# The Resume Metric Defense Standard, Seven Criteria per Figure You Sent

*Grade every figure on your sent resume pass or fail against seven criteria and leave with a one-page derivation sheet marking each failure restate, range, or retract.*

- Canonical URL: https://www.refolk.ai/candidates/guides/resume-metric-defense-standard
- Pillar: Interviewing
- Format: Standard
- Published: 2026-11-02
- Last reviewed: 2026-11-02
- Reading time: 20 min
- Keywords: how to defend resume metrics in an interview, how did you measure that interview question, explain the numbers on my resume, estimated metrics on resume how to justify, can I share numbers from my current employer in an interview, interviewer asked about resume percentage couldn't answer

## Key takeaways

- A resume figure is interview-ready only when it passes seven criteria: named baseline, comparison period, measurement source, own share, cross-document consistency, disclosure-safe phrasing, and a spoken derivation of 60 seconds or less.
- A 60-second derivation is about 120 to 160 words at a normal speaking pace, which sits at the top of the 30 to 60 second band practitioners give for factual answers.
- Interviewers in one study correctly detected only 13 to 23% of impression management tactics, while 52% of small employers in a CareerBuilder survey changed their mind about a candidate after a reference call, so a number that survives the interview can still fail later.
- In Refolk's index there are 13,743 US profiles tagged Behavioral Interviewing against 48 tagged Structured Interviewing, a ratio of 286 to 1, so most interviewers will probe your numbers in no fixed order.
- Multi-year growth must be compounded, not added: five years of 15% growth is about 101% total, not 75%.
- Every figure that fails any criterion gets one of three verdicts before the next round: restate it, downgrade it to a range, or retract it.

This standard is for candidates who already have interviews booked off a resume full of percentages, dollar figures, and counts. They need to know, before the first call, which of those numbers will hold up when someone asks "how did you measure that?" It grades each figure on the version you actually sent against seven written criteria, pass or fail. You finish with a one-page derivation sheet that marks every failing number restate, downgrade to a range, or retract.

## What counts as a defensible resume figure?

A figure is defensible when it passes all seven criteria below. Fail any one, and it needs a verdict other than pass before the next round.

Two terms first. A **figure** is any percentage, dollar amount, count, multiple, or rank on the resume. A **derivation** is the spoken chain from the starting state to the number: what it was, what it became, over what window, measured by what, and how much of it was yours.

The seven criteria, each with what it proves and what it looks like when it lies:

1. **Named baseline.** Proves the improvement started from a real, normal state. It lies when the baseline was a broken system or a cherry-picked low point.
2. **Comparison period.** Proves the before and after windows are like for like. It lies when the windows differ in length or season, so seasonality produces the lift.
3. **Measurement source.** Proves the number came from a system or an artifact, not from memory. It lies when the only source is "I remember the dashboard" and no export, report, email, deck, calendar, or ticket backs it up.
4. **Own share.** Proves what you personally did. It lies when a team result is written in the first person.
5. **Cross-document consistency.** Proves the figure is identical on the resume, LinkedIn, cover letter, any take-home, and in what a referee would say. It lies when a later edit on one surface quietly changed the number.
6. **Disclosure-safe phrasing.** Proves you can say the figure without handing over non-public information. It lies when a percentage on a small or public base lets someone work back to the employer's revenue.
7. **Spoken derivation in 60 seconds or less.** Proves you can explain the figure live. It lies when a memorised answer beats the timer and then falls apart on the second follow-up.

> **Rule:** Unknown counts as fail
>
> Every criterion is graded pass or fail, and a field you cannot fill from an artifact is a fail, not a partial pass. Two graders score separately before comparing, so the verdict reflects the evidence rather than the conversation.

The point of writing criteria this way is that two people grading the same figure should reach the same verdict. "I could probably explain it" is not a grade. "Baseline: named, from the Q2 export. Period: fail, windows differ by a month" is.

