the open materials as of 4 August 2026.

NeuroFrame uses both layers, and they are separate objects. The comparison sample is 14,850 people in real jobs across 21 industries, 21 functions and 8 grades, drawn from 500 companies whose executives are in the sample — not from the client list. The requirement benchmark is a separate artefact: 287 professions × 8 corporate lifecycle stages = 2,296 records, built from the Russian occupational-standards register and the Adizes lifecycle model.

The design choice there is deliberate and has consequences in both directions. Because the benchmark is built externally rather than from a client's successful incumbents, it works from day one — you do not need dozens of current employees in a role to start — and it does not structurally encode the composition of the staff you already have. The cost is the mirror image: the benchmark reflects an external occupational register rather than your company's specifics, 83% of its roles sit at specialist level, and top management is covered by nine professions.

One finding from that benchmark is worth stating because it is checkable arithmetic rather than a slogan. Requirements move with the company's stage of life. Computed across the 2,296 records, 99% of the 287 professions have requirement ranges at "Infancy" and at "Bureaucracy" that overlap by less than half, with a median range overlap of 0.40. Of 153 non-overlaps between the extreme stages, 152 fall on risk appetite. The travel of the range centre across the lifecycle is 18.8 percentiles for risk, 15.4 for openness, 11.5 for conscientiousness and 9.0 for cognition; result focus barely moves at all.

How NeuroFrame's approach differs — and what it does not buy you

The structural difference is the session. Instead of a battery of short discrete games, NeuroFrame is one continuous strategic simulation of 30 to 60 minutes, producing eight parameters — three in the cognition domain and five in the personality domain. A battery samples several separate abilities in a few minutes each; a single long scenario watches one person behave inside one situation as it develops.

What that buys, stated precisely. Dynamic simulations with delayed consequences measure complex problem solving — a construct that overlaps with general cognitive ability by roughly 18% of variance (r ≈ .43, 95% CI [.37; .49]; Stadler, Becker, Gödker, Leutner and Greiff, 2015, Intelligence 53, 92–101, a meta-analysis of 47 studies), meaning it contains something classical tests do not capture. Vigilance decrement — the decline in sustained attention as time on task grows — is among the most replicable effects in cognitive psychology at the group level. And a long session allows behaviour to be observed both fresh and fatigued; we treat that as observation, not as an individual fatigue score, because difference scores have poor retest reliability.

What it does not buy: higher predictive validity, and we will not claim it. On the current re-estimation by Sackett, Zhang, Berry and Lievens (2022), structured interviews sit at .42, job-knowledge tests at .40, work samples at .33, cognitive ability tests at .31, assessment centres at .29, situational judgement tests at .26 and conscientiousness questionnaires at .19 — high-fidelity simulations do not top that list. A systematic review of 34 studies (Ramos-Villagrasa et al., 2022) concluded that the game format does not offer sufficient advantages to be recommended in place of conventional methods, other than improved candidate reactions. Our honest position is comparable validity with substantially better candidate reactions — and a difference that lies in the form of the result rather than in the coefficient.

That form is the actual argument. Each of the 2,296 benchmark cells carries a human-readable rationale: 2,296 unique texts, 1.07 million characters, an average of 464 characters per cell, and 32% of them name the rule that fired outright. In 303 cells out of 2,296 (13%) the benchmark permits a high risk appetite alongside a low floor on cognition or progress monitoring — that is, the method can say "fits on every parameter and still warrants attention". A single aggregate match percentage cannot express that in principle. This is also the direction the 2026 rules described in the regulatory section above point in — not a ban on assessment, but explainability, documented bias checking and a human able to intervene; the sources for each of those rules are listed below.

One more structural property, with its cost attached. De-identification is architectural: the client receives codes and distributes them itself, so NeuroFrame never receives a name, a gender or an age. The model therefore cannot learn on protected attributes. The same design means we cannot compute an adverse impact ratio by gender or race on our own data, and it is the direct reason we have no ATS connectors. Finally, on genre: among the products we reviewed as of 4 August 2026 we did not find one that runs mass screening on a single continuous strategic simulation of 30+ minutes. We did not review the whole market — this is a statement about our search, not about anyone's product line. Per each vendor's own published materials as of 4 August 2026, the nearest products with one continuous session are Sova Immerse, which its product page states as a "15–30 min role simulation", and CapsimInbox, whose page states "15-60 minutes" — a different genre in both cases; Owiwi describes a single narrative, and we did not find a duration in its public materials. A preparation market exists around discrete, named tasks — a measurable property of the format: per Semrush, pulled 4 August 2026, search demand around Arctic Shores' individual mini-games runs to roughly 550 queries a month in total in the UK database, and "pymetrics games answers" to 110 in the US database. This is our own demand measurement rather than a vendor statement, so there is no link to give; we name the tool, the database and the date so it can be re-run. For Criteria specifically, JobTestPrep sells a preparation pack covering, in its own description, "3 of the 6 games" — Tally Up, Proof It and Resemble, of which Proof It is in the standard three-game version. We state that as a measurable property of discrete formats and draw no conclusion about any instrument's security.

How to read this page

This comparison is published by NeuroFrame — that is, by an interested party. For that reason every fact about the other product carries a link to a primary source, and we do not issue ratings at all. Where a question can only be answered by judgement rather than by a document, we say so and leave the judgement to you.

Method, stated so you can check it. Every figure about Criteria comes either from a document published by the vendor or from a named third-party source, with the date we checked it: 4 August 2026. Where the vendor's own documents disagree with each other — as they do on Cognify's duration — we print all the variants rather than picking the convenient one. Where a study sits behind a paywall and we could read only the abstract, we do not carry its conclusions onto any named product. Where we could not find something, we say we did not find it, not that it does not exist.

