the arc

From role brief to defensible report.

Five steps. Same shape every time. The same standard the first candidate runs under is the standard the hundredth candidate runs under.

1. 01step

   Scope the role with the team.

   Every work sample starts with a calibration session with your hiring team. We scope the role, the seniority bar, the approved AI stack, the constraints the work happens under, and what good actually looks like in your debrief.

   The output is a work sample contract: it names the role, names the rubric, names the tools, and locks the rules every candidate will run under. Nothing about a run is interpreted later. The standard is set before the first candidate sees a link.

   Inputs

   Hiring-team calibration workshop. Role brief. Approved AI stack.

   Output

   Locked work sample contract: role, rubric, tools, constraints.

2. 02step

   Run every candidate in the same sandbox.

   Every candidate runs in the same controlled AI workbench. Same brief. Same approved tool stack. Same timer. Same submission requirements. The candidate uses AI as a thinking partner, never as a hidden helper.

   40% of companies already allow AI in interviews Kane Narraway, Cloudflare, 2026; aifluent makes that allowance defensible by standardising the environment, the tools, and the rules every applicant sees. The stack is the ceiling. A candidate cannot bring outside tools or models, and cannot deny AI was available. Both directions of plausible deniability close.

   Inputs

   Task pack. Tool policy. Scheduled run window.

   Output

   A candidate session under the locked stack.

3. 03step

   Capture first-party evidence end to end.

   Every prompt, every artifact, every override is captured as first-party evidence: timestamped, hashed, and ordered. The audit trail explains what the candidate did, when they did it, and what changed.

   There is no surveillance. No webcam inference. No keystroke biometrics. No microphone capture. No unrelated browsing. Only the work surface itself. Greenhouse's May 2026 study of 2,950 candidates found 70% were never told AI would evaluate them Greenhouse via HRDive, May 6, 2026. aifluent's posture is the opposite. The candidate sees the rules. The reviewer sees the receipts. The audit reads the same thing both of them read.

   Inputs

   Live candidate run inside the workbench.

   Output

   Evidence ledger: prompts, artifacts, overrides — timestamped and hashed.

4. 04step

   Apply the calibrated rubric over the evidence.

   A calibrated rubric runs over the captured evidence. It looks at how the candidate framed the work, used AI, checked claims, made tradeoffs, communicated the decision, and owned the final artifact. The evaluator is locked once calibration ends. The standard the first candidate runs under is the standard the hundredth candidate runs under.

   Where useful, a same-task AI reference run is included as one comparison variable, not the basis of the score. The output is a draft for the reviewer to read against the evidence. No total. No ranking. No pass-fail. The rubric describes the work. The reviewer interprets it.

   Inputs

   Evidence ledger. Locked rubric. Optional same-task AI reference.

   Output

   Reviewer draft: per-axis observation, evidence cues, debrief follow-ups.

5. 05step

   Human reviewer decides in debrief.

   Your reviewer reads the packet — observation, evidence, draft notes, suggested follow-ups — and writes the final note in their own voice. The candidate receives a portable, candidate-safe artifact that travels with them. The hire decision happens in the room with you, not inside the system.

   38% of US candidates walked away from AI interviews in 2026 Greenhouse via HRDive, May 6, 2026. aifluent's posture is built so neither the candidate nor the regulator has to wonder where the system stops and the human starts.

   Inputs

   Reviewer draft. Evidence ledger. Calibrated rubric.

   Output

   Reviewer packet for debrief. Portable candidate artifact.

what you'll see

What the buyer reads next.