# The Market Pay Band Run, Three Public Records to a Floor, Target, and Stretch

*In about three hours, turn one role, level, and metro into a dated, sourced floor, target, and stretch number you can defend on a recruiter call.*

- Canonical URL: https://www.refolk.ai/candidates/guides/market-pay-band-public-records
- Pillar: Reading the market
- Format: Playbook
- Published: 2026-11-15
- Last reviewed: 2026-11-15
- Reading time: 22 min
- Keywords: how to figure out market salary for my role, how to research salary range before applying, what salary should I ask for, how to use BLS wage data for salary negotiation, H-1B salary data for salary research, posted salary range vs actual pay

## Key takeaways

- A defensible pay band combines three public records: 25 to 40 level-matched posted ranges, the government OEWS wage percentiles for your occupation code and metro, and employer-filed LCA wage records for your target companies.
- Do not use a posted range's midpoint as your target: in Glassdoor's match of 147,568 listings to reported pay, over 60% of in-range salaries sat below the range median.
- The median annual posted range is 40.5% of its midpoint, about as wide as the widest traditional internal pay band, so a posting usually describes a whole grade rather than the zone where people are actually hired.
- LCA prevailing wage levels I to IV sit at roughly the 17th, 34th, 50th, and 67th OEWS percentiles, which makes them the only public bridge between an employer's level label and a government percentile.
- Setting the floor at 80% to 85% and the stretch at 115% to 120% of the target gives a band spread of roughly 35% to 50%, the same range as traditional employer pay structures.
- In Refolk's index, 83.1% of UK profiles in the compensation title family use 'Reward' versus 0.2% in the US, so a posting search under only one title skips whole groups of employers.

This is the build to run before your first recruiter call: one target role, one level, one metro, turned into a pay band you can defend from public records rather than one salary site. It is for job seekers who need a number they can explain. You come out with a dated sheet holding a floor, a target, and a stretch, the spread between your sources, and a refresh date.

## What a defensible market pay band is

A defensible market pay band is three numbers: a floor (your walk-away line), a target (what you quote), and a stretch (the most you can justify). Each one traces back to public records that you can name and date. "I read it somewhere" does not count. "The metro median for this occupation code, cross-checked against 31 level-matched postings and 12 employer wage filings" does.

The method uses three sources. Each has a known bias, and the biases do not overlap:

- **Posted ranges.** The salary range an employer prints in a job posting. Plentiful, current, and wide.
- **OEWS percentiles.** The Occupational Employment and Wage Statistics survey is the federal wage survey run by the Bureau of Labor Statistics (BLS). It publishes wage percentiles by occupation code and metro area. Authoritative, but it blends every level of seniority and lags the market.
- **LCA filings.** A labor condition application (LCA) is the wage attestation an employer files with the Department of Labor before an H-1B petition. The disclosure files show the offered wage and the level the employer assigned to the job. Employer-specific and level-tagged, but they are offers, not payroll.

You will meet two more terms. An **SOC code** is the Standard Occupational Classification code, which is how OEWS and LCA files identify an occupation. **Range penetration** is how far a salary sits inside a range: (salary - minimum) ÷ (maximum - minimum). A salary at the minimum has 0% penetration, and one at the midpoint has 50%.

#### What each number actually measures

1. **Posted range** - A good-faith span that often covers a whole grade, several levels, or several locations
2. **OEWS percentile** - What all employers in the metro pay this occupation code, pooled over about three years
3. **LCA offered wage** - What one employer attested it would pay one worker at a stated level
4. **Final pay** - What the hire actually received, which none of the three public records discloses

*The three public sources sit at different distances from final pay, and none of them reaches it.*

A word on the method's origins. Compensation analysts build internal ranges by setting a midpoint and then applying a spread. I borrow their arithmetic here. I am not aware of any published standard that adapts this practice for job seekers, so treat the reconciliation rules below as my method, not an industry rule.

