Teams
Lock the Scorecards Before the Transcript and the Debrief
Interviewers should submit their scorecards before the debrief, and before reading anyone else's card, the transcript, or its automatic summary. The order does the work, not the deadline. Write for ten minutes straight after the call, file the card, then read the transcript and file a dated addendum if it corrects you, so the original card and the correction both stay in the file.
The takeThe 24-hour rule was written for a memory problem that no longer binds. Recording and transcription have made recall cheap, and the scarce thing now is an account of the round that nobody has already edited for you. A deadline records when a card arrived. It cannot see that the interviewer opened the summary first, which is how a card gets shaped by something other than the round, and it will pass every one of those cards without a flicker.
Where Olive fits
Open a role and see what the work shows
The six dimensions Olive reports on describe what capable AI work looks like on a team, and the report returns them as six written findings, each quoting the timestamped moment in the session it rests on. A human reviewer writes every word, and the candidate is granted the identical report, so the team and the candidate read one document.
Rank your shortlistWhen should a card be submitted?
Within about ten minutes of the call ending, and before three specific things: any other interviewer's card, the transcript, and the transcript's automatic summary. A card filed six hours later from unaided memory is worth more than one filed in twenty minutes after reading a summary, because only one of the two is that interviewer's own account of the round.
In practice that is a rule with three clauses, and it fits on a calendar invite:
1. Block five minutes after each round. Defend that block the way you would defend the interview itself, because it is part of the interview. 2. Write before you read. No cards, no transcript, no summary, no hallway conversation with the person who ran the previous round. 3. No silent edits after filing. Corrections go in a dated addendum naming what changed the view.
The third clause is the one teams drop, and it is the one that makes the other two auditable. A hiring file that quietly rewrites itself cannot show a later reader how the decision was reached, and it hides a specific pattern worth catching: the interviewer who always revises toward whoever spoke loudest in the debrief.
Why does the transcript have to wait?
Because the first account you take in shapes the one you write. That is the mechanism behind the independent-submission rule everyone already knows, and the transcript summary is now a second account, arriving early enough to be the one taken in first. It arrives fast, it is written in flat declarative sentences, and it carries no reminder that it is a selection from the round rather than the round itself.
The study underneath this does not show interviewers deciding in the first three minutes. In structured mock interviews with 189 accounting students, the interviewer's overall impression formed during the rapport-building small talk, before any structured question was asked, correlated .42 with that same interviewer's later structured score, falling to .25 and .24 when different interviewers supplied the later interview scores 1. That is a student sample in mock interviews, and much of the larger figure is one person's early read predicting their own later rating. What it does show is enough for the rule: what a person takes in first travels into what they record later.
A transcript summary sits in the same position, one step further along, with two extra advantages over an interviewer's memory. It is written down, so it feels like a record, and it has no visible author, so it feels like a finding. Whether a characterisation produced that way belongs in the hiring file at all is a live question, worked through in what to do when a notetaker's summary describes the candidate.
Read the cards before anyone speaks
Open the debrief by reading the submitted cards, most junior voice first, and discuss only the claims where two cards disagree. That sequence costs nothing to adopt and it removes the two ways a debrief fails: the senior person setting the frame in the first two minutes, and the meeting drifting into a shared impression nobody can point at afterwards.
The consensus meeting has a place in the evidence and it is not a flattering one. The meta-analysis comparing mechanical against holistic combination classifies group consensus meetings as a holistic method, on the grounds that what makes a method holistic is combining data by judgment, insight or intuition rather than by a rule applied the same way for each decision, and it derives a 25% reduction in correct hiring decisions at a selection ratio of .30 from the validity difference 2. That 25% is a derivation from meta-analytic validities rather than an outcome measured at an employer, and the review never tested consensus debriefs against averaged independent scores head to head.