## Which follow-up questions come after "how did you measure that?"

Expect baseline first, then your personal role, then method, then result. The numbers-specific chain this guide uses (baseline, period, source, attribution, durability) is my own composite of published interviewer guidance. It is not a published script.

The best public material on probing comes from the US Office of Personnel Management (OPM). Its structured interview guidance tells interviewers to decide before the interview how much probing to allow (none, limited, or unlimited) and to tie probes to each part of a STAR answer (Situation, Task, Action, Result). Its sample probes are "Who was involved? What specific actions did you take? What was the outcome?" OPM training also tells interviewers to "Seek clarification when candidates say 'we did…' or 'our group…'". That one instruction is why own share is the criterion most likely to be probed, even when nobody asks a numeric question.

The clearest published order comes from a practitioner list: "What was the condition when you started?" then "What part did you personally own?" then alternatives and disagreement, then "What did you measure?" and "What was the result?" Another practitioner guide recommends planned probes, meaning neutral follow-ups written in advance, citing Levashina and colleagues (2014).

#### The follow-up chain for a single resume figure

1. **Baseline** - What was the number when you started, and was that normal?
2. **Period** - Over what window, compared with what equal window?
3. **Source** - Which system, report, or export produced the number?
4. **Attribution** - Which part was yours, and who would confirm it?
5. **Durability** - Did the result hold afterward? Optional, not an established probe.

*Each probe tests one criterion, and a figure that fails early never reaches the later questions.*

I include durability because once the source checks out, "did it last?" is the obvious next question. But no public interviewer guidance treats it as a standard probe, so prepare one sentence for it and give it no more weight than that.

### Why the order will not be predictable

Most interviewers who probe your numbers are not following a structured script. In Refolk's index of professional profiles, the skill tags look like this:

| Skill tag | Country | Profiles |
|---|---|---|
| Behavioral Interviewing | US | 13,743 |
| Structured Interviewing | US | 48 |
| Structured Interviewing | UK | 4 |

**286:1 - US profiles tagged Behavioral Interviewing for each one tagged Structured Interviewing**

In Refolk's index, 13,743 against 48, and the top titles in a 25-profile behavioral sample were CEO (3) and Founder (2).

Skill tags are self-reported, so this shows who advertises a method, not who practises it. It still points one way. The US structured set includes titles like VP Talent Acquisition and Technical Recruiter, while the behavioral sample is led by founders and CEOs. In a startup loop, the person asking about your 40% may be a founder improvising. That means the follow-ups can arrive in any order, and your derivation sheet has to answer any of them first.

There is one more reason to take specific questions seriously. Research summarised for practitioners found that specific, situational questions led interviewers to recognise more impression management tactics. Impression management (IM) means the ways candidates shape how they come across, honestly or not. Precise probes on a number are exactly the kind of question that works.

## Why grade the numbers after the resume has gone out?

Because the version you sent is now a fixed record. The interviewer, your LinkedIn profile, and your referees will all be compared against it. And the bigger risk is usually not the live interview but the reference call that follows.

| Measure | Value | Source |
|---|---|---|
| HR managers who caught a resume lie | 75% | CareerBuilder, 2017 |
| Small firms checking references before interview | 25% | CareerBuilder |
| Small firms checking references after interview | 53% | CareerBuilder |
| Changed mind after a reference | 52% | CareerBuilder |
| IM tactics interviewers detected | 13 to 23% | Roulin et al., 2015 |

These rows come from different samples and years, so they are not strictly comparable. The CareerBuilder headline figure comes from 2,575 hiring and HR managers surveyed in 2017. Even so, the shape holds. Interviewers are poor at catching inflation live, detecting 13 to 23% of IM tactics in one study. But a majority of small employers check references after the interview, and about half have changed their mind after one. A figure that gets through the interview can still fail when a former colleague hears it read back.

