# The Post-Round Interview Record Standard, Graded Complete, Patch, or Redo

*Grade any single interview round's notes against eight stated criteria and reach the same Complete, Patch, or Redo verdict a second grader would.*

- Canonical URL: https://www.refolk.ai/candidates/guides/post-round-interview-record-standard-2
- Pillar: Interviewing
- Format: Standard
- Published: 2026-11-19
- Last reviewed: 2026-11-19
- Reading time: 23 min
- Keywords: what to write down after a job interview, interview notes template for candidates, how to keep track of multiple interview processes, can I record my job interview, keep my story consistent across interviews

## Key takeaways

- People recall only about 10% of a conversation right after it ends and about 4% a month later, so an interview record has to be written the same day, before any other call.
- After a 3 to 4 day delay, people recalled 6% of their own statements but only 3% of what the other person said, so a candidate loses the employer's claims and promises first.
- Trained investigators taking verbatim notes left out 57.3% of their own questions, which is why a candidate's record needs a quoted stem for every interviewer question.
- A round's grade is the lowest grade any of its eight criteria gets: one Redo makes the whole round Redo, and a gap you can only fill from memory gets marked UNKNOWN, never guessed.
- Federal law lets a party to a call record it, but sources count 9 to 13 all-party consent states, and a bot's name in the attendee list does not count as consent.
- Refolk's index counts 21,158 Technical Recruiters and 6,572 Recruiting Coordinators in the United States, a record-keeping layer of about 27,700 people that candidates face without a system.

This standard is for candidates who need a record of each interview round that they can trust weeks later, especially when several interview processes are running at once. It sets out eight criteria for one round's notes and a grading rule that returns Complete, Patch, or Redo. It also gives the same-day procedure and checklist that get a record to Complete.

## What a post-round record is, and why it needs a pass bar

A post-round record is a written account of one interview round, made right after it ends. It covers who was there, what they asked, every number spoken, what the employer committed to, and how the notes were captured. It needs a pass bar because the employer is already keeping a disciplined record of the same conversation, and a loose reflection page can't match it.

Two terms used throughout:

- **Round**: one scheduled conversation with the employer, anything from a recruiter screen to a single panel slot.
- **Loop**: the full sequence of rounds at one employer for one role.

The employer-side bar is public. The US Office of Personnel Management's structured interview guide says the interviewer's name and job title, the interview questions, and the scores and notes for each candidate must be recorded. It adds that notes should include actual behavioral examples so the ratings can be defended. A state guide built on it adds the date, time, place, and length of the interview. Your record should mirror the fields that matter to you: interviewer name and title, date and length, the questions, and the concrete examples you gave. Those examples are the evidence a panel scores.

A large professional layer keeps those records.

| Title | Country | Count | Exact-title share in sample |
|---|---|---|---|
| Technical Recruiter | US | 21,158 | 21/25 = 84% |
| Technical Recruiter | UK | 701 | 20/25 = 80% |
| Recruiting Coordinator | US | 6,572 | 18/25 = 72% |

**21,158 - People titled Technical Recruiter in the United States, in Refolk's index of professional profiles**

Add 6,572 US Recruiting Coordinators and the employer side has roughly 27,700 people whose job includes logging what candidates say.

The derived ratios show how the work is divided. There are 30.2 times as many US Technical Recruiters as UK ones, and 3.2 US Technical Recruiters for every US Recruiting Coordinator. The share column shows how many profiles in a 25-profile sample carried the title exactly. Treat it as a rough quality check on the count, not a census.

The candidate-side pages that rank for "interview notes template for candidates" are reflection checklists. They give you prompts, but none of them sets a time limit, a pass bar, or a consent check. This standard adds all three, and each round ends with one of three grades:

- **Complete**: all eight criteria pass. Use the record as it is.
- **Patch**: at least one gap, and every gap can be closed today from a document or another person.
- **Redo**: at least one gap that only your memory could fill, or a failed gate criterion.

