R Rubric

Interview Reports

Reports are how Rubric closes the loop. After a session ends, the app drafts a polished interview report from the scores and notes, opens it for review, and lets you edit before exporting.

How a report is generated

When you click End Session, Rubric packages the session state — final rubric scores, descriptors, every timestamped note, the candidate’s name, the elapsed time — and sends it to the configured AI provider with a structured prompt. The provider returns a Markdown report which Rubric persists alongside the session.

This means three things:

  1. You need an API key configured for auto-drafting. Without one, the session still saves; you can come back and trigger a draft later, or write the report by hand.
  2. The notes you took during the session matter. The model has access only to what you wrote and what you scored. Good notes produce good reports.
  3. The first draft is a draft. Treat it as the model’s best guess at what your scores meant. Always re-read it before sharing.

Report structure

A standard report has six sections:

  1. Executive summary — three to five sentences. The “what should I tell the hiring manager” version.
  2. Score breakdown — each rubric category, the score, and a short paragraph explaining the score grounded in your notes.
  3. Key observations — the moments from the round that mattered, with timestamps.
  4. Strengths — what the candidate did well, framed for performance feedback.
  5. Growth areas — what they didn’t do well, framed constructively (and honestly).
  6. Recommendation — Hire / No Hire / Follow Up, with the reasoning.

The Markdown source is fully editable on the Report screen. Edits are saved as you type.

The hiring recommendation

The model returns one of three values:

  • Hire — the candidate cleared the bar. The exec summary says so plainly and the score breakdown supports it.
  • No Hire — the candidate didn’t clear the bar. Same expectations, opposite direction.
  • Follow Up — the round didn’t produce enough signal. Usually because the candidate ran out of time, the round got derailed by an environmental issue, or one strong dimension and one weak dimension cancel. A Follow Up recommends a second round to resolve the ambiguity, often focused on the unclear dimension.

You can override the recommendation. The override stays in the saved report; the model’s original is preserved in the report metadata.

Editing a report

The Report screen renders the Markdown live and lets you edit the source directly. Common edits:

  • Tighten or sharpen the executive summary so it reads in your voice, not the model’s.
  • Add a sentence that the notes don’t capture (e.g., “the candidate was clearly thrown by the technical issues at the start; consider weighting that lightly”).
  • Soften or remove a phrasing you wouldn’t say to the candidate’s face.

Avoid editing scores. If a score was wrong, go back to the session, fix it there, and regenerate the report — that keeps the score and the prose in sync.

Re-generating a report

If the first draft missed the mark, you can regenerate. Reasons to regenerate:

  • You added or fixed notes after the session ended.
  • You changed a rubric score.
  • You changed the configured model in Settings and want to compare drafts.

Click Regenerate. The previous draft is preserved in the report’s revision history; you can compare or revert.

Exporting

A finished report can be exported in two formats:

  • Markdown — the raw .md file, paste-ready into a hiring tool, ATS, or shared doc.
  • PDF — a styled PDF with the letterpress brand applied, suitable for archiving or printing.

Reports are also persisted in your configured storage backend (see Storage Backends). Local filesystem stores them as .md next to the session JSON. The git backend can be configured to commit each finished report. The S3 backend writes them to the configured bucket.

Privacy

The session payload (scores, notes, candidate name, elapsed time) is sent to the configured AI provider when drafting. It is not sent anywhere else, and Rubric itself does not host or proxy any of this — the call goes from your machine directly to Anthropic or OpenAI. If you need an interview kept entirely off third-party servers, leave the API key blank and write the report by hand.