What is not known: nobody publishes how often accomplishment metrics specifically get checked. CareerBuilder's list of the most common lies covers embellished skills or responsibilities, employment dates, job titles, and degrees. Metrics are not separated out. Do not assume your numbers go unchecked just because no study counts them.

On how much a bad answer costs, the evidence is mixed. A Checkster survey of 400 job seekers and 400 managers found 29.4% of hiring managers would still hire someone who misrepresented themselves if there was a good explanation. That is a minority, not a safety net.

### Where this standard departs from resume-writing advice

Resume guides do their estimating before sending. This standard does it after, which changes what you are allowed to do. You cannot quietly rewrite a figure the interviewer already has in front of them. Cross-document consistency and disclosure safety therefore limit how you restate it, and a correction has to be said out loud.

The first job is knowing which version you sent. If you tailored several versions for different postings, that alone can take an evening. [Refolk](/candidates) keeps each tailored resume tied to the posting it went to, so the frozen version is already on file.

## How long should a spoken derivation take?

Sixty seconds or less per figure. At a normal speaking pace of 120 to 160 words per minute, that is about 120 to 160 words, the top of the band practitioners give for factual answers.

No research sets a norm for a single follow-up answer. The practitioner ranges agree with each other closely enough to work with:

| Context | Range | Source |
|---|---|---|
| Behavioral answer (UCSB, second-hand) | 60 to 90 s | interviewpilot.app |
| Warm-up or factual answer | 30 to 60 s | interview-prep.io |
| Quick screen answer | about 45 s | qcardai.com blog |
| Hard ceiling, uninterrupted | 3 min | interviewdrills.com |

A derivation is a factual answer, not a story. The full behavioral example around it can take 60 to 90 seconds. The "how did you measure that" follow-up should fit inside 60.

Video calls shorten the window. One practitioner puts it this way: "A 2-minute behavioral answer in a conference room is a 1:40 answer on a video call." Applying the same proportion, a 60-second derivation in person is roughly a 50-second derivation on video. Rehearse to the shorter number.

**Spoken derivation script, one figure**

```
The number is [figure as you will now say it].
We started at [baseline value], which was [normal state / why it was representative].
I compared [period A, length] against [period B, same length, same season].
The number comes from [system, report, or export], which I pulled [when / how].
My part was [specific action], roughly [fraction] of the result; [name, role] would confirm that.
It held at about [value] for [time] afterward. (Optional.)
If you want the rough version: [baseline] to [result] is [the arithmetic in one line].
```

*Fill the brackets from your rebuilt fields; read it aloud with a timer and cut until it lands under 60 seconds.*

The last line matters more than it looks. The pass test in the procedure is whether a peer can repeat the arithmetic back, and one plain line of math is what makes that possible.

## Can I share numbers from my current employer in an interview?

Share accomplishments and responsibilities. Do not share customer lists, product development, marketing strategy, pricing, or any figure that is both precise and non-public; abstract those into rounded numbers, ranges, or order of magnitude.

Employer-side legal guidance draws the same line from the other side. Interviewers should never ask about customer lists, product development, marketing strategies, or pricing, but may ask about accomplishments and responsibilities. The Sedona Conference's commentary on protecting trade secrets through the employment life cycle tells hiring employers to take care not to ask a candidate for any nonpublic information related to a prior employer. HR guidance is blunter about the candidate. Someone who volunteers pricing or client lists "has just answered the confidentiality question, and the answer is no." A precise figure can be true and still fail this criterion. Holding back is itself evidence of integrity.

What is not established: I found no standard confidentiality clause that explicitly allows you to share percentages. That a percentage is "safe" is a practitioner assumption, not a legal rule. Read the confidentiality clause in your own agreement. If it is unclear and the figure matters, an employment lawyer can tell you in one conversation.

Practitioner substitutes that work:

- Round the number.
- Give a range instead of a point.
- State the order of magnitude.
- Describe the rate in words, e.g. "High-single-digit revenue growth" instead of "8.4% CAGR".