## How fast interview memory fades

Fast, and unevenly. People recall about 10% of a conversation right after it ends and about 4% a month later. Exact wording, proper names, and the other person's words go first.

| Delay | Finding | Study |
|---|---|---|
| Immediate | About 10% recalled | Stafford et al. |
| Short delay, unprompted | 20% or less of details | Interview recall review |
| 2 days | Verbatim memory still present | Verbatim memory study |
| 3 to 4 days | 6% own, 3% other's statements | Ross and Sicoly, via Brown-Schmidt and Benjamin |
| 5 days | Verbatim memory greatly reduced | Verbatim memory study |
| 1 month | About 4% recalled | Stafford et al. |

These studies measure recall in different ways, so compare the direction across rows, not the exact values. Three findings shape the criteria below:

- **Names fade faster than ideas.** Recall and recognition of proper names declined faster than recall of concepts. A name you are confident about is not evidence that you have it right.
- **The other side's words fade first.** After 3 to 4 days, people recalled 6% of their own statements and 3% of what the other person said. In an interview, the other person is the employer, so their claims and promises are the first things you lose.
- **Writing early works.** In an NSF-funded study, written recall done immediately was more complete and more accurate than oral recall a week later.

Where the evidence stops: no one has published an hour-by-hour forgetting curve for interviews. The 15-minute window in this standard comes from practitioner guidance. Pin's interviewer guidance says to write up within 15 minutes. Jobshinobi recommends a 10-minute debrief immediately after. Jobfoxly says to do it straight after the interview, before anything else. Treat 15 minutes as a bright line two graders can check against a timestamp, not a threshold the research proves. The research does support writing the same day, before another conversation competes for the same memory. Nothing in it supports waiting until day 2.

## The eight criteria and how to grade them

Grade each of the eight criteria Pass, Patch, or Redo using the triggers below, then give the round the lowest grade any criterion got. The triggers are written so that two graders looking at the same record and the same calendar reach the same result.

One test sits behind every trigger:

- **Patch** means the gap can be closed today from a source outside your head: the calendar invite, the email thread, the company page, LinkedIn, the applicant portal, a lawfully made recording, or a short question to the recruiter.
- **Redo** means only your memory could close the gap and the write-up window has passed, or a gate criterion failed. The two gates are time to write-up and consent.

> **Rule:** The round takes its lowest grade
>
> One Redo makes the whole round Redo, however good the rest is.
> Never close a Redo gap from memory. Mark the field UNKNOWN so you never quote it later as fact.

### 1. Time to write-up (gate)

- **Pass**: the write-up starts within 15 minutes of the calendar end time and before any other interview, recruiter, or loop-related call.
- **Patch**: it starts later than 15 minutes, but on the same day and before any other loop conversation.
- **Redo**: it starts after another interview conversation, or on the next day or later.
- **What it proves**: the record reflects this round and only this round.
- **When it lies**: the note says "same day" but was written after the next call. Compare the file timestamp with the calendar.

### 2. Verified interviewer names and titles

- **Pass**: every person in the round is listed, with spelling and title matched to the invite, the company page, or LinkedIn, and the source noted.
- **Patch**: a name or title comes from memory and hasn't been checked.
- **Redo**: doesn't apply. A coordinator or an invite can always close this gap.
- **What it proves**: your follow-up and your prep for the next round go to the right person.
- **When it lies**: a confident spelling from memory. Names decay fastest, so confidence proves nothing.

### 3. Near-verbatim questions and follow-ups

- **Pass**: every question has a quoted stem, meaning its opening words as the interviewer said them. Follow-ups sit under their parent question, and each has a one-line note of the example you gave.
- **Patch**: some questions are logged only as gist, but the list is complete. You know how many questions there were and what each one was about.
- **Redo**: there is no question list, only an impression such as "mostly system design, went well".
- **What it proves**: you can prepare the next round against what this panel actually probes.
- **When it lies**: your own paraphrase inside quotation marks. If you can't say the question in the interviewer's words, it is gist.

**57.3% - Share of their own questions that trained investigators left out of verbatim notes**

The same note-takers missed only 25% of the interviewees' details, because attention goes to answers, and that is why this criterion exists.

### 4. Every number and date said aloud

- **Pass**: every figure spoken in the round is logged with its speaker. That includes pay expectations, any band, level, headcount, start date, notice period, and metrics you quoted from your own work.
- **Patch**: a figure has no speaker, or a figure is missing but an email or a lawful recording can restore it.
- **Redo**: pay or level came up and you can't state what either side said.
- **What it proves**: you know your anchor and can stay consistent with it in the next round.
- **When it lies**: only your own numbers are logged and the recruiter's band is missing.

