CLUELYSAFE

Interview integrity  /  Remote first-round screens  /  Est. 2026

Your candidate has a second screen you cannot see.

For Heads of Talent Acquisition, Directors of Recruiting, and the VP of Engineering who personally runs the technical screen at a 50–1,000 person company — and who is not going to stop hiring remotely this quarter.

Said up front, not in a footnote: no behavioral tell in this category has peer-reviewed validation. Not mine, not anyone’s. So this kit is built so you never accuse anyone — you change the interview format instead, and the integrity note never touches the hire/no-hire math.
Get the kit — $499 Book a 30-min scoping call

Card checkout, in your name, on Stripe. All eight files arrive by email within one business day — there is no instant download, and I would rather say that here than have you refresh an empty inbox. Need a PO or net terms? Use “Request an invoice” instead.

FIG. 01

Latency, then fluency.

Response latency trace  ·  30 s window  ·  illustrative, not measured Prompt: “walk me through the migration you led”
Response latency trace comparing an unassisted answer with an assisted answer An illustrative drawing, not a recording or a measurement study. Two lanes over a thirty second timeline. Lane A, an unassisted candidate, starts speaking almost immediately with ragged amplitude, seven separate speech runs, six audible pauses and four self-corrections. Lane B, the same question routed through a hidden assistant, shows roughly four seconds of complete dead air, then one uninterrupted block of even-amplitude speech for about 25 seconds with zero self-corrections. The pairing of the dead-air gap with the unbroken fluency that follows is the tell. TRC-01 / LANE A:B t0 = QUESTION END SRC: OPERATOR OBSERVATION — NOT PEER-REVIEWED DRAWN TO SCALE — NOT MEASURED DATA A B UNASSISTED — SAME QUESTION ROUTED THROUGH A HIDDEN OVERLAY uh— no, sorry, 2022 let me back up I got that wrong 7 SPEECH RUNS · 6 PAUSES · 4 SELF-CORRECTIONS ~4 s DEAD AIR — NO FILLER, NO BREATH, NO “UM” 1 SPEECH RUN · 0 PAUSES > 400 ms · 0 SELF-CORRECTIONS THE TELL IS THE PAIRING. A long silence is nerves. Fluency is competence. Dead air immediately followed by 25 s of unbroken, evenly-paced speech is neither. 0s 5s 10s 15s 20s 25s 30s TIME AFTER QUESTION ENDS

Scroll the figure sideways →

Fig. 01 — Response latency trace. Illustrative reconstruction of a pattern experienced interviewers describe, drawn to scale on a real 30-second axis. It is an operator observation, not a measurement study. What it teaches is the only thing worth teaching here: a single signal proves nothing, and a pairing of signals is what you actually probe next. The kit gives you the 60-second confirming probe for each one.
6% of 3,000 job candidates in Gartner’s 2Q25 survey admitted to interview fraud — posing as someone else, or having someone pose as them. Gartner predicts one in four candidate profiles worldwide will be fake by 2028. Source: Gartner
72.4% of recruiting leaders now conduct interviews in person to combat candidate fraud. Google, Cisco and McKinsey have reintroduced in-person rounds; Gartner’s Emi Chiba says at least half of applications for remote IT roles may be fraudulent. Source: Computerworld / Gartner
$149.99 per month is Cluely’s listed “Pro + Undetectability” plan, which explicitly includes remaining hidden during screen sharing. Pro is $19.99/mo. Free clones on GitHub advertise the same thing for $0. Source: Cluely pricing
38.5% of candidates were flagged for cheating behavior across 19,368 interviews on one vendor’s platform, July 2025–January 2026 — 48% in technical roles vs 12% in sales. Vendor-reported platform data, not peer-reviewed research. Treat it as directional. Source: Fabric (vendor data)
02 / THE MOMENT

The hiring manager put it in writing.

