Before anything else: candidate data
Sharpen your rubric first
Every interview is implicitly evaluating something. The “something” needs to be explicit, written down, and shared across the loop. If your team doesn’t have a rubric, write one before running this workflow — the artifacts depend on it. A minimum-viable rubric has four parts:
The last row is the most underrated and the highest-leverage. A gap that closes in three months on the job is a different hiring decision from a gap that doesn’t close at all — and most loops conflate them. Be explicit: “weak system design at senior level = barrier. Weak presentation skill = coachable in role.” Force the distinction.
Also decide before opening the role: which competencies each interview probes. Don’t have five interviewers all asking behavioral questions about ownership. Allocate competencies so each is probed by at least two interviewers (for calibration) but not all of them (for efficiency).
What the artifact looks like
A worked example of one competency from a competency interview — a senior engineer being evaluated on system design:The competency interview template
Save it as a workspace template
1
Start from the people-and-team interview template, or write from scratch
There is no built-in interview template, so the fastest path is to paste the prompt above into a new task on a real interview and iterate in the Composer until the output lands.
2
Customize per interview type
A behavioral interview probes different competencies than a system design interview. Save a separate workspace template for each interview type in your loop, with the competency list pre-filled.
3
Save with Workspace visibility
Set visibility to Workspace so every interviewer in the loop produces the same shape. See Custom templates. Calibration depends on shared structure.
Run it on a single interview
1
Confirm consent at the start of the call
“I’m using Earmark to take notes so I can be present. Is that okay?” If the candidate declines, switch to handwritten notes for this interview.
2
Pre-seed with the right template
For the interview type and the specific competencies being evaluated. See Before a meeting.
3
Be present
The point of the workflow is to preserve presence during the interview. Take no manual notes. Engage. Probe follow-ups. Ask the questions a real conversation produces.
4
Five-minute cleanup
Within 30 minutes — interview signal decays fast:
- Strip vague language. “Candidate seemed engaged” → “candidate asked clarifying questions on three of the five scenarios.” If the model wrote “good energy,” cut it.
- Sharpen the Coachable/Barrier reads. For every gap the model flagged, confirm the call yourself. This is your judgment, not the model’s.
- Write the hire recommendation yourself. Don’t let the model write this. The model can summarize evidence; the recommendation is your call. Letting the model write it moves the most important judgment in hiring to a system that wasn’t in the room.
- Tighten the scorecard-ready notes. This is what lands in the ATS.
5
Push to ATS — same day
Within 24 hours at the outermost; ideally within the hour. The longer the gap between interview and ATS push, the more memory bias creeps in. Paste the Scorecard-ready notes into the ATS scorecard, ratings into the competency fields, risks into the notes-for-next-interviewer field. Earmark does not write to ATS systems directly — the artifact is structured so it pastes cleanly.
Variations
Same skeleton, two close relatives that handle the bookends of the loop.Recruiter or hiring-manager screen
For the 30–45 minute conversation that decides whether to invest the full loop. Lighter than a competency interview, broader in scope.Reference check
Reference checks are underutilized and underdeveloped at most companies. A structured one is often the single highest-signal artifact in the loop.The calibration debrief
The debrief is where decisions actually get made. It is also one of the most underutilized meetings in hiring — typically run on fading memories, dominated by whoever speaks first or loudest. The structure that works:- Hiring manager frames briefly — 60 seconds on the candidate and the loop’s coverage. Not the recommendation yet.
- Each interviewer offers their independent recommendation before discussion. The single most important debiasing move. Some panels do this in writing simultaneously. Without it, the first speaker anchors the room.
- Walk through the divergences. Where the loop disagreed, dig in. What did the dissenting interviewer see? What did they probe that others didn’t?
- Probe every Coachable/Barrier read. For each gap the loop flagged, the panel discusses: do we agree it’s coachable? Coachable by whom, and on what timeline?