### 5. Employer claims and next step with a date

- **Pass**: the employer's claims (team, scope, timeline, promises to send you something) are listed with who said each one. The next step has an owner and a date, logged in the record and in one other place.
- **Patch**: the next step has no owner or no date. The fix is to ask for a date in your follow-up.
- **Redo**: doesn't apply.
- **What it proves**: you can hold the employer to its own timeline and compare loops on facts.
- **When it lies**: "We'll be in touch" logged as a next step.

### 6. Open questions

- **Pass**: a list of what you still need to know, each item with who to ask and by when, or an explicit "none".
- **Patch**: the section is blank.
- **Redo**: doesn't apply.
- **What it proves**: gaps turn into next-round questions instead of disappearing.
- **When it lies**: a blank section read as "none". Blank means you never looked.

### 7. Capture method and recording consent (gate)

- **Pass**: the capture method is logged (handwritten anchors, a typed debrief, a recording, or an AI notetaker). If anything was recorded, every party's state is logged and spoken consent from everyone was captured. Every number in an AI summary has been checked against the audio.
- **Patch**: no method is logged but nothing was recorded, or an AI summary was used and its numbers haven't been checked yet.
- **Redo**: a recording exists without logged consent from every party, or you don't know which states the parties were in.
- **What it proves**: the record is reliable and doesn't expose you when you use it.
- **When it lies**: a bot shown in the participant list was treated as consent.

### 8. Link to the resume version sent

- **Pass**: the record links to the exact resume file sent for this application.
- **Patch**: the link is missing, but you can retrieve the sent file from the portal or the application email.
- **Redo**: no copy of the version you sent exists anywhere, so you can't know what the panel read.
- **What it proves**: when an interviewer asks about a line, you know which wording they were looking at.

If you tailor with [Refolk](/candidates), each posting already has its own resume version. The link points to that version, not to a master file you have edited since.

## Can I record my job interview?

Often yes, but the answer depends on where every participant is, not just where you are. Federal law lets a party to a conversation record it. A minority of states require consent from every party, and on a call across state lines, the stricter state's law can reach you. This section is a standard, not legal advice. If you're unsure, don't record.

#### The layers a recording decision passes through

1. **Federal floor** - A party to the call may record it, unless the purpose is a crime or tort
2. **All-party states** - Every party must consent first; sources count 9 to 13 such states
3. **Hybrid states** - One rule for phone calls, another for in-person conversations
4. **Cross-state calls** - The other party's state law can apply to you
5. **Platform notice** - Indicators and bot names inform people; they are not consent

*A recording is only safe when it clears every layer, and the platform's notice at the bottom never substitutes for consent at the layers above.*

Sources disagree on the number of all-party states. Counts run from 9 to 13, and the gap comes from hybrid states that treat phone and in-person conversations differently:

| State | In-person conversation | Phone call |
|---|---|---|
| Missouri | All parties | One party |
| Oregon | All parties | One party |
| Connecticut | One party | All parties |
| Nevada | One party | All parties |

Cross-state calls matter most for candidates. The Reporters Committee for Freedom of the Press gives this example: someone in Virginia recording a call would not break Virginia law, which needs one party's consent for phone calls, but could be liable under Maryland law. Under 18 U.S.C. 2511(4)(a), unlawful interception carries up to 5 years in prison. Washington offers a model for doing it right: consent is obtained when one party announces the recording to everyone else in a reasonably effective way and the announcement itself is recorded.

The meeting tools tell people that a recording is happening. Whether that counts as consent is a separate question:

| Mechanism | Notice to others | Counts as consent? |
|---|---|---|
| Zoom host recording | On-screen notice (OK or Leave) | A gate, if the disclaimer is enabled |
| Zoom participant local recording | Indicator for everyone | Not stated |
| Third-party notetaker bot | Shows as a named participant | "Visibility is not consent" |
| Greenhouse Notetaker | Consent collected at scheduling | Yes, required before the bot joins |

A few details from the platforms' own documentation and from secondary sources:

- Greenhouse says phone and in-person interviews aren't recorded even when its notetaker is enabled. Admins can set consent rules by office and department, so read the scheduling email for each loop.
- Google Meet's explicit-consent setting is reported to be off by default.
- A vendor blog reports that Zoom's AI Companion disclaimer appears only for some features. I haven't verified this, so check the notice you actually see.
- A third-party comparison says some interview-intelligence tools send candidates a notice with an opt-out. That claim hasn't been checked against the vendors' own documentation.

