July Week 2 · Goodfit

Read every integrity signal for an assessment in one report with one verdict, catch candidates running an AI copilot in their browser, hand every candidate a different mix of coding questions, see the full history of who changed what on a job, and verify a candidate's ID straight from the government source with DigiLocker.

Proctoring

One integrity report, one verdict, one place to look

Everything we watch during an assessment — the camera, the browser, the answers, the voice — used to live in two separate panels with their own language for "suspicious". Now every signal lands in a single report with a single verdict.

Each assessment attempt carries one Integrity card in its header — HARPA AI extension, AI-written answer (95%) + 35 more flags. Open it for the full report.
The report opens with the verdict in words, not a score to decode — and when a face doesn't match, the two photos sit side by side at the top so you can judge it yourself in a second.
A timeline of the whole session shows every flag on the moment it happened, labelled — Earbuds, Looking away ×2 — and for voice interviews, tap any moment to jump the recording straight to it.
A What we checked grid lists all eight checks for that assessment type, so you see what passed as clearly as what didn't. Nothing wrong? You get "All 8 checks passed" instead of a wall of empty tables.
Below that, everything with its evidence attached: AI-written answers with the exact quoted passage, voice-to-lip match, the flagged camera frames, and a tidy event log you can filter to browser, camera, microphone, or tab.

An Integrity card on an assessment header — a meter bar, a risk label, and a one-line worst concern naming the HARPA AI extension — One card per attempt — the verdict and the single worst thing we found, before you open anything.

The Integrity Report opened on a coding attempt, with a colored verdict band, a plain-English summary, and a row of headline finding chips — The verdict in a sentence, and the findings worth your attention right under it.

A session timeline with a suspicion curve, colored flag dots on the flagged moments, and labelled lanes underneath for each signal — The whole session on one timeline — every flag on the second it happened.

An Integrity Report for a clean attempt showing an All 8 checks passed hero instead of empty tables — When a candidate is clean, the report says so plainly — no hunting for the absence of bad news.

Proctoring

We now catch AI copilots running in the candidate's browser

A candidate can install Monica, HARPA, Sider or the like and have an AI assistant answer questions in a floating panel — same tab, no tab switch, nothing for the old checks to see. We now catch it.

If a candidate had a known AI assistant open during their assessment, it turns up in the Integrity Report by name and logo — nothing for you to switch on, and nothing the candidate is told.
We separate used from merely installed: an assistant that was open and answering is a red, high-risk flag; one that was installed and never opened is an amber "installed, no visible use" — worth knowing, not an accusation.
Every flag comes with the evidence behind it — how long the assistant was on screen, and what it was showing while the candidate worked.
Covers all four assessment types — coding, skill-based, psychometric, and voice interviews.

A replay frame from a candidate's coding assessment with the Monica AI panel open next to the question, mid-answer — Exactly what we're catching — an AI panel open beside the question, in the candidate's own recording.

An Integrity Report event log with a Monica AI row naming the extension and the minutes it was visible, above a group of paste attempts — The extension named, timed, and sitting right next to the paste attempts it explains.

A What we checked grid on a coding attempt where every check passed except the browser one, which names the HARPA AI extension — The case this was built for — an otherwise spotless attempt with one AI assistant running through it.

Assessments

Give every candidate a different coding test from the same set

Identical question sets leak — one candidate finishes, the questions land in a WhatsApp group, and the next twenty walk in prepared. Now you can attach a bigger set and hand each candidate a random slice of it.

Split your questions into Fixed — everyone gets these — and a Pool everyone draws from. Click the pill on any question card to move it between the two.
Set how many to draw from each difficulty: 1 easy, 2 medium, 1 hard. Every candidate gets the same shape of test, just not the same questions.
A summary bar spells it out as you build: "Each candidate receives 6 questions: 2 fixed + 1 easy + 2 medium + 1 hard from the pool". No mental arithmetic.
Attach up to 24 questions, deliver up to 6 — and reshape the pool right up until a candidate starts; your changes count for everyone yet to begin.
Scoring is unchanged — the cutoff is still "solve X of the questions delivered", and it now counts against what the candidate actually got, not everything you attached.

A coding assessment builder with a Fixed section and a Pool section below it, each pooled difficulty having a stepper for how many questions to draw — Fixed questions on top, the pool underneath, and one stepper per difficulty.

A tooltip on the Pool section header explaining that each candidate receives a random draw from these questions — Hover anything you're unsure about — the builder explains itself as you go.

Jobs

See who changed what on a job, and when

Someone dropped the cutoff on the coding round, or a stage vanished from the pipeline, and nobody could say who or when. Every job now has an Activity page that reads like a sentence.

Open it from the clock icon on the job's applications board — a plain-English timeline of every change, newest first.
Pipeline changes read like people talk: "Lowered cutoff 2→1 on Clawbot Coding Assessment", "Added 'Web Interview' as Stage 6", "Moved 'CEFR English' to Stage 3". Open any row for the stage-by-stage detail.
Job changes are there too — published, closed, made public or private, details edited, duplicated.
Every row names the person who did it and the exact time, alongside a friendly "10 min ago".
Filter to Pipeline or Details & visibility when you're chasing one kind of change.

A job Activity page showing a timeline of changes, each with who made it, how long ago, and an expandable stage-by-stage detail — Every change to the job, in a sentence, with a name and a time against it.

The job applications board action strip with a new clock icon labelled Job Activity — One clock icon on the board you're already looking at.

Proctoring

Verify a candidate's ID straight from the government source with DigiLocker

Our photo check proves the candidate's face and name match the ID they're holding up — but not that the ID itself is real. DigiLocker closes that gap: the photo and name come from UIDAI, so the document can't be forged.

On the ID step, the candidate picks Connect DigiLocker (recommended) or Take a photo of your ID instead — the fallback stays for anyone who doesn't use it.
They give consent in DigiLocker, and we receive their Aadhaar photo and name from the government source. We match the photo against the selfie they just took and the name against the applicant on file.
A pass sends them straight into the interview. A mismatch is flagged for you to review — same as today.
You see it in the same verification panel you already use: a via DigiLocker badge and the photo labelled as pulled from DigiLocker.

Opt-in, and we keep almost nothing
Off until you turn it on — per organisation, then per job. Nothing changes on your existing jobs. We store only the Aadhaar photo (encrypted) and the name we matched against; never the Aadhaar number, date of birth, or any DigiLocker login.