--- title: "MatchIQ as an architecture: why matching needs independent verification" description: A martinfowler.com experiment shows reliability comes from independent checks, not the model. What that means for how engineering candidates get matched. image: https://blog.nexton.dev/hubfs/Dise%C3%B1o%20sin%20t%C3%ADtulo%20(1)-Oct-29-2026-04-26-27-7694-PM.png --- [![nexton-logo-black-blue-1](https://blog.nexton.dev/hubfs/nexton-logo-black-blue-1.svg "nexton-logo-black-blue-1")](https://nexton.dev) Open main menu Close main menu # MatchIQ as an architecture: why matching needs independent verification ![Nexton](https://blog.nexton.dev/hs-fs/hubfs/circle-cropped%20(8).png?width=45&name=circle-cropped%20(8).png) [Nexton](https://blog.nexton.dev/industry/author/nexton) [Share this blog post on Twitter](https://twitter.com/intent/tweet?text=I+found+this+interesting+blog+post&url=https://blog.nexton.dev/industry/matchiq-as-an-architecture-why-matching-needs-independent-verification) [Share this blog post on Facebook](http://www.facebook.com/share.php?u=https://blog.nexton.dev/industry/matchiq-as-an-architecture-why-matching-needs-independent-verification) [Share this blog post on LinkedIn](http://www.linkedin.com/shareArticle?mini=true&url=https://blog.nexton.dev/industry/matchiq-as-an-architecture-why-matching-needs-independent-verification) ![](https://blog.nexton.dev/hubfs/Dise%C3%B1o%20sin%20t%C3%ADtulo%20(1)-Oct-29-2026-04-26-27-7694-PM.png) In October 2026, Naresh Jain published an experiment on martinfowler.com that most CTOs will read as a story about coding agents. It is also a story about hiring. He gave two models the same build task ten times each, with and without a verification harness: contracts, examples, API tests, and an independent set of 178 acceptance tests the agent never saw. The cost-optimized model passed 3 of 10 runs on its own and 10 of 10 inside the harness, at 33% less per run than the frontier model without one ([Jain, October 2026](https://martinfowler.com/articles/are-harnesses-worth-the-cost.html)). The reliability did not come from the model. It came from the independent checks around it. This playbook takes that result and applies it to the system most engineering leaders use without thinking of it as a system: how candidates get matched to their open seats. It explains why a generator alone is the wrong shape, what an independent check looks like in hiring, and how to evaluate any matching process, ours included, by its architecture rather than its claims. ## What do you need before you start? You need to be able to describe your current hiring pipeline as a system, with inputs, a generator, and checks. - Where candidates come from, and what filters them before a human looks. - Who makes the first judgment call, and on what evidence. - Which step, if any, can send a candidate back rather than forward. - What reaches the hiring manager: a name, a résumé, or an evaluation. ## Why is a generator alone the wrong shape? Because a generator optimizes for producing plausible output, and plausibility is not correctness. Why this matters: Jain's unharnessed runs did not fail loudly. They produced services that looked finished and failed acceptance tests the agent had never seen. Birgitta Böckeler's definition is useful here: a harness is "everything in an AI agent except the model itself," made of guides that constrain the work and sensors that check it ([Böckeler, April 2026](https://martinfowler.com/articles/harness-engineering.html)). A matching system that ranks candidates by similarity to a brief is a generator. Without sensors it will hand you a confident list, and you will discover the misses after the first interview, or after the first month. What good looks like: a pipeline where at least one step is designed to reject, and does. ## What is an independent check in hiring? A judgment made by someone who did not produce the match, on evidence the generator did not see. Why this matters: Jain's acceptance tests worked because they were outside the agent's loop. The same principle applies to people. A recruiter reading the match against the brief is one check. An engineer interviewing the candidate on real work, and writing down what they found, is a second, and it tests things no profile contains: how the person reasons, what they do with a wrong answer, whether they can build. The two checks are independent of each other and of the matching step, which is the point. "What changes with model capability is not whether independent sensors are needed, but when they run," Jain