It is also not an argument for cancelling the meeting, and a reader who takes it that way has inverted it. A debrief is how evidence gets surfaced, challenged and corrected. What the classification catches is the final step, when the combining is a conversation ending in a shared feeling instead of a stated rule. Keep the meeting and move the combining out of it. Who runs that meeting, and whether they hold a vote in it, is a separate decision taken in who chairs the debrief and whether the recruiter votes.
What about a stage nobody watched?
Name it out loud before the discussion starts, and say what it did and did not show. A round with no human observer produces no card to lock, so nothing in the submission rule protects it, and a room that is not told will treat its output as though somebody watched. That is how an unexamined result acquires authority in a hiring decision.
Independent rating is rarer than the vocabulary around it suggests. A content analysis of 104 structured interviews, drawn from the studies published between 1997 and 2010, found an average of 5.74 of the 15 recognised components of interview structure in use, and the component barring discussion of applicants or answers between interviews was described in 4% of them 3. That counts what researchers wrote into published studies rather than what employers do, so read it as a ceiling rather than a floor. A practice that thin in the literature is not a solved problem in the field, whatever the vendor guides say.
Getting a panel to look for the same things in the first place is its own job, and it comes before any of this, in getting an interview panel to judge AI use the same way.
On Monday, add one field to the form: what had you read before writing this. Four values cover it, which are nothing, another card, the transcript, and the summary. Check it next week against the timestamps on the notetaker's emails, and count how many cards were written after the summary went out. If the answer is most of them, the deadline was never the control.
Common questions
How long after the interview should a card be filed?
Ten minutes, and the number matters less than the sequence. Write before you talk to anyone, before you open the transcript, and before the summary email arrives. If the loop runs back to back, block five minutes between rounds specifically for this and treat that block as part of the interview rather than as paperwork. A card filed late from unaided memory beats a fast one written from somebody else's account of the same conversation.
Can an interviewer change a card after the debrief?
Not the original. File an addendum instead, dated, naming what changed the view and what evidence produced the change. A hiring file that rewrites itself cannot show a later reader how a decision was reached, and the person most likely to want that history is the one who wrote it. An addendum also exposes a pattern worth seeing: the interviewer whose revisions always run toward the loudest person in the room.
Is the transcript useless?
The transcript is useful in second position and only there. Read it after filing, and use it for what memory is genuinely bad at: exact wording, the order things were said in, and whether you asked the question you think you asked. If it corrects you, file an addendum. What it should not supply is the emphasis, because a summary is a selection from the round and it does not arrive labelled as one.
What if only two people interviewed the candidate?
Independent submission gets cheaper, not less necessary. Two cards written independently, with the junior one read first, still stop the more senior of the two from setting the frame, which is the entire mechanism. In a two-person loop that ordering discipline carries more of the weight, because there is nobody else in the room to notice the drift, and the junior person has fewer ways to recover the ground once it is set.
Does this apply to the recruiter screen?
Yes, and the recruiter screen is the round where independent submission is easiest to skip. A screening call produces the first written account of a candidate, and everyone after it reads that account before meeting the person. Filing it before the hiring manager weighs in costs nothing and makes the screen's claims checkable later against what the panel actually saw. It is also a round where a summary can end up standing in for a person's notes.
References
- 1. Initial Evaluations in the Interview: Relationships with Subsequent Interviewer Evaluations and Employment Offers homepages.se.edu Supports the claim that an impression formed before the structured questions began tracked the same interviewer's later structured score (.42, falling to .25 across interviewers), stated without the decides-in-three-minutes reading the paper does not support.
- 2. Mechanical Versus Clinical Data Combination in Selection and Admissions Decisions: A Meta-Analysis gwern.net Supports the claim that a group consensus meeting is classified as holistic combination, and that the 25% figure is a Taylor-Russell derivation at a .30 selection ratio rather than a measured employer outcome.
- 3. The Structured Employment Interview: Narrative and Quantitative Review of the Research Literature doi.org Supports the claim that the component barring discussion of applicants or answers between interviews was described in 4% of the 104 interviews analysed, against an average of 5.74 of 15 structure components in use.
3 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.