Watch the base. A percentage on a public or small base can identify the business: if the employer's revenue is public, "grew revenue 12%" is a dollar figure in disguise. Ask whether the figure plus the employer's name can be worked back to something non-public.

#### Disclosure risk against derivation strength

Horizontal axis runs from Derivation weak to Derivation solid. Vertical axis runs from Public or abstracted to Precise and non-public.

| Quadrant | What it means |
| --- | --- |
| Precise, non-public, weak derivation | Retract it; nothing about it survives a probe |
| Precise, non-public, solid derivation | Downgrade to a range or order of magnitude |
| Public or abstracted, weak derivation | Restate with the method named, or downgrade to a range |
| Public or abstracted, solid derivation | Pass; keep it on the sheet as written |

*A number you can derive but should not disclose gets abstracted, while one you can neither derive nor safely disclose gets retracted.*

Sources disagree on whether to sanitise first or rebuild the math first. I rebuild first, privately, and sanitise before speaking. You cannot pick a safe range without knowing the true number it has to contain.

## The grading procedure, step by step

Run this once per application, starting from the file you sent. The time estimates are mine, not measured: about 5 minutes to freeze the version, about 15 to list the figures, and 10 to 20 minutes per figure to rebuild the fields.

#### From sent resume to derivation sheet

1. **Freeze the version you actually sent** - Pull the exact file submitted for this application and match it to the posting. Done means one frozen file per application, not your latest master draft.
2. **List every figure in a numbered table** - Extract every percentage, dollar amount, count, multiple, and rank, one row each. Done means nothing numeric on the page is missing.
3. **Rebuild baseline, period, source, and own share** - Reconstruct each field from exports, reports, emails, slide decks, calendars, and tickets. Done means all four fields are filled or marked unknown.
4. **Cross-check every other document** - Compare each figure against LinkedIn, the cover letter, any take-home, and what a referee would say, with an ex-colleague if possible. Done means no figure differs between documents.
5. **Run the disclosure check** - Flag client names, exact dollars, and proprietary details, then pick an abstraction. Done means each figure is marked public, abstracted, or withheld.
6. **Time a spoken derivation with a peer** - Explain each figure aloud to a peer acting as interviewer, who then asks two unscripted follow-ups. Done means 60 seconds or less and the peer can repeat the math back.
7. **Grade independently and assign a verdict** - You and a second grader score each criterion separately, then compare. Every figure ends as pass, restate, downgrade to range, or retract.
8. **Write the one-page derivation sheet** - One line per figure with the number as you will now say it, its four rebuilt fields, and its verdict. Done means it fits one page and stays open in the loop.

Notes on the steps that people tend to rush:

- **Step 3** is where most of the time goes, and where "unknown" is the honest answer more often than you expect. Write it down. An unknown field becomes a downgrade or a retraction in step 7. It does not get filled from memory in step 6.
- **Step 4** is the referee test. Read the resume line to an ex-colleague and ask what number they would have given. If Refolk drafted your cover letter from the same work history as the resume, the figures should already match. Check the LinkedIn profile anyway, because that is the surface you edit by hand.
- **Step 7** borrows the structured-interview practice of recording independent ratings before any group discussion. If you have no second grader, grade alone, wait a day, and grade again without looking at the first pass.

## How this goes wrong: failure modes and false passes

Most failures are false passes: the criterion looks met on paper and breaks on the first or second probe. A Blind interviewer puts the stakes plainly: "That is why you probe people when you interview them. I catch people in lies all the time about their metrics." Test for each of these directly.

