Interview feedback that later hires can audit

Interview feedback that later hires can audit is a structured, dated record of who said what about a candidate, against which job, with a decision attached. It is remarks and ratings a human typed, not a chat opinion that evaporates. This page owns that record. It does not define applicant-tracking software — that stays on what an ATS is. It does not design pipeline columns as candidate states — that stays on hiring pipeline stages. Week-by-week hiring operations stay on the recruitment hiring checklist. Product context is Dayzen recruitment, which includes an interview remark and rating portal for evaluators. It does not rank or score candidates with AI.

If feedback lives only in chat, a calendar note, or memory, you cannot reconstruct why someone advanced or was declined. Rehires and “we already met this person” need a trail, not a vibe.

What “auditable” actually means

Audit here is operational, not a courtroom script. A later recruiter, HR lead, or hiring manager should be able to open the application and answer:

• Which job this feedback belongs to (requisition, not a loose name).
• Which interview event it belongs to (date, round label you actually use, who was in the room).
• Who wrote the remarks, as a named evaluator, not “the panel.”
• What they observed against the role’s must-haves, in words.
• What rating or recommendation they submitted, if you use one.
• What decision the organisation recorded after that evidence (advance, another named round, hold with a date, or reject).
• When those fields were written, so a recap written two weeks later is visible as late.

That is the minimum unit. Adjectives with no owner, a chat emoji, or a private score never filed are not units. If you cannot point at the record, you have folklore, not feedback.

Do not confuse this with AI ranking. Ordering a list by last updated or a number a human typed is still a human process. Dayzen’s evaluator portal captures remarks and ratings. It does not choose the hire. Humans move the application and own the decision.

Structured notes beat freeform novels

Freeform “thoughts after the call” produce three problems: interviewers write about different things, they describe the person they liked instead of the work, and they bury the recommendation so recruiters must interpret tone.

A workable structure is short and role-shaped. Reuse the same headings for every interviewer on that requisition:

1. Role fit against must-haves — evidence for or against the skills the job description named, not a new secret list.
2. Work sample or problem discussed — what they actually did in the conversation, not “good communication.”
3. Risks or unknowns — what you still do not know, with a suggestion for the next round if there is one.
4. Recommendation — advance, reject, or hold, in a word the pipeline already understands.
5. Rating, if you use one — a shared scale (for example 1–5 on named dimensions), submitted in the same portal as the remarks.

Keep the scale the same across interviewers for that job. If one person rates “culture” out of 10 and another rates “technical” out of 5, you have two private languages. Publish dimensions on the requisition before the first interview. Changing the scale mid-loop makes older notes incomparable. Structured does not mean robotic: two or three sentences per heading is enough. No transcripts, no character sketches. If a comment cannot be tied to the job, it does not belong in the hiring record.

Consistency is a process, not a personality trait

Inconsistent feedback is usually a design failure. Typical causes:

• No shared headings, so each interviewer invents a genre.
• No shared “what good looks like” for the role, so ratings float.
• Some interviewers file in the portal and others mail a recap that never gets attached.
• Panel interviews with one spokesperson and four silent attendees who later disagree in a hallway.
• A hiring manager who overwrites interviewer notes with a summary that erases dissent.

Fix the design. Require every named evaluator to submit remarks before the debrief. The debrief can still produce one organisational decision; it should not pretend four people said the same thing. Preserve individual records and add a short decision note that cites them. If one interviewer is strongly negative and the group advances anyway, that fact should survive.

Keep the same bar for the same job. Do not raise it for a later, more charismatic candidate and retrofit older notes. If the bar changed, write it as a requisition change and tell people already in interview. When the spec moves, old evidence needs a dated context, not a quiet rewrite.