A candidate who crushed the remote technical screen collapsed at the onsite — or in week one. Now the hiring manager has asked you, in Slack, in writing, what your interview process actually does about real-time AI assistance.

There is no answer, because there is no process. The next remote panel is on the calendar inside ten days and nobody wants to run it blind. Or: an exec forwarded the Cluely coverage into the hiring channel with “are we exposed to this?” and it landed on your desk with no budget cycle attached.

Which is exactly why the answer has to cost under $500, arrive today as files in your inbox, and be something you can hand a panel on Tuesday — not a vendor demo in Q4.

The unsolved problem is not detection. It is that a hiring manager suspects something and nobody in the room knows what they are procedurally or legally allowed to do next. That is a documentation problem. This kit is the documentation.

Fig. 02 — Tells matrix  ·  instrument panel 12 tells  /  4 classes  /  each with a benign explanation
Four-quadrant matrix of the twelve observable tells A two-axis instrument panel. The horizontal axis is signal durability, from short-lived to structural. The vertical axis is false-positive risk, from low to high. Four quadrants hold the four tell classes: gaze and eye-line behaviour (three tells, highest false-positive risk, lowest durability), lexical fingerprinting (three tells), latency before fluency (three tells), and answer shape (three tells, most structural and lowest false-positive risk). Each quadrant lists its tells and the benign explanation that is explicitly not a tell. MTX-02 READ WITH THE PROTOCOL, NOT ALONE SIGNAL DURABILITY  →  STRUCTURAL (SURVIVES THE NEXT TOOL) FALSE-POSITIVE RISK  →  HIGH CLASS 01 · HIGHEST FALSE-POSITIVE RISK GAZE & EYE-LINE 01  Sustained off-camera fixation, one axis 02  Left-to-right sweep on every answer 03  Eye-line returns on a fixed cadence NOT A TELL: a second monitor with the JD open, notes on paper, ADHD, or a bad webcam angle. CLASS 02 · DRIFTS AS MODELS CHANGE LEXICAL FINGERPRINT 04  Register jumps mid-answer, then returns 05  Spoken enumeration: “three key factors” 06  Textbook terms absent from the résumé NOT A TELL: a rehearsed candidate, an ESL speaker with formal training, or someone who preps hard. CLASS 03 · SEE FIG. 01 LATENCY → FLUENCY 07  Consistent gap before every question 08  Silence with no filler, breath or restart 09  Zero self-correction across a long answer NOT A TELL: a deliberate thinker, a slow connection, or a candidate who was taught to pause first. CLASS 04 · MOST DURABLE, LOWEST FP RISK ANSWER SHAPE 10  Accepts a false premise you planted 11  Cannot supply a number from own résumé 12  Sanitised failure story, no discovery moment This quadrant is where the question bank lives. It fails next year for the same reason it fails today. OBSERVE — NEVER ACT ON ALONE CORROBORATE BUILD THE INTERVIEW AROUND THIS PROBE, DO NOT SCORE

Scroll the figure sideways →

Every tell in the protocol ships as four fields: what it looks like, the benign explanation that is not a tell, the 60-second confirming probe you run next, and the false-positive risk. Each is labelled vendor-observed or operator judgment. Nothing here is presented as science, because none of it is.
03 / MANIFEST

Eight files. Here is every one.

Page counts, question counts, file formats. If it reads like an operations deliverable rather than an ebook, that is because it is one — and because you deserve to know exactly what you are buying before you buy it.

01

Live-Interview Signal Protocol

Twelve observable tells across four classes: gaze and eye-line, latency-before-fluency, answer-shape, and lexical fingerprinting. Each tell is written as four fields — what it looks like, the benign explanation that is not a tell (nerves, ESL, thinking style, bad webcam), the 60-second confirming probe you run next, and the false-positive risk. Every tell is labelled vendor-observed or operator judgment, and the first 200 words state plainly that no peer-reviewed validation exists for any of them.