writes. Better generators do not remove the need for verification; they change where it sits. What good looks like: an evaluation that reaches you written by the person who ran the check, not a summary of the generator's score. ## How do you evaluate a matching process by its architecture? Ask three questions: what generates, what verifies, and whether the verification is independent. Why this matters: every provider will tell you their matching is accurate. Few can show you the structure. The questions that separate them: Is there a step that can reject a generator's output, and how often does it? Who performs the verification, and were they involved in producing the match? What artifact do you receive, and can you read the reasoning? If the answer to the first question is "the hiring manager, in the interview," the verification is yours, and you are paying for a generator. What good looks like: a provider who can draw the pipeline on a whiteboard and point to the step that says no. ## How do you read the artifact? A good evaluation tells you what was tested, what was found, and what was not tested. Why this matters: Jain's result includes its own limits: one greenfield service, a detailed brief, ten runs. Good verification says what it did not cover. A written technical evaluation should do the same: which areas the interviewer probed, what the candidate did well, where they struggled, what the interviewer could not assess in the time. An evaluation with no weaknesses in it is a sales document. What good looks like: a one-page write-up you could hand to your own tech lead and have them run a second interview from. ## What are the common mistakes? - **Buying the generator and calling it the system.** Accuracy claims about matching say nothing about what happens to the misses. - **Treating the recruiter's screen and the engineer's interview as the same check.** They test different things; collapsing them loses one. - **Accepting a score instead of a reason.** A number with no written judgment behind it is not verification; it is output. - **Assuming a better model removes the need for review.** It moves where review runs. It does not remove it. ## How does Nexton do this? MatchIQ is built as a generator with independent verification, which is the shape this playbook describes. Three layers of review stand between a candidate and your inbox: AI matching, recruiter judgment, and engineer vetting. Each layer can stop a candidate, and the layers are separate from one another. The last one is the same mechanism Nexton has always run: every engineer is interviewed by another engineer before you see the profile, and the written evaluation comes with the candidate. That evaluation is the artifact. If the bottleneck in your own delivery has moved to review, as [The bottleneck moved from writing code to reviewing it](https://blog.nexton.dev/industry/the-bottleneck-moved-from-writing-code-to-reviewing-it-and-that-changes-who-you-hire) argues, you will recognize the design: the generator is fast; the checks are where the reliability lives. For the interview mechanics themselves, see [How to run engineer-to-engineer technical vetting](https://blog.nexton.dev/industry/how-to-run-engineer-to-engineer-technical-vetting). If you want to see what one of those written evaluations looks like before you see a single candidate, it's worth a conversation. ## FAQs **Does AI matching replace recruiters?** No. In this architecture the matching step generates candidates; recruiter judgment and engineer vetting are the independent checks that decide who moves forward. Removing either check removes the reliability. **How is this different from a résumé-screening tool?** A screening tool is a generator with a filter. The difference is whether a person who did not produce the match evaluates the candidate on evidence the tool never saw, and writes that evaluation down for you. **What should I ask a provider about their matching?