## What each public source measures, and how it misleads

Every source answers a slightly different question, so every source can mislead in its own way. The table shows what each one proves and what it looks like when it lies.

| Source | What it proves | How it lies |
|---|---|---|
| Posted ranges | What employers say they will pay right now | Ranges that span several levels inflate the stretch, and hires cluster low |
| OEWS percentiles | Metro pay for the occupation across all employers | Mixes levels, pools three years of panels, and older releases cap high earners |
| LCA filings | One employer's attested offer at a stated level | Prevailing wage is a legal floor, not the market, and final pay is not shown |

### Posted ranges

Of the three, posted ranges are the most tempting source and the most misread. Glassdoor matched 147,568 US full-time listings to salaries that employees reported for the same job, employer, and city. Reported pay fell inside the posted range 67% of the time, below it 22% of the time, and above it 11% of the time. Over 60% of the in-range salaries sat below the range median. Answering "posted salary range vs actual pay" in one line: actual pay usually lands inside the range, and usually in its lower half. As far as I can find, nobody has published a replication of that study, so treat the exact shares as one good measurement rather than a law.

Postings that carry pay are also a selected sample. An NBER working paper found that only 13.5% of all job postings include any wage information, and the median firm discloses pay in under 1% of its posts. A sample of 25 to 40 posted ranges therefore over-represents the firms that do disclose.

### OEWS percentiles

OEWS covers about 830 occupations in about 530 metropolitan and nonmetropolitan areas. Each release is pooled from six semiannual survey panels spanning about three years. BLS itself says the survey is less useful for looking at changes over time. Older releases also top-coded high wages: any percentile at or above $115.00 per hour or $239,200 per year was shown only as "equal to or greater than" that cap. Newer releases publish specific values for many of those cells, but you should still check each cell you use for a top-code marker.

### LCA filings

Each LCA row carries the offered wage (WAGE_RATE_OF_PAY_FROM and WAGE_RATE_OF_PAY_TO), the unit of pay (WAGE_UNIT_OF_PAY), the PREVAILING_WAGE, the prevailing wage level (PW_WAGE_LEVEL, I to IV), the worksite location, and the SOC_TITLE. Under 20 CFR 655.731, the employer must pay the greater of the actual wage (what it pays similar workers) or the prevailing wage. The regulation adds that the actual wage "is not an average of the wage rates paid to all workers employed in the occupation." The files are released as quarterly cumulative files.

The prevailing wage levels also give you a bridge between an employer's level label and a government percentile.

| Anchor | Approximate percentile |
|---|---|
| LCA Level I | ~17th |
| LCA Level II | ~34th |
| LCA Level III | ~50th |
| LCA Level IV | ~67th |
| Where most hires land in a posted range | Below its 50th |

That bridge matters. No other public record ties an employer's own level assignment to a government percentile.

## How wide a real pay band is compared with a posted range

A real hiring band is narrower than most posted ranges. The median annual posted range is about as wide as the widest traditional internal pay band. Measured against its midpoint, the median annual posted range is 40.5% wide. Internal bands at employers with traditional structures span 36.5% at the narrowest and 51.3% at the widest, measured against the minimum.

To compare those figures you have to put them on the same basis. The last column converts every width to a percentage of the range minimum, using (1 + w/2) ÷ (1 - w/2) - 1.

| Source | Width as reported | Basis | As % of minimum |
|---|---|---|---|
| NBER, job postings | 28% | Midpoint | 32.6% |
| NLx, annual postings | 40.5% | Midpoint | 50.8% |
| NLx, transparency states | 40% | Midpoint | 50.0% |
| WorldatWork traditional, narrowest | 36.5% | Minimum | 36.5% |
| WorldatWork traditional, widest | 51.3% | Minimum | 51.3% |

What to take from this:

- **The median posting describes a whole grade, not a hiring zone.** At 50.8% of its minimum it almost matches the widest traditional band. That is one reason hires cluster in the lower half.
- **Pay transparency laws are a weak filter.** Annual ranges posted in states with transparency laws had a median width of 40%, against 43% in the 36 states without such laws. A secondary account of the Glassdoor study found ranges in transparency-law states about as accurate as ranges posted voluntarily. The likely reason is that these laws require a good-faith range but cap neither its width nor how many levels it covers. Colorado's text asks for "the highest and lowest amounts the employer in good faith believes it would pay." California asks for a good-faith estimate of what the employer "reasonably expects to pay for the position upon hire." Minnesota treats open-ended ranges as non-compliant. Trackers disagree on how many states have these laws, so check the statute text for any state you rely on.
- **Hourly postings are tighter.** The median hourly range is 12.5% wide. Do not judge an hourly range against annual-range benchmarks.
- **The 60% line is a rule of thumb.** No peer-reviewed rule says when a range covers several levels. I flag any annual posting wider than 60% of its midpoint. One secondary source traces a 60% line to a New Jersey proposal, which I could not verify at the primary source.