1. **Broken baseline.** The improvement is real but measured against a starting state nobody would call normal. One Blind thread mocks "Improved pipeline performance by 1000x…", the joke being that the original pipeline was a wreck. The false pass is that a baseline exists but was cherry-picked. *Check:* would a peer who worked there call that starting point the normal state?
2. **Seasonal comparison.** The lift comes from the calendar, not from you. An interviewer's reflex is "How did you prove causality? So much seasonality." *Check:* the before and after windows are the same length and fall in the same season.
3. **"We" passing as "I".** The figure reads fine on paper and then fails OPM's "we did" probe. *Check:* state your share as a fraction, and name someone who would confirm it. If shared credit is the main issue, the Shared-Credit Resume Line guide covers splitting and scoping the line itself.
4. **Added instead of compounded.** Forum advice suggests that after five years of 15% growth you can say you increased revenue 75%. That adds the years. Compounded, 1.15 to the fifth power is about 2.01, so the true figure is about 101%. The added version understates here, but the same mistake can overstate in other cases, and either way the math breaks when someone redoes it. *Check:* recompute every multi-year figure.
5. **Identifiable percentage.** A percentage passes the "no exact dollars" test but identifies the business because its base is public or small. *Check:* can the figure plus the employer name be worked back to revenue?
6. **Post-send edit.** You improved the LinkedIn wording after applying, and now the two surfaces disagree. *Check:* edit timestamps, against the file you actually sent.
7. **Misquoted statistic.** If you cite research to justify a correction, cite it accurately. CNBC wrote that "just 12 percent of HR managers will call a dishonest candidate back". The CareerBuilder press release ties the 12% to something else: candidates who did "something unusual or outrageous". Do not repeat the CNBC version.
8. **Rehearsed derivation.** A memorised answer beats the 60-second timer and breaks on the second probe. *Check:* the peer in step 6 asks two follow-ups you did not script, in an order you did not choose.

> **Watch out:** The timer is the easiest criterion to fake
>
> Passing 60 seconds on a script you have read aloud ten times proves fluency, not understanding. The derivation only passes if it survives two unscripted follow-ups and the peer can repeat the arithmetic back.

> A number you cannot rebuild from an artifact is a claim about your memory, not about your work.

## Restate, downgrade, or retract: assigning the verdict

Every figure that fails any criterion gets exactly one of three verdicts. Restate if the evidence supports a corrected exact number. Downgrade to a range if it supports only a band. Retract if the evidence cannot carry any version of the claim.

| Verdict | When | How to say it in the loop |
|---|---|---|
| Pass | All seven criteria met | As written |
| Restate | Rebuilt number differs but is solid | "The cleaner number is X; here is why" |
| Downgrade to range | Only a band is supportable, or disclosure requires it | "Roughly X to Y, estimated from Z" |
| Retract | No artifact, no share, or unsafe to discuss | "I would not stand behind that figure; what I can show is..." |

For estimates, practitioners recommend conservative figures labeled "approx." or "estimated", ranges like "an estimated 15 to 20%", and naming the method, for example "estimated using meeting calendar and average attendance". The common line on fabrication is short: "Never fabricate a number you cannot explain when asked." No regulator or industry body has published a standard for estimated resume metrics. All of this guidance comes from practitioners.

### Why correct it before you are challenged

No interview study measures whether one unexplainable figure lowers trust in your other claims. The nearest evidence is a 262-participant lab study by Schweitzer, Hershey, and Bradlow. It found that trust harmed by deception never fully recovers, even after a promise, an apology, and consistent trustworthy behaviour, while trust harmed by untrustworthy actions alone could be repaired. My inference, from trust games rather than hiring data: pulling a figure before it is challenged looks more like the repairable case than defending it until it breaks. So when a failing line comes up, correct it the first time, in your own words.