> **Watch out:** Platform consent tools protect the employer, not you
>
> A bot's name in the attendee list is not consent, and in all-party states every participant must agree before recording starts.
> No interview vendor publishes a rule for a candidate's own recorder, so if you record from Virginia and the recruiter is in Maryland, the Maryland exposure is yours alone.

If you do record, ask before you press the button, then repeat the question once the recording is running so the consent is on the file:

**Spoken recording consent script**

```
Before we start, I'd like to record this conversation so my notes are accurate. It's only for my own reference and I won't share it. Is everyone here okay with that?
[If everyone says yes, start recording and repeat:]
For the record, I've started recording with everyone's agreement. Is that still okay with everyone?
```

*Say it after everyone has joined; if anyone hesitates, say "No problem" and don't record.*

You don't need to record to pass this standard. Anchor notes plus a debrief inside 15 minutes satisfy criterion 7, as long as you log the method.

## The post-round procedure, step by step

The procedure takes a round from the consent decision before it starts to a graded record the same day, in about 30 minutes of active work spread around the call. Steps 1 to 7 come from practitioner guidance. Step 8, the grading, is new to this standard; I found no public source that documents it.

#### From consent check to a graded record

1. **Decide on recording before the round** - Check the state of every party and the meeting platform's consent behaviour, then decide whether to record and ask out loud if you will. Done in about 5 minutes when the consent, or the decision not to record, is logged.
2. **Capture anchors only during the round** - Write down repeated priorities, numbers, constraints, names of systems and stakeholders, and anything you promised to send. In technical rounds, skip live notes and write everything right after the call.
3. **Debrief immediately after** - Within 15 minutes of the calendar end time, and before anything else, write the questions with quoted stems, every number with its speaker, the employer's claims, and the next step.
4. **Verify names and titles** - Match every person's spelling and title against the invite, the company page, or LinkedIn, and note the source. Done in about 5 minutes when every person is confirmed.
5. **Log the next step and date twice** - Write the next step with its owner and date in the record and in one other place, such as your tracker or calendar.
6. **Link the resume version and update the tracker** - Link the exact resume file you sent for this application and update the loop's tracker row with the round and its grade. About 2 minutes.
7. **Draft the follow-up from the notes** - Use the verified names, the employer's claims, and the logged next step to draft the thank-you or follow-up note.
8. **Grade the record Complete, Patch, or Redo** - Grade each criterion Pass, Patch, or Redo, give the round the lowest grade, close Patch items the same day from documents, and mark Redo gaps UNKNOWN.

Two ordering points. First, in the debrief, write the employer's words before your own. Their claims, their numbers, and the interviewer's follow-ups are what you lose first, and your own answers are the easiest part to reconstruct. Second, practitioners disagree about notes during the round. One camp says to capture anchors live. Another says to take no notes in technical rounds and write everything right after the call. Either way passes, provided step 3 starts inside the window.

## How a post-round record goes wrong

Most bad records don't look bad. They look complete and hold a confident error that surfaces weeks later, in a pay conversation or a next-round answer. Each failure mode below passes a casual read, so each comes with a check a second grader can run.

1. **Late write-up that still says "same day".** It looks like a tidy record dated today, but it was written after the next recruiter call, and the second conversation has blended into the first. *Check:* compare the file's created timestamp with the calendar end time and with the start of the next call.
2. **Names spelled from memory.** It looks like a full roster with titles. Names decay faster than ideas, so the false positive is a confident wrong spelling, which then goes into your thank-you note. *Check:* every name has a noted source (invite, company page, or LinkedIn).
3. **Gist written up as verbatim.** It looks like a numbered question list. In practice the stems are your summaries, and the follow-ups, which are where interviewers actually probe, are missing. Trained note-takers omitted 57.3% of their own questions, so this blind spot is normal, not careless. *Check:* every question has a quoted stem, and every round with a follow-up has one logged under its parent question.
4. **Only your own numbers.** It looks like a careful log of what you said about pay and notice period. The recruiter's band, or the level they mentioned, is missing, and that number is the anchor for everything that follows. *Check:* every figure has a speaker tag, and at least the pay discussion has entries from both sides.
5. **A visible bot treated as consent, or your home state's rule assumed.** It looks like a recording with a clean transcript. In fact no one agreed out loud, or the recruiter was in an all-party state. *Check:* the record lists every party's state and the time when spoken consent was captured.
6. **"We'll be in touch" logged as a next step.** It looks like the next-step field is filled in, but it has no owner and no date, so you can't tell when silence has become a signal. *Check:* the next step names an owner and a date. If either is missing, the criterion is Patch, whatever else it says.
7. **An AI summary accepted as the record.** It looks like the most complete record in your files. The risk is that it changed a number or invented a commitment, and the polish hides that.