PDF · 16 pages12 tells
02

AI-Resistant Question Bank

Sixty questions across four tracks — software engineering, data/analytics, ops & program management, customer-facing — built on five mechanisms that make real-time model assist actively hurt the candidate: (1) résumé-anchored questions about specifics no model can know, (2) questions containing a deliberately false premise a model will confidently accept, (3) interrupt-and-redirect at the 20-second mark, (4) “what did you get wrong and how did you find out” failure narratives, (5) artifact walkthroughs of a document the candidate wrote. Each question ships with three columns: the human answer shape, the AI-assisted answer shape, and the follow-up that splits them.

Google Sheet + CSVGreenhouse / Lever / Ashby import60 questions · 4 tracks
03

Integrity-Separated Scoring Rubric

Five competence dimensions on 1–4 behavioural anchors, plus a structurally separate integrity observation column that is recorded independently and never mathematically blends into the hire/no-hire score. Panel-aggregation formulas are pre-built. The competence decision and the integrity concern stay two different records — which is the design your legal team will ask for.

XLSX5 dimensions · 1–4 anchors
04

Candidate-Facing AI-Use Policy Pack

Three drop-in policy versions — AI permitted with disclosure / permitted for prep but not live / prohibited during live rounds — each roughly 150 words. Each version ships with the scheduling-email paragraph, a 40-word verbal script the interviewer reads at minute zero, and the one-line acknowledgment field to add to your ATS application form.

DOCX + plain text3 versions
05

The 20-Minute Controlled Re-Interview

The runbook for the format you switch to when a screen goes sideways. Structured, artifact-based, and built so that any real-time overlay produces a worse answer than no overlay. Includes the invite copy, the minute-by-minute timing sheet, the three question slots, and the pass/fail read. Schedulable inside 24 hours with no new software.

PDF runbook20 min · 3 slots
06

Escalation & Documentation Script

Six pages for the moment you suspect. The non-accusatory live pivot (“I want to change format for the next fifteen minutes”), the two-independent-signals rule before any action is taken, a timestamped observation log template, debrief language that keeps the panel out of defamation exposure, and the two decision paths: re-interview under controlled format, or decline on competence grounds without ever stating cheating as the reason.

PDF · 6 pages2 decision paths
07

Panel One-Pager

A single printable page for the interviewer’s second monitor: the 12 tells, the confirming probes, and the pivot script. Designed to be readable during a live call, at a glance, without scrolling.

PDF · 1 page, printable
08

Twelve months of updates, emailed

When a new overlay tool ships or a tell stops working, you get the revised files at no cost. I send them from my own address. There is no portal to log into, no renewal, and no seat to provision.

12 months includedEmail, no portal
Fig. 03 — Sample page: lexical fingerprint markup Constructed example — not a real candidate
Constructed transcript example with model-phrasing giveaways highlighted A written-for-this-page transcript example, not a recording of any real candidate, with the invented specifics covered by solid bars so only the phrasing is readable. Four phrases are boxed in amber and annotated in the right margin: an enumerated preamble, a hedged balancing clause, a summary closer, and a textbook term that never appears on the candidate's own resume. A note at the bottom states that phrasing alone is not evidence and must be paired with a second independent signal. TX-03 / CONSTRUCTED EXAMPLE — NO REAL CANDIDATE HIGHLIGHT = PHRASING ONLY, NOT EVIDENCE Q  Tell me about the outage in Q3. A  Sure. So meaningfully improved reliability across the stack. and the rollback path was already documented. On balance but the blast radius stayed contained. was implemented To summarise the postmortem closed the same week. there are three key factors On balance, was implemented To summarise, MARGIN NOTES Spoken enumeration. People rarely count out loud unprompted. Written-register hedge, spoken. Passive voice about work they claim they personally did. Summary closer on a 40-second answer nobody asked to summarise. RULE: phrasing is a prompt to probe, never a finding. Two independent signals are required before any action — and the only action authorised is changing the format.