> A posted range is a grade, not a hiring zone. Most hires land in its lower half.

## The three-hour build, step by step

The build runs about three hours across seven steps, and almost half of that time goes into collecting postings. The time estimates are my own, from running the build, not from a published source. Do the steps in order, because each one depends on the one before it.

#### Market pay band build

1. **Fix the target and code it** - Write down one role, one level, and one metro, then map the title to an O*NET-SOC code by checking the duties, not the title. About 20 minutes. Done means the SOC code and metro area name are recorded.
2. **Pull the OEWS metro percentiles** - Record P10, P25, P50, P75, and P90 for that code and metro, the release vintage, and any top-code marker. About 15 minutes. Done means one dated row.
3. **Collect 25 to 40 level-matched postings** - Start with states that require pay in postings, and log the minimum, maximum, and width of each range. Drop open-ended and multi-level ranges. About 75 minutes. Done means a cleaned list with the median minimum, midpoint, and maximum.
4. **Pull LCA records for target employers** - Filter to Certified status, your SOC code, and your worksite metro, then annualize WAGE_RATE_OF_PAY_FROM using WAGE_UNIT_OF_PAY. About 30 minutes. Done means an offered-wage median and the prevailing wage level mix.
5. **Normalize every figure to annual base** - Convert everything to annual base pay on the 2,080-hour full-time basis that OEWS uses, and strip variable pay. About 10 minutes.
6. **Reconcile into floor, target, and stretch** - Take one candidate target from each source, choose where they converge, and set the floor at 80 to 85% and the stretch at 115 to 120% of the target. Record the spread between sources. About 20 minutes.
7. **Date the band and set a refresh** - Write the build date and schedule refreshes. LCA files update quarterly, OEWS annually, and postings constantly. About 5 minutes.

### Step 1: fix the target and code it

Write the target as a single line, for example "senior compensation analyst, Chicago metro". Then look the title up in O*NET OnLine's job coding tool and read the duties listed for each candidate code. Pick the code whose duties match your work, not the one whose title matches. Some titles crosswalk to several codes, and BLS gives "Project Manager" as an example.

The level is where people fool themselves, because it feels like a stretch to call yourself mid-level and a stretch to call yourself senior. Settle it before you collect any data. Read three postings at each level you are considering and check which one your history actually proves. [Refolk](/candidates) scores how well your own history fits a specific posting. That gives you an outside read on whether you sit at Level II or Level III before you lock the level in.

### Step 2: pull the OEWS metro percentiles

Pull the row for your SOC code in your metro. Record all five percentiles, the release vintage, and whether any cell carries a top-code marker. If the metro row is suppressed or missing, use the state row, and label it as a state figure in your sheet.

### Step 3: collect 25 to 40 level-matched postings

Search the job boards under every title variant your role goes by. Start with postings in states that require pay in postings, because they give you more ranges per hour of searching. Then add postings from your target metro. For each posting, log the minimum, the maximum, the unit, and the width, where width = (max - min) ÷ midpoint.

Drop these:

- **Open-ended ranges**, such as "from $X" or "up to $Y".
- **Multi-level ranges**, meaning annual ranges wider than about 60% of their midpoint, or any range with "varies by location" wording.
- **Level mismatches**, where the required years or scope clearly sit a level above or below yours.
- **Variable-pay roles** where the base figure is not separated out.

Stop at 25 if your role is common and the medians have stopped moving. Push toward 40 if the ranges scatter widely.

