Hire or Pass
Debriefs go wrong when the loudest interviewer wins or when everyone waits to see what the manager thinks. This skill forces the evidence into a fixed shape — per competency, strongest example and strongest counter-example, from whom — and defines the four possible outcomes before the discussion starts. The output is a distribution over those outcomes and, when it is close, the specific extra signal to go get.
Boundary. This is decision support. A person makes and owns the hiring decision, applies the company's process and the law where they are, and is accountable for it. Never let a distribution replace that.
Where this sits
After hiring-rubric (which defines the competencies) and engineering-hiring-rubric, and after the interviews. This one runs the debrief decision.
What This Skill Produces
- The evidence state — per must-have competency: strongest example, strongest counter-example, which interviewer saw it, rubric scores
- The decision —
hire / hire_at_different_level / more_signal / pass with probabilities and confidence
- Act or hold — thresholds applied; a close hire/pass is a
more_signal by definition
- The signal plan — when holding: the competency to test, the format, who runs it, by when
- The debrief record — one paragraph for the file: evidence summary, decision, who decided
Required Inputs
Ask for these; do not infer from "vibes" or an interviewer's overall thumbs:
- role — title, level, the 3–5 must-have competencies from the rubric
- evidence — per competency: strongest observed example and strongest counter-example, with the interviewer who saw it
- scores — rubric scores per interviewer per competency
- references — if taken: confirmed or contradicted what
- risks — gaps, level mismatch, red flags with the evidence behind them
- process — any missed interviews or untested competencies
Framework: Four Defined Outcomes
Thresholds. Act when confidence ≥ 0.65 and the winning option's probability ≥ 0.7. A close hire / pass is a more_signal outcome: name the competency and who tests it. Never decide on the distribution alone.
Evidence, not impressions. "Strong communicator" is an impression. "Explained the outage timeline to a non-engineer in the panel without being asked — seen by R." is evidence. Only evidence enters the state.
Getting probabilities. Any calibrated decision model that returns a probability per defined option serves this contract; the repo ships an adapter (node integrations/jev/decide.mjs hire-or-pass --state state.json, contract in integrations/jev/decisions/). Without one, the assistant estimates and labels it. Either way the decision is the hiring manager's.
Output Format
Debrief decision: [role, level] · candidate [initials]
Evidence by competency
Decision
Confidence: 0.xx · Source: [model / estimate]
Verdict: [act: option] or [hold → more_signal: competency, format, owner, by when]
Debrief record: [one paragraph — evidence summary, the decision, who made it]
Quality Checks
Anti-Patterns
- Overall thumbs replacing per-competency evidence
- Anchoring on the manager — collect scores before anyone speaks
- "Culture fit" as an undefined competency
- Deciding a coin flip — 0.5 / 0.5 hire/pass is a
more_signal, always
- Letting the model hire — it produces evidence about evidence; a person decides
Example Trigger Phrases
- "Run the debrief — three interviewers, split panel."
- "Hire or pass on this candidate? Here are the scorecards."
- "We're at hire vs. pass 50/50 — what do we do?"