Scroll the figure sideways →

This transcript is written, not recorded. No real candidate, no real interview, nothing redacted from anyone — the bars are there so you read the phrasing and nothing else. What is real is the page shape: the margin-note column and the two-independent-signals rule at the bottom appear on every markup page in the kit.
04 / PRICE

One price. Whole team. Forever.

Tier 1 · Interview Integrity Kit

$499

One-time · not a subscription

  • All eight assets, emailed within one business day of payment.
  • Unlimited internal use across your whole company. No seat count.
  • No candidate minimum. No contract. No renewal.
  • Twelve months of updates, emailed directly.
  • Card checkout or invoice. Files arrive by email within one business day.
Buy the kit — $499 Request an invoice

Card checkout is live. Files arrive by email within one business day — not instantly. If your team needs a PO or net terms, request an invoice instead. It is priced at $499 rather than $549 to sit under a $500 discretionary-approval limit, if your team has one. At 200 interviews a year that is $2.50 an interview in year one, and $0 after.

Tier 2 · Team Calibration

$1,950

One-time · live, 90 minutes, up to 8 interviewers

  • The $499 kit included at no additional cost.
  • Ten questions written against your two open reqs before the call.
  • Live walkthrough of two of your recorded or role-played screens.
  • Written 2-page calibration memo inside 48 hours.
Capacity: I take four of these a month. Not scarcity marketing — I am one person and I run these myself.
Book the 30-min scoping call Pay & schedule — $1,950 Request an invoice

Order of operations: 30-minute scoping call → payment → session scheduled. Paying by card sends you straight to my calendar afterwards. The call is the filter. If the kit alone is enough for you, I will say so on the call and you will not spend $1,950.

Tier 3 · Detector — in development

$0

Not available. Nothing to install. There is no extension, no download, no screenshot, and no ship date. If and when it exists, I will email the people who asked. That is the entire offer.

Email me if it ships

04.1 · What the alternatives cost, at list price

Every figure below was taken from the vendor’s own published page. Links included so you can check me.
Line itemPriceUnitSource
ClueLySafe Interview Integrity Kit $499 One-time, entire company, unlimited interviews This page
SHRM Talent Acquisition Specialty Credential $1,830 – $2,105 Per person (member / non-member) AIHR
AIRS recruiter certifications $795 – $995 Per person (bundle $1,695) AIHR
AIHR Sourcing & Recruitment Certificate $1,125 Per person AIHR
Tech Recruiter Certification 2.0 — cheapest in AIHR’s roundup $595 Per person, one-time AIHR
Automated proctoring (market range) $3 – $15 Per attempt — forever, per candidate Proctor365
Live / hybrid proctoring (Proctorio ~$5, Honorlock ~$12, ProctorU ~$20 first hour) $15 – $35 Per session Proctor365
Talview Quote only Unit-based, ~150–200 candidate minimum to start Talview
Cluely — the tool on the other side of the table $19.99 – $149.99 Per month; the $149.99 plan buys invisibility during screen share Cluely
Free Cluely clones on GitHub (OpenCluely, Vysper, Pluely) $0 Self-described as invisible to screen sharing and recording. OpenCluely and Vysper are open source; Pluely now ships as closed binaries with a free tier GitHub
05 / TIMELINE

The honest timeline.

No onboarding. No implementation project. No portal. This is what actually happens, hour by hour.