**Posting log, one row per posting**

```
posting_id,employer,title_as_posted,level_evidence,metro,state,pay_min,pay_max,pay_unit,annual_min,annual_max,width_pct,flag
width_pct = (annual_max - annual_min) / ((annual_max + annual_min) / 2)
flag values: OPEN_ENDED | MULTI_LEVEL | MULTI_LOCATION | LEVEL_MISMATCH | VARIABLE_PAY | KEEP
summary row: median annual_min | median midpoint | median annual_max | median width_pct | n kept
```

*Paste the header into a spreadsheet, fill one row per posting, and drop every flagged row before taking medians.*

### Step 4: pull LCA records for target employers

Download the latest quarterly cumulative LCA disclosure file from the Department of Labor's OFLC performance data page, and keep the record layout open next to it. Filter to Certified status, your SOC code, your worksite metro, and the employers you are targeting. Annualize the offered wage before you compare anything (see step 5). Record the median annualized offered wage and the count of rows at each prevailing wage level. If at least a handful of rows sit at your level, compute a median for that level alone as well.

### Step 5: normalize every figure to annual base

OEWS annual estimates assume a full-time, year-round schedule of 2,080 hours, so multiply hourly figures by 2,080. Convert any LCA row whose WAGE_UNIT_OF_PAY is not annual to the same full-time basis. Remove bonus, commission, and equity from every figure. This band is base pay only.

> **Rule:** Annualize before you compare
>
> No figure enters the reconciliation until it is annual base pay on a 2,080-hour basis. An hourly LCA row read as annual is the most common false outlier in this build.

## How to reconcile three sources into a floor, target, and stretch

Pull one candidate target from each source, choose the target where the level-matched candidates converge, and then apply a fixed spread to get the floor and stretch. Do not average the three sources together. Each measures a different population, and an average hides which source is wrong.

#### From three sources to one band

1. **OEWS candidate** - Read the percentile for your level using the LCA anchors, for example about P34 for Level II
2. **Posting candidate** - Median min plus 40% of the gap to median max, not the posted midpoint
3. **LCA candidate** - Annualized offered-wage median at your level, or the overall median noted with its level mix
4. **Choose target** - Take the point where the candidates converge, with the best-matched source winning ties
5. **Apply spread** - Floor at 80 to 85% of target, stretch at 115 to 120%
6. **Record spread** - Write down the lowest and highest candidate and the gap between them

*Every source yields its own candidate target, and the band is set only after you see where they converge.*

### The three candidate targets

- **Candidate A, OEWS.** Find your level's percentile with the LCA anchors: about the 17th percentile for Level I, the 34th for Level II, the 50th for Level III, and the 67th for Level IV. OEWS publishes P10, P25, P50, P75, and P90, so for anything in between you interpolate on a straight line. For Level II, for example, take P25 plus 36% of the gap to P50. The interpolation is my own method. BLS does not publish a 34th percentile.
- **Candidate B, postings.** Use 40% range penetration on your cleaned medians: median minimum + 0.4 × (median maximum - median minimum). The 40% is my own inference from Glassdoor's finding that over 60% of in-range pay sits below the midpoint. It is not a sourced constant.
- **Candidate C, LCA.** Use the annualized offered-wage median at your prevailing wage level. If too few rows match your level, use the overall median and note the level mix next to it. A Level II-heavy mix pulls the figure down.

### Setting the band

Choose the target where the candidates converge. If two candidates agree and one sits apart, go with the two and investigate the third. When two candidates are close but you have to choose, take the one from the better level-matched source.

Then set the floor at 80% to 85% of the target and the stretch at 115% to 120%. Compensation teams build ranges the same way, and roughly 90% of organizations tie their range midpoints to the 50th percentile. The arithmetic checks out: with the floor at 80% of the target, the band spread measured from the floor is 50%. At 85%, it is about 35%. Both fall inside the traditional band widths in the table above. If you prefer to work from a spread, the formula is floor = target ÷ (1 + spread/2).

### Reading the spread between sources

The spread is the gap between your lowest and highest candidate target, as a percentage of the lowest. I have no published threshold for it. My working test: if all three candidates fall between your own floor and stretch, the sources agree well enough to defend. If one falls outside, find out why before the call.