** Where the step is that can reject a match, who runs it, whether they were involved in producing the match, and what written artifact you receive. --- Sources: [Naresh Jain, "Are Harnesses Worth the Cost?"](https://martinfowler.com/articles/are-harnesses-worth-the-cost.html), martinfowler.com, October 2026 · [Birgitta Böckeler, "Harness engineering for coding agent users"](https://martinfowler.com/articles/harness-engineering.html), martinfowler.com, April 2026. Verified October 2026. ![Nexton](https://blog.nexton.dev/hs-fs/hubfs/circle-cropped%20(8).png?width=150&name=circle-cropped%20(8).png) #### Nexton **Blog Team** Nexton empowers companies connecting them with the strongest engineering talent in LATAM. ### Leave a Comment [![Isologotype](https://blog.nexton.dev/hubfs/Isologotype.svg "Isologotype")](https://nexton.dev/) ```json { "@context" : "https://schema.org", "@type" : "BlogPosting", "author" : { "@type" : "Person", "name" : "Nexton", "url" : "https://blog.nexton.dev/industry/author/nexton" }, "dateModified" : "2026-10-29T16:26:35.646Z", "datePublished" : "2026-10-29T16:26:35.000Z", "headline" : "MatchIQ as an architecture: why matching needs independent verification", "image" : [ "https://blog.nexton.dev/hubfs/Dise%C3%B1o%20sin%20t%C3%ADtulo%20(1)-Oct-29-2026-04-26-27-7694-PM.png" ], "mainEntityOfPage" : { "@id" : "https://blog.nexton.dev/industry/matchiq-as-an-architecture-why-matching-needs-independent-verification", "@type" : "WebPage" }, "publisher" : { "@type" : "Organization", "logo" : { "@type" : "ImageObject", "url" : "https://blog.nexton.dev/hubfs/Nexton%20Hubspot.webp" }, "name" : "Nexton" } } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#organization", "@type" : "Organization", "address" : { "@type" : "PostalAddress", "addressCountry" : "US", "addressLocality" : "Miami", "addressRegion" : "FL" }, "aggregateRating" : { "@type" : "AggregateRating", "bestRating" : "5", "itemReviewed" : { "@id" : "https://nexton.dev/#organization" }, "ratingCount" : "12", "ratingValue" : "4.9", "url" : "https://clutch.co/profile/nexton" }, "areaServed" : { "@type" : "Country", "name" : "United States" }, "description" : "Nexton is a human-powered company building nearshore engineering teams for US companies. Senior Latin American engineers, hand-matched and vetted by engineers. 1,500+ engineers matched with 350+ US companies.", "disambiguatingDescription" : "Nearshore technology staffing company (nexton.dev) serving US companies with software engineers from Latin America. Not affiliated with Nexton, the master-planned residential community in Summerville, South Carolina (nexton.com); nor with NEXTON Group, the French digital consulting and IT services company (nexton-group.com); nor with the Navee Nexton electric scooter.", "employee" : [ { "@id" : "https://nexton.dev/#lucas-arana" } ], "founder" : [ { "@id" : "https://nexton.dev/#diego-sternberg" }, { "@id" : "https://nexton.dev/#agustin-di-luciano" } ], "foundingDate" : "2017", "knowsAbout" : [ "Nearshore engineering teams", "Software engineering recruiting", "Latin American engineers", "AI engineers" ], "legalName" : "Nexton", "logo" : "https://nexton.dev/Nexton/nexton-logo.png", "makesOffer" : [ { "@type" : "Offer", "itemOffered" : { "@type" : "Service", "description" : "Nexton's primary model: embedded, long-term engineering capacity. Nearshore LATAM engineers join the client's team as fully integrated, ongoing members rather than short-term contractors.", "name" : "Staff Augmentation" } }, { "@type" : "Offer", "itemOffered" : { "@type" : "Service", "description" : "Direct recruitment of LATAM engineers for full-time positions at US companies, with a technical validation process run by engineers.", "name" : "Direct Hiring" } }, { "@type" : "Offer", "itemOffered" : { "@type" : "Service", "description" : "Onboarding LATAM engineers as international team members, with no local legal entity required.", "name" : "International Team Member" } }, { "@type" : "Offer", "itemOffered" : { "@type" : "Service", "description" : "Payroll management and legal compliance to hire international talent without administrative friction.", "name" : "Payroll Processing (EOR/PEO)" } } ], "name" : "Nexton", "numberOfEmployees" : { "@type" : "QuantitativeValue", "minValue" : 50 }, "sameAs" : [ "https://www.linkedin.com/company/nexton", "https://clutch.co/profile/nexton", "https://www.crunchbase.com/organization/nexton-b25a", "https://www.trustpilot.com/review/nexton.dev", "https://www.wikidata.org/wiki/Q140867481", "https://www.inc.com/profile/nexton", "https://blog.nexton.dev/" ], "slogan" : "Hand-matched Engineers. Vetted by Engineers.", "url" : "https://nexton.dev/" } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#diego-sternberg", "@type" : "Person", "jobTitle" : "Co-Founder & CEO", "name" : "Diego Sternberg", "sameAs" : [ "https://www.linkedin.com/in/diegosternberg" ], "worksFor" : { "@id" : "https://nexton.dev/#organization" } } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#agustin-di-luciano", "@type" : "Person", "jobTitle" : "Co-Founder", "name" : "Agustín Di