Ownership: who may write, who may see, who may decide

Name three hats, even if one person wears two of them in a small company:

• Evaluator — writes remarks and ratings for interviews they ran. They own accuracy of what they observed. They do not silently edit someone else’s notes.
• Recruiter of record — makes sure the event exists, the right people are invited, and feedback is filed before the stage can exit. They chase missing notes. They do not ghost-write a rating for a manager who skipped the form.
• Decision owner — usually the hiring manager for that requisition, sometimes with a named approver. They record advance/reject/hold after reading the filed notes. They may add a decision rationale. They do not delete evaluator remarks to make the file look unanimous.

Access should follow that split. Interviewers see the candidate and job they assessed. Recruiters see the loop. Hiring managers see their requisition. Do not broadcast raw notes company-wide “for transparency.” If an interviewer is recused (they referred the candidate, or they will be a peer with a conflict), write the recusal so later readers do not assume independence.

The decision record is separate from the remarks

Remarks are evidence. The decision is an organisational act. Mixing them is how “I had a good chat” becomes an accidental offer. After feedback is in, the decision record should state:

• The outcome for this application: advance, reject, hold until a date, or withdraw if the candidate left.
• Who decided, and when.
• Which feedback IDs or evaluator names they relied on.
• If they overrode a recommendation, one sentence why — not a novel, not silence.

Pipeline stages still hang on the application. This article does not redesign those columns. Do not advance on a verbal “we liked them” while the portal is empty. Empty notes plus an advanced stage is a lying board. The stages article owns what a column means. This page owns the evidence that should exist before you leave Interview.

A hold is not a missing decision. If you wait on another round, an assignment, or the candidate’s notice, write the wait and a review date. A hold without an owner is not a professional delay. Closing aged applications as rejected is a sibling topic.

Where the record should live

Put remarks on the application (or the interview event tied to it) in the recruitment system you run — not in a parallel spreadsheet, not only in mail. Attachments can support notes; they should not replace structured fields. A recording nobody will replay still needs a written summary the next interviewer can read in five minutes.

Dayzen’s interview remark and rating portal is that filing surface. Debrief from it. Copying notes into a slide deck that becomes the “real” discussion is how deck and record diverge. If leadership wants a shortlist summary, generate it from filed notes. Do not add a scoring model “to help ranking.” Dayzen does not sort candidates with AI. If you use a human rating, keep dimensions visible. Similar ratings mean read the remarks: the number is a shorthand, not a model verdict.

What later readers actually do with the trail

Six months later, typical uses:

• Reapplication: the same person applies again. You should see prior rounds and whether the gap they failed has changed — without pretending last year’s interview is this year’s.
• Internal challenge: a hiring manager asks why a referred candidate was declined. You can show filed reasons against the JD, not a shrug.
• Handoff between recruiters: someone goes on leave. The replacement should not restart Interview from zero because notes were in a private inbox.
• Calibration: two interviewers routinely disagree. You can see the pattern in writing and coach the process, not the rumour mill.

None of that requires invented “quality of hire” statistics. It requires that the file exists. An HR review of hiring files will look for the same objects: owner, date, job, evidence, decision. This page is not legal advice on what those reviews must contain. It is the habit that makes a review possible.

Failure modes that look like hiring speed

Symptom	What is missing	Repair
Offer after one “great call”	Filed remarks and a decision note	Block Offer until the portal has the evaluator record
Two interviewers, one paragraph	Individual ownership	Each evaluator submits; debrief cites both
Notes appear after reject email	Timestamp honesty	Do not backfill as if they were contemporaneous
Chat consensus, empty ATS	System of record	Paste is not filing; submit in the portal
“AI can rank the panel scores”	A product you should not claim	Humans read remarks; Dayzen does not AI-rank

Speed that skips the record is a loan you repay when someone asks why. The ATS guide still owns the category of system. The pipeline article still owns stage names. The hiring checklist still owns how you run a structured interview as an ops step. Use this page when the argument is whether later you can reconstruct the judgement. File remarks. Name the owner. Date the decision. Leave ranking to people, not to a model you should not advertise.