Technical hiring stack for standardized engineering screening

This is the stack for the technical screen — the stretch between “the resume looks right” and “four engineers spent an afternoon on an onsite.” It assumes you already have recruiters and an ATS, and that your constraint is engineering hours, not applicant volume.

Two things changed the shape of that screen. Candidates now use AI assistants while they answer, so a test that measures unassisted recall measures a skill nobody will use on the job. And identity stopped being free: the U.S. Justice Department’s January 2025 indictment in the North Korean remote-IT-worker scheme found that over 300 U.S. companies, several of them Fortune 500, had unknowingly hired workers operating under false identities. A screening pipeline designed before either of those is scoring the wrong thing and verifying nothing.

The stack answers both with four layers: one async assessment platform that narrows the funnel cheaply — CodeSignal or HackerRank, not both — Karat for the live screen you do not staff, and Greenhouse as the identity gate and the structured record everything writes back to.

One thing changed since the last version of this page, and it changes a purchase. HackerRank’s July 2026 release moved AI Fluency scoring off chat logs and onto IDE telemetry, which means the async layer can now say something about how a candidate directs a model. That narrows the gap this stack used Karat to close, and it makes the assessment-only variation a live option for more teams than before rather than a downgrade.

If you want the whole hiring loop rather than the technical screen specifically, that is the AI-augmented recruiting stack. This one goes deep on one step.

How the pieces fit

• CodeSignal is the calibrated async filter. Its Coding Score is a 600-850 scale built to be comparable across candidates and across reqs, which is what turns a screen result into a number a hiring manager can argue with. AI Proctoring with ID verification ships on all three Hire tiers, Build included. The credit definition is now published and it is the aggressive one: a credit burns when a candidate starts an assessment or interview, not when they finish and not when you send the invite, so your bill tracks starts rather than completions. The buying detail that decides the tier: ATS integrations start at Grow, so the $79/mo entry tier writes nothing to Greenhouse and a recruiter copies results by hand.
• HackerRank is the same layer instrumented for AI use rather than against it. Candidates work in an agentic IDE and pick their model — Claude, Gemini, or GPT — and the AI Fluency evaluation reads IDE activity: which suggestions they accepted, rejected, or rewrote, surfaced through a Diff View of initial versus final code. Plan Mode has them shape an approach before writing anything. A set of 159 plan-build-review repository tasks drops candidates into production-style codebases — MERN, Spring Boot, Django, Go, .NET — to resolve support tickets. Chakra, its AI interviewer, runs conversational screens and opens an inline code editor mid-interview. Candidates with too few AI interactions get no AI Fluency grade at all rather than a low one, which is the right call and also means the score is missing on the quiet candidates you most want to read.
• Karat is the live technical screen you do not staff. Its network of trained Interview Engineers runs the 45-60 minute screen on your behalf, 24/7, and returns a scored written report against a fixed rubric; Karat reports over 600,000 interviews delivered. Its NextGen Interviews format, launched 10 December 2025, puts an AI assistant inside the interview on a multi-file project and has the Interview Engineer probe the candidate’s reasoning while they use it. That is a different instrument from HackerRank’s: telemetry records what a candidate did with a model, an Interview Engineer asks why and pushes back when the answer is thin. Karat now sells four lines — Direct Hiring, Partner Talent, AI Readiness and Hiring Insights — and only the first is this stack.
• Greenhouse is the system of record and the identity gate. Scorecards and interview kits are first-class, which is what makes assessment results comparable rather than filed. The plan ladder is now Core / Plus / Pro, and the verification layer ships as Real Talent, an add-on rather than a tier inclusion. Inside it, talent matching reaches Core, but fraud detection and identity verification with CLEAR are Plus and Pro only. Fraud detection runs a candidate’s phone, email, IP and location through IPQualityScore and returns 26 signals in three groups — high-risk, weak, and authenticity markers. CLEAR runs 60+ security signals including government ID authentication and a selfie check with liveness detection, inside the candidate’s MyGreenhouse profile, and CLEAR keeps the identity data rather than passing it to Greenhouse. Greenhouse MCP is in open beta on all three tiers, and Voice AI is sold outside the plan ladder entirely.

The handoffs that make it a stack

Each arrow is an event, not a habit.