- Hiring manager makes the call with the loop’s input weighted appropriately. Calls that contradict the loop’s aggregate read are fine — sometimes the hiring manager sees something the loop doesn’t — but they should be made explicitly and with reasoning, not implicitly.
What this workflow doesn’t do
Earmark refines artifacts within a single interview. Three things hiring teams want — loop-level synthesis across multiple interviewers, post-hire retrospectives comparing interview reads to actual performance, and rubric refinement based on patterns across past loops — all require synthesis across many meetings. Not a one-click action today. Workarounds:- Loop-level synthesis before a calibration debrief. Paste each interviewer’s Scorecard-ready notes and gap reads into a single Customize-context document. Run a synthesis task that surfaces per-competency aggregate reads, divergences across interviewers, and remaining open questions. Fifteen minutes; produces the substrate the debrief actually needs.
- Post-hire retrospectives at 6 and 12 months. The structured original artifacts are still in the workspace. Pair them with the new hire’s actual performance reads — compiled manually from recent 1:1 artifacts using the people and team meetings workflow — and run an external synthesis on the pair. What did the loop predict? What’s actually happening? Which coachable gaps closed?
- Rubric refinement. Quarterly, the talent lead or hiring manager reviews the post-hire retrospectives by hand and updates the rubric, the question bank, and the coachable-vs-barrier list. The rubric becomes a living document over years.
- ATS push is paste-based, not integrated. The structured Scorecard-ready notes section is designed to drop into Greenhouse, Lever, Ashby, or Workable cleanly. Workspace templates ensure every interviewer produces the same shape.
Common pitfalls
- Vague qualifiers in scorecards. “Smart, good fit, didn’t vibe.” If the artifact contains these without specific behavioral evidence, the workflow has failed. Non-negotiable: evidence, not impression.
- Filling out the scorecard the next day. Memory decay starts immediately. Same-day push, ideally same-hour. The longer the gap, the more bias creeps in.
- Letting the model write the hire recommendation. The model can summarize evidence. The recommendation is your judgment.
- Conflating coachable gap with barrier. The most common hiring mistake. Force the explicit call.
- No calibration across the loop. Different interviewers using different competencies and different language. The loop’s value depends on shared structure. Fix the rubric before standing up the workflow.
- Skipping loop-level synthesis before the debrief. A debrief without the synthesis is the same debrief everyone’s been running for decades, with a recording added. Run the synthesis first, every time.
- Anchoring in the debrief. The first speaker tends to anchor the room. Mitigate by collecting individual recommendations before discussion — in writing if possible.
- Reference checks as box-checking. A poorly-run reference check is worse than no reference check — it generates false confidence. Probe for specifics; pay attention to hesitations.
- No close-the-loop habit. Without post-hire retrospectives, the rubric never improves. Calendar them; do them.
- Treating interview signal as homogeneous. Different competencies are observable to different degrees in interview settings. Technical depth: very observable. Long-term resilience: barely observable. Weight the synthesis accordingly.
- Querying interview artifacts for ad-hoc Q&A. Interview artifacts are restricted access. Not part of company-wide ad-hoc Q&A corpora. Keep them inside the hiring loop.
- Retention drift. It’s easy to keep candidate artifacts longer than your retention policy. Audit annually. Delete what should be deleted.
- “Culture fit” as a competency. If your rubric has “culture fit” as a dimension with no specific behavioral definition, it’s a vector for bias. Replace with specific behaviors (“communicates disagreement directly,” “shares credit,” “asks for help when stuck”) that match what you actually mean.
Where to go next
- Workflows — the general shape this is an instance of
- People and team meetings workflow — for the broader manager-relationship workflows including 1:1s and skip-levels
- Temporary meetings — for sessions with tighter retention than the workspace default
- Security and privacy — workspace isolation, no training on your data, retention controls
- Custom templates — visibility, sharing, and edit permissions
- Before a meeting — pre-seeding the right template for the interview type
- Composer — for tuning the prompt before saving as a workspace template
- Ad-hoc Q&A workflow — for the privacy discipline around restricted-access corpora like interviews