> **Watch out:** A fluent summary is not a verified one
>
> AI notetakers can hallucinate and misstate what was said, and a well-written summary makes a wrong number more believable, not less.
> Spot-check every number and date in the summary against the audio before the record can pass criterion 7.

There is also a structural false positive: grading generously because the round went well. The grade measures the record, not the interview. A strong round with a missing pay figure is still Redo on criterion 4. If you can, have someone else grade one of your records cold against the triggers. If their result differs from yours, the trigger you disagreed on needs a sharper reading, not a softer one.

## How to keep track of multiple interview processes and keep your story consistent

Keep one graded record per round, link each one to its loop's tracker row, and read the previous round's record before every new conversation. Consistency comes from reading back what you said, not from remembering it.

Parallel loops are the normal case. A Robert Half survey of more than 1,000 US job seekers found 59% had received two or more offers at the same time, and 58% made their decision in two days or less. An Australian survey of more than 1,200 people found more than 60% were in at least three interview processes at once. I found no current US figure for the share of candidates running more than one loop, so treat parallel loops as common but unmeasured in the US. A secondary summary of ZipRecruiter data puts the scale per hire like this:

#### Rounds behind one US hire

| Stage | Figure | Note |
| --- | --- | --- |
| Applications | 16 | median per recent US hire, over 5 weeks |
| Interviews | 5 | each one needs its own graded record |
| Offers | 2 | decided in days, using what the records hold |

*A median recent US hire went through 5 interviews across 16 applications, so the rounds to keep track of add up across loops.*

The same secondary source reports a median time to hire of 30 days for business roles and 40 days for technical roles. Those are long enough gaps that the first round of a loop is a month old by the time you negotiate.

The consistency risk is concrete, though the evidence for it is anecdotal. One self-described recruiter wrote, "recruiters keep very detailed notes so we know exactly what salary expectations you shared with us." Another commenter told a candidate who wanted to revise their number, "you already accepted the range upfront," and said they saw no reason for the employer to go higher unless the job's scope had changed. No published case study ties inconsistent figures to a lower offer or a lower level, so don't overstate the risk. The mechanism is plain, though. The other side's record is exact, and a candidate who can't recall the figure they gave Company A is negotiating against a better record than their own.

> The recruiter already has a record of this round. The only open question is whether you do.

One graded record feeds four later jobs:

#### Where one round's record goes next

1. **Graded record** - Eight criteria, Complete or Patched the same day
2. **Follow-up note** - Verified names, employer claims, the next step and its date
3. **Next-round prep** - Quoted questions and follow-ups show what this panel probes
4. **Pay conversation** - Every figure with its speaker, including their band and your anchor
5. **Loss diagnosis** - Questions, answers, and the resume version, read together after a rejection

*A Complete record is written once and read at least four more times, so every gap in it repeats downstream.*

For loss diagnosis, compare the record with how well you fit the posting before the round. Refolk's fit score for that posting tells you how strong the match was on paper, and the record tells you what happened in the room. If the fit was high and the record shows a weak answer to a quoted follow-up, you know what to fix. If the fit was low, the round may not have been the problem.

Use this skeleton for every round. Sections are in debrief order, with the employer's words first:

**Post-round record**

```
ROUND RECORD
Company / role / requisition ID:
Round: (e.g. recruiter screen, panel 2 of 4)
Date | calendar end time | write-up started at:
Capture method: anchors by hand / typed debrief / recording / AI notetaker
Consent: parties' states: ___ | asked aloud at __:__ | answer: ___

EMPLOYER CLAIMS (who said it)
- [Name]:

NUMBERS AND DATES SAID ALOUD (speaker on every line)
- [Them: Name]:
- [Me]:

QUESTIONS (quote the stem, nest follow-ups)
1. "..."
   - Follow-up: "..."
   - My example:

NEXT STEP
- Owner: | Action: | Date: | Also logged in:

PEOPLE (verified)
- Name | Title | Source checked: invite / company page / LinkedIn

OPEN QUESTIONS
- Question | Ask whom | By when

RESUME VERSION SENT
- File or link:

GRADE: Complete / Patch / Redo
- Patch items and fix source:
- Fields marked UNKNOWN:
```

*One file per round; fill sections in this order and leave nothing blank, write "none" or UNKNOWN instead.*

## Checklist and keeping the standard current

Before you call a round's record done, every line below should be true. Run it at the end of step 8.