An application lands in Greenhouse. Fraud detection scores it on application signals before a recruiter opens it, and high-risk signals route to manual review rather than to auto-reject — the signals are rules over phone, email, IP and location, not a model verdict on the person. On Pro those reports run automatically; on Plus somebody has to run them. That is the difference between a gate and an intention, and it is worth the Pro quote if your recruiting coordinators are already at capacity. Candidates who clear go to the async assessment: Greenhouse triggers the CodeSignal or HackerRank invite from the stage transition, and the returned score writes back onto the candidate record. A score above your cut line advances the candidate to Karat, which is booked from the same stage and returns a scored report and a pass/fail into Greenhouse within hours rather than after a scheduling round-trip. A Karat pass triggers the CLEAR check before the onsite is scheduled — the last cheap moment to confirm the person who passed the screen is the person who will show up. Candidates are marked verified for the stage where they completed it, and that flag is searchable, so a coordinator can filter the onsite list rather than remember. The onsite runs against Greenhouse scorecards, and stage-conversion reporting closes the loop: if 80% of candidates who clear the async assessment fail the Karat screen, the cut line is wrong, not the candidates.

Why this combination

Because a technical screen has three failure modes and no single vendor covers all three. An async test scales and, since July 2026, can report how a candidate used a model — but it reads behavior, not reasoning, and a candidate who accepts the right suggestions for the wrong reasons scores the same as one who understood them. A human interview reads that difference and costs engineering hours you are trying to protect. And neither one verifies identity — that is an ATS-layer job because it has to happen on the application record, before anyone spends a credit or an interview slot.

The load-bearing rule: one assessment vendor, one ATS, and the ATS owns the trigger. Running CodeSignal and HackerRank side by side gives you two incomparable scores on the same funnel and two renewals, and the Coding Score’s whole argument is that it is comparable. Triggering assessments from the assessment tool instead of from Greenhouse gives you candidates sitting in a queue that no ATS report can see. That is why the ATS tier matters more here than the assessment tier does.

What it costs

Budget $30K-$95K in year one and $25K-$70K at steady state for 40-80 engineering hires, with Karat as the dominant line and Greenhouse quoted on top.

• CodeSignal Hire publishes $79/mo billed annually ($99 monthly) for Build with 60 annual credits, and $479/mo billed annually ($599 monthly) for Grow with 420 annual credits and basic ATS integrations. Pro is quoted and carries advanced fraud prevention, enterprise ATS connections and role-based access control. Overage runs $20 per credit and annual credits do not roll into a new term. Vendr puts the median Pro contract at $24,394 across 62 purchases, in a range of $8,000 to $66,742.
• HackerRank publishes Starter at $79/mo billed annually ($948/yr, 60 attempts) or $99/mo monthly, and Pro at $419/mo billed annually ($5,028/yr, 360 attempts) or $529/mo monthly. Overage is $20 per attempt, or $15 pre-purchased on Pro annual. Enterprise is quoted and is the only tier with API access, the HackerRank MCP server, SSO/SCIM and the desktop proctoring app. Platform users are unlimited at every tier, so putting the whole panel on the submission costs nothing.
• The two entry tiers now cost the same money for the same volume — $948/year for 60 assessments, $15.80 a candidate, on either vendor. The second tier is nearly a tie too: CodeSignal Grow is $13.69 per credit and HackerRank Pro is $13.97 per attempt. Price is not the deciding variable at this size; what the tier carries is. HackerRank reaches ATS integration $720/year cheaper because it sits one tier lower on its ladder.
• Karat does not publish a rate card. Per-interview pricing runs roughly $350-$450 at low volume and $200-$280 above 2,000 interviews a year, with premium formats — system design, staff-level, NextGen — carrying a 20-40% premium. Vendr transaction data puts the median buyer contract at $175,695 a year, which tells you the shape of the customer base rather than the shape of your bill. Volume commitments are binding: commit to 500 interviews, use 300, pay for 500. Budget $5K-$25K for onboarding and rubric setup and 3-5% annual escalators.
• Greenhouse is quote-only across Core, Plus and Pro. Vendr’s data across 873 purchases puts the median contract at $26,587/year in a range of $10,137 to $74,492, with buyers negotiating an average 16% off the first quote. You have it already or this is the wrong stack — but price two line items, not one: the Plus tier and the Real Talent add-on. Neither is optional here, because Core with Real Talent gets you matching and no identity gate.