**One-page derivation sheet**

```
Application: [company, role, posting date] | Sent file: [filename, date sent]
# | Resume says | Say | Baseline | Period | Source | My share (confirmer) | Disclosure | Verdict
1 | [as sent] | [as you will say it] | [value, normal?] | [A vs B, same length/season] | [system/export] | [fraction] ([name]) | [public / abstracted / withheld] | [pass / restate / range / retract]
2 |  |  |  |  |  |  |  |
3 |  |  |  |  |  |  |  |
Correction line for any retracted figure: "The resume says [X]. I went back to the data, and I would not stand behind it. What I can show is [Y]."
```

*One line per figure, in resume order; the "say" column is the only version you speak in the loop.*

## Keeping the sheet current through the loop

The sheet is a living record for one application. Re-check it after every round, because each conversation can add a follow-up you could not answer and each referee you name becomes a check on your own share.

After each round, add any probe that caught you off guard as a new line under the relevant figure, and rebuild that field before the next round. If a figure's verdict changes, change it on every surface that still carries the old number. That is the same consistency test from step 4, run again. For preparing the non-numeric lines on the same resume, the Sent-Resume Question Sheet traces a single tailored version through a loop line by line, and this sheet slots into it.

#### Before the next round

- [ ] The derivation sheet is built from the exact file sent for this application, not a later draft.
- [ ] Every percentage, dollar amount, count, multiple, and rank on that file has a numbered line.
- [ ] Every line has baseline, period, source, and own share filled from an artifact or marked unknown.
- [ ] No figure differs between the resume, LinkedIn, cover letter, and any take-home.
- [ ] An ex-colleague or named referee would give the same figure and the same share.
- [ ] Every figure is marked public, abstracted, or withheld, and no abstracted figure can be worked back to revenue.
- [ ] Every multi-year figure has been recomputed as compounded, not added.
- [ ] Each derivation lands in 60 seconds or less in person and about 50 seconds on video.
- [ ] A peer has asked two unscripted follow-ups per figure and repeated the math back.
- [ ] Two independent grades exist for every figure, and every failure is marked restate, downgrade to range, or retract.
- [ ] The correction line for each retracted figure is written out and has been said aloud once.

The sheet stays open on the desk during every call. It is not there to read from. It is there so that when someone asks "how did you measure that?", the answer is already graded.

## Frequently asked questions

### How do I answer "how did you measure that?" in an interview?

Give the derivation in one breath: the starting value and what it was, the before and after windows, the system the number came from, and the part you personally did. Keep it under 60 seconds, roughly 120 to 160 words. If you only know a range, say so and name the method, for example an estimate built from calendars and average attendance. Then stop and let the interviewer pick the next probe.

### How do I justify an estimated metric on my resume?

Say it is an estimate and name the method before anyone asks. Practitioner guidance recommends conservative figures labeled approx. or estimated, ranges such as 15 to 20%, and a stated method. The test is whether you can rebuild it from artifacts like exports, emails, or tickets. If you cannot explain the estimate when asked, downgrade it to a range or retract it before the next round.

### Can I share numbers from my current employer in an interview?

Share accomplishments and responsibilities, but not customer lists, product development, marketing strategy, pricing, or precise non-public figures. Use rounded numbers, ranges, or order of magnitude instead, such as high-single-digit growth rather than an exact rate. HR guidance treats volunteering confidential detail as a failed integrity test, so restraint helps you. No standard clause explicitly permits percentages, so read your own agreement.

### What should I do if I couldn't answer a question about a resume percentage?

Correct it at the next contact rather than hoping it passes. Rebuild the figure from artifacts, then restate it, downgrade it to a range, or retract it, and say so plainly in the next round or a short note to the recruiter. Lab research on trust suggests deception does lasting damage while a corrected mistake is more repairable, though that evidence comes from trust games, not hiring.

### Should I update my LinkedIn numbers after I've already sent my resume?

Only if the change makes LinkedIn match the version you sent, or if you are retracting a figure everywhere. An edit made after sending that changes a number creates a mismatch an interviewer or referee can spot. Check the sent file first, then align every document to the verdict on your derivation sheet, not to whichever draft is newest.

---

*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/resume-metric-defense-standard*