Fig. 04 — Deployment rail T+0 → Month 12
Deployment timeline from purchase to month twelve A horizontal rail with six amber event markers. At T plus zero you pay and all eight files arrive by email within one business day. At T plus thirty minutes the panel one-pager is printed and on the second monitor. At T plus two hours the policy paragraph is pasted into the scheduling template and the ATS acknowledgment field. At T plus one day the question bank CSV is imported into your applicant tracking system and the controlled re-interview is schedulable. At T plus three days the first panel runs with the new format. From month one through month twelve, revised files arrive by email whenever a tell stops working. RAIL-04 NO IMPLEMENTATION PROJECT T+0 Pay. All 8 files emailed same day. T+30 min Panel one-pager printed, on the second monitor. T+2 hrs Policy paragraph into scheduling email + ATS field. T+1 day CSV imported to Greenhouse / Lever / Ashby. Controlled re-interview is now schedulable. T+3 days First panel runs the new format. Month 1–12 Revised files arrive by email when a tell stops working. NOT INCLUDED: software to install, a seat to provision, a security review to pass, or a call you have to take before you get the files.

Scroll the figure sideways →

The only thing standing between purchase and use is the time it takes to read a 16-page PDF and print one page. That is deliberate: the buyer who needs this has a panel on the calendar inside ten days.
06 / STRAIGHT ANSWERS

Ask me the hard ones.

Seven objections, answered in full. Nothing is collapsed behind an accordion, because hiding the hard answer is how you end up with a refund request.

Q.01

It’s a PDF and a spreadsheet. Why is it $499?

Because of what it replaces, not what it weighs. The cheapest per-person recruiting credential in AIHR’s roundup is $595 and the SHRM Talent Acquisition credential is $1,830 for members, $2,105 for non-members (AIHR) — each certifies one person and none of them cover real-time AI assist. Proctoring runs $3–$15 per attempt automated and $15–$35 live (Proctor365), and Talview publicly states a 150–200 candidate minimum just to get started (Talview).

The kit is $499 once. No seat count, no candidate minimum, no contract, and you use it on every interview your team runs forever. At 200 interviews a year that is $2.50 an interview in year one and free after that.

Q.02

These ‘tells’ sound like junk science. If I flag a candidate on eye movement and I’m wrong, I have a legal problem.

Correct, and that is exactly why the kit is built the way it is. There is no peer-reviewed validation for any behavioral tell in this category — I say that in the first 200 words of the protocol, and I label every tell as either vendor-observed or my own operator judgment.

So the kit is designed so that you never accuse anyone. Integrity observations are recorded in a structurally separate column that never enters the hire/no-hire math. The escalation script requires two independent signals before you do anything, and the only action it authorises is changing the interview format — not making a claim. The decline path never states cheating as the reason. That is the piece your legal team will actually read, and the piece I have not seen shipped anywhere else in this category.

Q.03

Why wouldn’t I just buy a detection tool instead?

Because you cannot blocklist software you cannot enumerate. Cluely sells a $149.99/month plan whose stated feature is remaining hidden during screen sharing (Cluely pricing), and there are free clones on GitHub — OpenCluely, Vysper, Pluely — that describe themselves as undetectable in video calls, screen shares and recordings (GitHub).

Detection tooling is a signature race against an attacker set with zero marginal cost to fork. The question bank is not a signature race: a question anchored to a specific number on the candidate’s own résumé, or built on a false premise a model will confidently accept, fails for the same structural reason next year that it fails today. Buy detection later if you want. Buy something that works during Tuesday’s panel now.

Q.04

Google, Cisco and McKinsey just went back to in-person. Shouldn’t we do that instead?

For final rounds, probably yes — Gartner data reported by Computerworld shows 72.4% of recruiting leaders are already conducting interviews in person to combat candidate fraud (Computerworld).

But you are not flying candidates in for a 45-minute first screen, and you are not going to stop running remote screens this quarter. In-person is a control for the last round. This kit is the control for the four rounds before it — which is where the funnel volume and the exposure actually sit.

Q.05

I’ve never heard of you. Where are the customer logos and the case studies?

There are none, because this launched this month and I will not invent them. What I will do instead is be specific: exact page counts, exact question counts, exact file formats, exact payment terms, and an exact cap on how many live sessions I take.