The break-even test is engineering hours. A screen you run internally costs an engineer roughly 90 minutes end to end including prep and debrief. At 300 screens a year that is 450 engineer-hours; at a $110/hr fully loaded cost, $49,500 of engineering time against roughly $105K of Karat at the $350 rate. Karat wins on hours only when your engineers are the constraint and their hours have somewhere better to go. If they don’t, run the live screen internally and buy the assessment platform alone for about $5K.

Common variations

• Assessment platform only, no Karat. Under 20 engineering hires a year the interview-as-a-service math does not close — you are paying a $200-$450 unit cost to protect hours you are not short of. This variation got stronger in July 2026: with AI Fluency reading IDE activity, HackerRank Pro at $5,028/year answers “can this person direct a model” well enough for many mid-market bars, and your engineers run the reasoning check on the shortlist. Revisit when a hiring manager starts declining interview requests, or when a bad hire traces back to a candidate who scored well and could not explain their own code.
• Swap Greenhouse for Ashby. Ashby’s native analytics answer the stage-conversion question this stack depends on without a reporting upgrade, and both HackerRank and CodeSignal integrate with it. Choose it when you are selecting an ATS now rather than replacing one — but confirm the identity-verification path first, because Real Talent and the CLEAR check are Greenhouse-specific and there is no equivalent add-on to buy.
• Karat only, no async layer. Works when your applicant volume is small and senior-heavy and every candidate deserves a live screen. It removes the calibrated score, so hiring managers argue from interview reports instead of numbers. Swap back the moment applicant volume per req passes about 50.
• CodeSignal to HackerRank on the instrument, not the price. The old reason to move — HackerRank’s published overage against CodeSignal’s undefined credit — is gone: both are $20 a unit and CodeSignal now states that a credit burns on start. Move because you want AI-direction telemetry in the async layer, or because you want ATS integration one tier down. Stay on CodeSignal when the 600-850 Coding Score is already the number your hiring managers argue from; replacing a calibrated scale mid-year costs you comparability across every req you have already run.

When this stack is the right pick — and when it isn’t

Pick it when you hire 20-200 engineers a year, your engineers are the scarce resource, at least some hiring is remote, and you already run structured hiring — structured interviewing is the precondition, not the output. It is also the right pick when someone has asked you to prove the person interviewed and the person hired are the same human, because the CLEAR and IPQualityScore layers are the only part of this stack that answers that.

Skip it in three cases. Under 20 engineering hires a year the Karat commitment and the Plus-plus-Real-Talent line both fail on unit economics — buy the assessment platform and nothing else. If your hiring is high-volume hourly or frontline rather than engineering, this is the wrong category entirely; that is the high-volume recruiting stack. And if your loop has no rubric — no defined cut line, no scorecard, no agreed bar — buying assessment adds a number nobody trusts. Fix the loop with the interview loop builder first.

What this stack does NOT replace

• Sourcing. Nothing here finds candidates. The pipeline this stack filters has to arrive from somewhere — inbound, or the AI sourcing stack.
• The onsite and the hiring decision. Karat returns a pass/fail on technical ability. Architecture depth, collaboration, and role fit stay in your loop, and the debrief stays human — see interview loop design.
• Background screening. CLEAR verifies that the applicant is the person on the ID. It does not check employment history, education, or criminal record; that is a separate purchase and a separate vendor.
• Your quality-of-hire measurement. This stack produces a screen score and a conversion rate. Whether the people you hired are good at the job is measured after they start, against performance data the ATS does not hold.
• Your obligations when AI scores a candidate. Chakra, the AI Fluency evaluation and CodeSignal’s AI Interviewers all rank people. Where AI-assisted evaluation feeds a hiring decision, disclosure, bias-audit and notice rules — NYC LL 144 and the EU AI Act among them — attach to the employer, not the vendor. Greenhouse states plainly that its fraud signals are rules over objective inputs and not a definitive assessment of a candidate’s authenticity, which is a deliberate design choice on their side and a fact you should be able to repeat to a regulator on yours.