Scorecards
Decisions you can defend.
A scorecard is the write-up each interviewer files after a round. Here the criteria live on the round, nobody reads another card before filing their own, and a filed card cannot be changed — so two people who sat the same interview rated the same things, independently, and the record of it lasts.
macOS app, Windows next · free for ten interviews a month, no card
Three rules, all enforced where they cannot be bent.
Each of them is checked by the server, not asked of the interface, so a card looks the same whether it was filed in the app or came in from your ATS.
- The criteria live on the round
- Not on the interviewer, not on the candidate. Two people who sat the same round rated the same things, and two candidates who sat it are comparable.
- Nobody reads another card before filing their own
- Anchoring on the first opinion in the room is the failure structured interviewing exists to prevent. Until you have filed, the panel’s cards are not shown to you — enforced on the server, not hidden in the interface.
- A filed card cannot be changed by the person who filed it
- Otherwise the first rule is decoration: file a placeholder, read everyone else, revise. A filed card is the record; the server refuses the edit and points to an admin.
Criteria that ask the right shape of question.
Each criterion on a round has a name, guidance for the interviewer, and a kind — so a yes-or-no question is not forced into a one-to-four, and a template imported from your ATS keeps the shape it had there. Mark one required and a card cannot be filed without it.
- A rating
- One to four, no middle — so a verdict is a verdict.
- Yes or no
- Did they do the thing.
- A choice
- From options you wrote, or that came in with an imported template.
- Free text
- Where the evidence is a sentence. What they did, not how they seemed.
The write-up that is missing is chased for you.
After the hours you set — eighteen by default, or two, or two days — the interviewers who have not filed get an email with the form and the transcript on the same screen, and your queue shows who still owes one. Nobody writes “gentle reminder”.
The chase, like every reminder here, goes out once however many times the clock runs, and never inside the quiet hours you set for the person receiving it.
Then the debrief: disagreement first.
One candidate’s whole loop on one page — every round, each interviewer’s verdict beside the others, the write-ups still owed counted rather than guessed at. The rounds where the panel split are shown first, because agreement needs no meeting and the only thing worth anyone’s time is the round two people saw differently. Reading it is free and needs no paid seat.
Then a summary of every round, and a score against the job.
Every finished interview gets a summary drafted from its record — the transcript, the timeline of what was run, what the candidate produced and the cards filed so far — with the record quoted beside each finding rather than adjectives about the person, and a Regenerate button for when the next card comes in. Above the rounds sits the candidate’s overall assessment: each requirement of the job rated on one scale, an overall out of 100, and beside it how much of the job was assessed at all.
The job is a list of requirements — drafted from the role’s brief when it has one, written by hand when it does not — each marked must or nice, weighted three and one unless you change them, and kept in history like the criteria. A criterion on a round can point at the requirement it is evidence for, which is how an interviewer’s mark reaches the score.
An interviewer’s mark comes first. Where the round’s criteria point at a requirement, the filed marks decide its rating and the names sit beside it; the assistant’s reading of the transcript is used only where nobody marked, and the table says which it was. The overall is the weighted mean of the requirements that were assessed, scaled so that below the bar everywhere is 0 and exceeds everywhere is 100. Coverage is the share of the job’s weight that was assessed at all, and it is kept apart on purpose: a 90 on a third of the job reads as 90 with 33% coverage, and a requirement nobody assessed says so in its row and is never counted as a zero.
| Rating | What it says | On its own, out of 100 |
|---|---|---|
| not assessed | Nobody rated a criterion linked to it and nothing in the record spoke to it. | Nothing. A gap in coverage, never a zero in the score. |
| below the bar | The floor of the scale. A no on a yes-or-no question lands here. | 0 |
| partly meets | Some of it, not the whole. | 33 |
| meets | Did the thing. A yes on a yes-or-no question lands here. | 67 |
| exceeds | More than the round asked for. A 4 on a one-to-four rating, or a 5 on a five-point template imported from your ATS. | 100 |
Two interviewers who disagree are shown disagreeing. When two marks on the same criterion sit two or more steps apart, both are listed with their names, the row is flagged, and the middle of them is the number used — the flag is the finding, not the number.
Everything the assistant wrote says so, above it: “Drafted by the assistant from the transcript and 2 scorecards”, “Drafted by the assistant from 3 round summaries and 5 scorecards”. Every quote in the table is checked against the record and dropped if it is not there. Where an interviewer marked, the number is arithmetic on the marks and the model never touches it; where only the transcript speaks, the rating is the assistant’s own and the row says so. The prose is written last, from the finished table, so regenerating it cannot move a mark-based score.
Whoever may read the debrief reads it — owner, admin, recruiter, and the hiring manager on the free viewer seat. Only an owner, admin or recruiter can ask for a draft, because drafting reads every card and the transcript and spends the organisation’s model budget. A round’s summary is its cards in other words, so it follows the same rule as the cards: nobody who has not filed their own card for a round sees its draft. The Security seat sees none of it. The draft is the one place a transcript is handed to a model; where it goes, and the setting that turns it off, are on the security page.
Connected to your ATS.
Scorecards already written in your ATS — Greenhouse, Lever, Ashby, Workable or SmartRecruiters — come in with the candidate, filed against the interview they were written for where the ATS names one and otherwise on the stage, and matched to the criteria on the round by the ATS’s own id or an exact name — an attribute with no match is kept, never dropped, and a “mixed” that has no place on a one-to-four scale keeps its label rather than being invented as a number. From Workable and SmartRecruiters a card is a rating and a note, with no marks per criterion, because neither vendor exposes them. Cards written here stay here until write-back exists, and that is said here rather than found out later. What comes in from each ATS is on the integrations page.
The verdict is the same five points every ATS uses under different names — strong no, no, mixed, yes, strong yes — so a verdict survives the trip in without being invented or lost.
The last thing that should ever expire.
Structured criteria, filed independently by named assessors, are the evidence that a hiring decision was made on the job. So when retention is built — it is designed, per organisation and per class of data, with a legal hold, and not yet built — scorecards are what it keeps and interview content is what it expires. Every change to a round’s criteria is already kept as a version, so “this candidate was assessed under these criteria” is answerable later.
Two candidates, one round, the same questions.
Ten interviews a month, free, on the whole product. Scorecards and the debrief are on every plan; the hiring manager reads them on a free seat.