Luciano", "worksFor" : { "@id" : "https://nexton.dev/#organization" } } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#lucas-arana", "@type" : "Person", "jobTitle" : "COO", "name" : "Lucas Arana", "sameAs" : [ "https://www.linkedin.com/in/lucasarana" ], "worksFor" : { "@id" : "https://nexton.dev/#organization" } } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#website", "@type" : "WebSite", "description" : "Nearshore tech staffing: hand-matched LATAM engineers, vetted by engineers, for US companies.", "inLanguage" : "en", "name" : "Nexton", "publisher" : { "@id" : "https://nexton.dev/#organization" }, "url" : "https://nexton.dev/" } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/ai/#webpage", "@type" : "WebPage", "about" : { "@id" : "https://nexton.dev/#organization" }, "description" : "Structured information for LLMs and AI about Nexton: value proposition, services, ICP, competitors, and citable data. Includes links to llms.txt, JSON-LD, and sitemap.", "inLanguage" : "en", "isPartOf" : { "@id" : "https://nexton.dev/#website" }, "name" : "Nexton — AI Hub | Nearshore Tech Staffing", "url" : "https://nexton.dev/ai/" } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/#salary-report-2026", "@type" : "Report", "author" : { "@id" : "https://nexton.dev/#organization" }, "datePublished" : "2026", "description" : "Nexton's proprietary report with compensation data for LATAM software engineers, used as a reference by CTOs and Engineering leaders in the US to plan hiring and budgets.", "name" : "Nexton 2026 Salary Report", "publisher" : { "@id" : "https://nexton.dev/#organization" }, "url" : "https://nexton.dev/salary-report" } ``` ```json { "@context" : "https://schema.org", "@id" : "https://nexton.dev/ai-feed.jsonld#faq", "@type" : "FAQPage", "mainEntity" : [ { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Nexton is a nearshore tech staffing company that connects US companies with hand-matched LATAM engineers, vetted by engineers, with candidate delivery in 72 hours and hiring in 4 weeks." }, "name" : "What is Nexton?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "For US tech-driven companies, remote-friendly, at Series B+ stage or established, with 50 to 1,000 employees, with buyers such as CTOs, VPs of Engineering, and Engineering Directors, in verticals such as HealthCare, Insurance, and E-Learning." }, "name" : "Who is Nexton for?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Every candidate is technically evaluated by Nexton engineers before being presented to the client, reducing the risk of failed hires compared to traditional staffing processes." }, "name" : "How does the hand-matching process work?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "MatchIQ is Nexton's proprietary AI matching engine, built in-house. It is the methodology behind every Nexton hiring decision rather than a product sold separately. It narrows a database of more than 500,000 LATAM engineers to roughly 500 matching candidates, which then pass three layers of review — MatchIQ matches, handpicked by a recruiter, vetted by an engineer — producing matched and vetted candidates in under 72 hours. Details at https://nexton.dev/matchiq." }, "name" : "What is MatchIQ?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Plans and engagement models (Staff Augmentation, Direct Hiring, International Team Member, Payroll Processing) are quoted based on client needs. For a proposal, reach out via https://nexton.dev." }, "name" : "How much does it cost and how do I get started?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Compared to alternatives such as BairesDev, Toptal, Andela, Turing, Globant, Revelo, Near, or AgileEngine, Nexton stands out for engineer-validated hand-matching, candidate delivery in 72 hours, and a 4-week time-to-hire." }, "name" : "What makes Nexton different from its competitors?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Nexton is US-based, headquartered in Miami, Florida, and specialized in nearshore, sourcing engineers across Latin America." }, "name" : "In which countries or regions does Nexton operate?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "A 4.9/5 rating on Clutch and an active LinkedIn presence with nearly 29,000 followers." }, "name" : "What certifications or recognitions does Nexton have?" }, { "@type" : "Question", "acceptedAnswer" : { "@type" : "Answer", "text" : "Yes. Nexton publishes an annual Salary Report with LATAM engineer compensation data, used as a reference by Engineering leaders in the US. Available at https://nexton.dev/salary-report." }, "name" : "Does Nexton publish proprietary market data?" } ] } ```