We also did not do several things on purpose. We give no legal opinion on whether any named product complies with the ADA, Title VII, NYC Local Law 144, the EU AI Act or any other regime — we describe the requirement and cite the source. We make no claim about documentation available to clients under contract, since we have not seen it. And we use no evaluative adjectives about another company's product anywhere on this page; if you find one, it is an error and we want to know.

Criteria, Cognify, Emotify and Revelian are trademarks of Criteria Corp or its affiliates. Harver, pymetrics, HireVue, Arctic Shores, FDM Group, Sova, Capsim, Owiwi, JobTestPrep, Vendr and all other product and company names on this page are trademarks of their respective owners. All are used here in a nominative comparative sense only. If you find a factual error, a dead link or a figure that has since changed, write to us — corrections to this page are published with the date they were made.

Questions

Cognify and NeuroFrame are both game-based. Are they interchangeable?

Not really — they are different genres. Cognify is a battery of short discrete mini-games: three in the standard version, stated at 10 minutes of timed play plus tutorials, mapped to CHC constructs and reported as an overall percentile plus three sub-percentiles (Criteria Information Brief ©2025). NeuroFrame is one continuous strategic simulation of 30–60 minutes producing eight parameters, three cognitive and five personality. A battery samples several separate abilities quickly; a single long scenario observes one person's behaviour as a situation develops. They answer overlapping but not identical questions, and it is entirely reasonable to run both.

Which of the two has higher predictive validity?

Neither of us can honestly answer that from published data, because the coefficients were obtained on different samples against different criteria and are not directly comparable. What is publicly checkable: Criteria's brief reports convergence with its own RCAT at r = .301–.54 and a peer-reviewed study reporting r = .29 against supervisor ratings at N = 49; we report R² 0.46 and AUC 0.77 on a 3,000+ validation sample and have no peer-reviewed publication. For context on the field, the current re-estimation (Sackett et al., 2022) puts structured interviews at .42 and cognitive ability tests at .31 — simulations do not top that list, and a systematic review of 34 studies (Ramos-Villagrasa et al., 2022) found the game format's main documented advantage is candidate reactions, not validity.

We hire Arabic-speaking candidates in the Gulf. What do I actually need to ask?

Two separate questions, because the answers differ. First: in which language will the candidate take the assessment? Per Criteria's language sheet of 18 March 2026, Cognify is listed for Turkish (English (US) is the product page's default) and UCognify for twelve languages; Arabic is listed for the EPP and the UCAT. Second: in which language will the hiring manager read the result? The Candidate Summary Report is available in 20 languages including Arabic. Emotify is not listed under any language in that sheet, but we found no emotional-intelligence category in the sheet's "By Test" section, so ask the vendor directly rather than assuming. On our side, the client handbook is Russian-only, which is its own limitation.

Does Criteria publish a bias audit, and does it matter if a vendor does not?

As of 4 August 2026 we did not find a bias-audit summary specific to Cognify or Emotify on the vendor's site — that is the result of our search, nothing more. It matters less than most buyers assume: under 6 RCNY §§ 5-301 and 5-303 the duty to commission and publish an annual audit falls on the employer or employment agency, not on the tool's supplier. A vendor audit is a voluntary maturity signal. It also proves less than it seems: across the 116 audits studied in FAccT 2025, 53% contained at least one impact ratio below 0.8 and 83% reported missing demographic data. We have no formal bias audit either, and our de-identified-code design means we cannot compute one without the employer joining codes to demographics.

Is the EU AI Act a problem for either tool right now?

The dates are the part most published material gets wrong. Annex III, point 4(a) does classify recruitment and candidate evaluation as high-risk — but the obligations of Chapter III, Sections 1–3 were deferred to 2 December 2027 for Annex III systems by Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026. Transparency duties under Article 50 remain at 2 August 2026. The Chapter II prohibitions have applied since 2 February 2025 and were not softened. We describe the requirements and cite the sources; which of them cover which specific products is deliberately outside the scope of this page — that is a legal qualification, and a question for your counsel rather than for a vendor comparison.

Can candidates prepare for these assessments?

Preparation markets exist around discrete, named tasks generally — that is a measurable property of the format, not a judgement about any instrument. Per Semrush, pulled 4 August 2026, search demand around Arctic Shores' individual mini-games runs to roughly 550 queries a month in total in the UK database, and "pymetrics games answers" to 110 in the US database — our own demand measurement, not a vendor statement, which is why we name the tool, the database and the date instead of giving a link. For Criteria specifically, JobTestPrep sells a preparation pack covering, in its own description, "3 of the 6 games" — Tally Up, Proof It and Resemble — of which only Proof It is in the standard three-game version. We draw no conclusion about how any instrument holds up under that; if it matters to you, ask each vendor what evidence they have on practice effects. On our side, note that the vendor's own Cognify brief reports a weak-to-moderate positive relationship between weekly gaming hours and scores on Grid Lock, and characterises it as minimal.

You are an interested party. Why should I trust this page?

You should not trust it — you should check it. That is why every fact about Criteria carries a link to a vendor document or a named third-party source with the date we checked it (4 August 2026), why we print all the conflicting durations from the vendor's own files instead of picking a convenient one, why we say "we did not find" rather than "does not exist", and why we use no evaluative adjectives about another company's product anywhere on this page. The section on when to choose Criteria rather than us is the longest one here and lists real gaps: no peer-reviewed publications, no formal bias audit, α of 0.69–0.77, no ATS connectors, a Russian-only handbook, no verbal or numerical scales. If you find an error, tell us and we will publish the correction with its date.