The evidence layer for AI interviews

What the layer records today

• Signed turn ledger. Each interviewer question and candidate answer is persisted server-side as a sequenced turn and signed over a canonical serialization, so a transcript's origin and ordering can be verified after the fact -- not reconstructed from a browser blob.
• Evidence packets. Interviews resolve into a recruiter-reviewed packet: summary, scorecard, transcript excerpts, per-question timings, flagged risks, and unresolved questions. The packet is the decision artifact humans review; scores never advance a candidate on their own.
• Governed rubrics. Roles interview against versioned, content-hashed rubrics with an approval workflow. A launched interview pins the exact rubric and question-set version it ran against, so the evidence names the standard it was measured by.
• Retention, holds, and erasure. Evidence ages out on configurable retention windows, legal holds pause destruction per company or candidate, and candidate erasure is transactional with a zero-remaining verification -- a partial erasure cannot exist.

Integration surface

The layer is opening up in deliberate stages. Statuses below are current and honest -- nothing is sold before it is real.

• Workspace exports (Recording today). Evidence packets and analytics leave the product through recruiter-facing exports today.
• Evidence specification (Recording today). The evidence model -- turns, signatures, packets, lifecycle -- is published so partners can evaluate it before any API commitment.
• Evidence API (Self-serve (Growth & Enterprise)). Programmatic access to packets, ledger verification, interview invitations, and signed webhooks. Workspace admins provision credentials from Settings.
• Interview API and Embed (Self-serve (Growth & Enterprise)). Interview-invitation links can be minted from partner systems today. Embedding review surfaces remains a separate default-off experiment.
• Rubric review embed (Experiment, default off). Review-only rubric embedding behind a separate default-off gate. Not a published self-serve contract.

Why an evidence layer

Hiring decisions made with AI increasingly need to be explainable after the fact: what was asked, what was answered, what standard it was scored against, and who reviewed it. A product can show a score; a layer can prove where the score came from. Klevian is built layer-first so that proof exists for every interview, whether it is reviewed in our product or, eventually, consumed through an API.