Here is my actual record — I spent 20+ years on enterprise delivery at Fortune 500s, and I currently run enterprise systems and automation at a trucking company. I have also personally built the kind of AI tooling this kit defends against, which is why the countermeasures target the assist channel and not a specific binary.

Q.06

Won’t this be obsolete the next time a new overlay tool ships?

The tells will drift. The question mechanisms will not. A model cannot know the headcount you managed in 2022, it will confidently accept a false premise you plant in the question, and it produces a sanitised failure narrative when you ask what you got wrong and how you found out. None of that depends on which binary is running on the candidate’s second machine.

You also get 12 months of updates by email — when a tell stops working I revise the file and send it. No portal, no login, no renewal.

Q.07

Our talent team already has a structured rubric and an interview guide.

Then skip the rubric and use the question bank, the escalation script, and the 20-minute controlled re-interview — those are the three assets that do not exist in a standard interview guide, because standard guides were written before real-time assist existed.

If you want a straight answer on whether the kit adds anything to what you already have, book the calibration call and I will tell you on the call, before you buy.

07 / OPERATOR

I build the tooling this kit defends against.

I’m Daniel Graham. I bootstrapped AgentHive Inc. in February 2025 — no outside capital, no staff, no investor timeline. Before that I spent 20+ years on enterprise delivery at Fortune 500s, which is where I learned that the hard part of any control is never the detection. It is the paperwork that lets a human act on it without creating a second problem.

Today I’m Manager of Enterprise Systems & Automation at a trucking company, running Copilot Studio agents and Trimble, B2W and Samsara integrations in production. In that role I drove an SSO rollout to 76.6% completion. That is a real number off a real dashboard, not a rounded marketing figure — if it had finished I would have written 100%.

I know precisely how a real-time assist channel is wired, because I have wired them. That is why this kit targets the channel rather than chasing a specific binary, and it is why I am comfortable telling you that the behavioral tells are the weakest part of the product and the escalation script is the strongest.

No logos on this page. No testimonials. Nothing borrowed. If that costs me the sale, it costs me the sale.

Talk to me for 30 minutes Email me a question

Verifiable record

  • 20+ yrsenterprise delivery, Fortune 500
  • 76.6%SSO rollout completion, real deployment number
  • Feb 2025AgentHive Inc. bootstrapped, no outside capital
  • 4 / molive calibration sessions, hard cap, because I am one person

Specificity in place of social proof

  • 16page protocol
  • 12tells, each with its benign explanation
  • 60questions across 4 tracks
  • 3policy versions, ~150 words each
  • 6page escalation script
  • 1business day to delivery

One more thing, on the record: in March 2026 Cluely’s CEO publicly admitted he had lied about the company’s revenue numbers (TechCrunch). I include that not as a dunk but as a warning that cuts both ways — claims from vendors on both sides of this market need checking, including mine. Every number on this page links to its source. Cluely began as Interview Coder, built by a Columbia student who was suspended over it and used it to obtain an Amazon internship; it raised $5.3M in April 2025 (TechCrunch) and a further $15M from a16z in June 2025 (TechCrunch). That is the funding behind the thing sitting on the other side of your next screen share.

08 / DECIDE

You cannot block software you cannot enumerate.
You can build an interview it fails.

$499, one-time. Eight files, in your inbox within one business day. Your whole team, every interview, no expiry. Fourteen days to change your mind — and you keep the files either way.

Get the kit — $499 Or ask me first — 30 min

Before you buy, confirm all six

  • You run remote first-round screens and intend to keep doing it.
  • Someone has asked, in writing, what your process does about AI assist.
  • You have a panel on the calendar inside the next ten days.
  • You would rather change the format than accuse a candidate.
  • You can approve $499 without opening a procurement cycle.
  • You are fine with a product that tells you which parts of it are weak.

Six for six — buy it. Five or fewer — book the call and let me talk you out of it.