#### Post-round record, done means

- [ ] The write-up's start timestamp is within 15 minutes of the calendar end time and earlier than any other loop call.
- [ ] Every person in the round has a name and title matched to a named source.
- [ ] Every interviewer question has a quoted stem, with follow-ups nested under their parent.
- [ ] Every number and date spoken in the round has a speaker tag, including the employer's figures.
- [ ] The next step names an owner and a date, and the same entry appears in a second place.
- [ ] The open questions section lists items with an owner and a deadline, or says "none".
- [ ] The capture method is logged, and any recording has every party's state and the time of spoken consent.
- [ ] Every number in any AI summary has been checked against the audio.
- [ ] The record links to the exact resume file sent for this application.
- [ ] The round's grade equals the lowest criterion grade, Patch items are closed today, and Redo gaps are marked UNKNOWN.

Four inputs to this standard change over time. Re-check them by mechanism rather than trusting a value you read once:

- **State recording law.** The count of all-party states differs by source because of hybrid states. Before recording anyone in a state you haven't checked, look up that state in the Reporters Committee for Freedom of the Press recording guide, and confirm whether the rule differs for phone and in-person conversations.
- **Platform consent behaviour.** Defaults and disclaimers change with product releases, and admins can configure them per office. Read the notice you actually see in each meeting, not a description of it.
- **Employer notetaker policies.** Consent rules for employer-run bots are set in the employer's applicant tracking system and can vary by office. The scheduling email is the most reliable place to find what applies to your round.
- **Loop length.** Median time to hire is reported per period by hiring platforms. Check the current figure when deciding how long a loop's records need to stay readable, and keep them at least until the loop closes.

Recalibrate the standard itself every few loops. Pull three old records at random and grade them cold. If you grade any of them higher than you did on the day, that criterion's trigger has drifted, and the fix is to reread its trigger, not to relax it.

## Frequently asked questions

### Can I record my job interview?

Often yes, but it depends on where every participant is, not just where you are. Federal law lets a party to a call record it. Some states require consent from all parties, though, and if the recruiter is in one of those states, that state's law can apply to you. Ask out loud before you start, get a spoken yes from everyone, and capture that yes on the recording. If you aren't sure, don't record. Anchor notes plus an immediate debrief still meet the standard.

### What should I write down after a job interview?

Write down eight things: the time you started writing, each interviewer's verified name and title, every question with its opening words quoted, every number and date said aloud with who said it, the employer's claims and the next step with an owner and date, your open questions, how you captured the notes, and the resume version you sent. Do the employer's words first, because those fade fastest.

### How soon after an interview should I write my notes?

Start within 15 minutes of the scheduled end, and always before your next interview or recruiter call. No one has published an hour-by-hour forgetting curve for interviews, so 15 minutes is a practitioner bright line, not a proven threshold. The research does show that recall right after a conversation is already low, around 10%, and that writing early beats recalling later. Nothing supports waiting until the next day.

### Should I take notes during a technical interview?

Keep it to anchors at most. Practitioners disagree here. One camp says to jot down repeated priorities, numbers, constraints, and promises during the call. Another says to take no notes during technical rounds and write everything right after you hang up. Either way passes this standard if the full debrief starts within 15 minutes. What fails is splitting your attention so that both your answer and your notes get worse.

### How do I keep my story consistent across interviews?

Read your last round's record before every new conversation instead of relying on memory. Recruiters say they keep detailed notes on things like the salary expectations you gave. So the figures, dates, and examples you used earlier are already on file on their side. Log every number with who said it, and before the next round, read back what you said. Change a figure only on purpose, and have a reason ready, such as a change in scope.

### Is an AI notetaker summary good enough as my interview record?

Not without review. AI notetakers can hallucinate and misstate what was said, and a summary that reads cleanly can still have a wrong number in it. Use the summary as a draft. Check every number and date against the audio, add the quoted question stems, and log the consent you got from each participant. An unreviewed summary grades Patch at best, and any recording made without everyone's consent grades Redo.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/candidates/guides/post-round-interview-record-standard-2*