#### Is the band ready to use

Horizontal axis runs from Candidates fall outside the band to Candidates fall inside the band. Vertical axis runs from Thin or mixed-level sample to 25 or more level-matched postings.

| Quadrant | What it means |
| --- | --- |
| Real disagreement | Find the misleveled source, usually the posting sample or the LCA level mix |
| Defensible band | Quote the target, keep the floor private, and bring the source trail |
| Not a band yet | Recheck the SOC code and collect more postings before any call |
| Agreement by luck | Treat it as provisional and widen the sample before relying on the floor |

*A band is ready only when the sample is level-matched and the sources land inside your own floor-to-stretch range.*

**Market pay band record**

```
MARKET PAY BAND
Role / level / metro:
SOC code and title:
O*NET duties checked against my work (Y/N):
Built on:
Refresh on:

SOURCE 1 - OEWS metro percentiles (vintage: ____)
P10 / P25 / P50 / P75 / P90:
Top-code marker on any cell (Y/N):
Percentile for my level (LCA anchor):
Candidate target A:

SOURCE 2 - Posted ranges (n kept = __, n dropped = __)
Median min / median midpoint / median max:
Median width (% of midpoint):
Candidate target B (40% penetration):

SOURCE 3 - LCA filings (employers: ____ ; n = __)
Annualized offered-wage median:
PW level mix I / II / III / IV:
Candidate target C:

RECONCILIATION
Target chosen, and which sources support it:
Floor (80 to 85% of target):
Stretch (115 to 120% of target):
Lowest candidate __ / highest candidate __ / spread __%
Any candidate outside floor-to-stretch, and why:
```

*Fill one record per role, level, and metro, and keep it with the date and source trail intact.*

## How this build goes wrong

Most bad bands come from a few predictable mistakes, and the band looks fine on paper until a recruiter pushes on one number. Each item below says what the mistake looks like, why it happens, and how to check for it.

1. **Multi-level ranges inflate the stretch.** An annual posting wider than about 60% of its midpoint usually covers several levels or locations. Leave it in the sample and your median maximum climbs. Check every width and search the posting text for "varies by location".
2. **Using the posted midpoint as the target.** More than 60% of in-range pay sits below the midpoint, so a midpoint target lands too high and you price yourself out at the screen. Use the 40% penetration discount and label it as an inference.
3. **The wrong SOC code.** A title-matched code can point you at a different occupation's wages. Check the O*NET duties against your actual work. If two codes fit, pull both and note which one your target employers use in their LCA filings.
4. **Comparing OEWS releases.** Each release pools about three years of panels, so an apparent year-over-year "raise" between releases is an artifact. Use one release and never compute a growth rate from two.
5. **Treating the LCA prevailing wage as the market rate.** The prevailing wage is a legal floor, and Level I sits near the 17th percentile. Use the offered wage, and remember it is still an attestation, not payroll.
6. **Leaving hourly LCA rows unannualized.** An hourly figure read as annual creates an absurd outlier that drags the median. Filter on WAGE_UNIT_OF_PAY before taking any median.
7. **Searching under one title.** The same job goes by different titles depending on where the employer is headquartered (details below).
8. **Including variable-pay roles.** Glassdoor found posted ranges were least accurate for sales and tipped work. For these roles, build the band on base pay only and treat the posted range with extra suspicion.

> **Watch out:** A posted minimum is not a guaranteed floor
>
> In Glassdoor's match, reported pay fell below the posted range in 22% of listings. Set your floor from the reconciliation, not from the lowest number an employer printed.

### The title problem, measured

The title problem is the failure I see skipped most often, and Refolk's index makes it measurable. In Refolk's index of professional profiles, 4,845 US profiles hold Compensation Analyst, Manager, or Partner titles, against 124 in the UK. Reward Analyst, Manager, or Partner titles reverse the pattern: 610 in the UK and 11 in the US.

| Title family | US | UK | US:UK |
|---|---|---|---|
| Compensation Analyst/Manager/Partner | 4,845 | 124 | 39:1 |
| Reward Analyst/Manager/Partner | 11 | 610 | 1:55 |
| Share using "Reward" | 0.2% | 83.1% | |

**83.1% - Share of UK profiles in the compensation title family that use "Reward"**

In Refolk's index the US share is 0.2%, so a one-title search on either side of the Atlantic misses a whole group of employers.

The split follows the employer's headquarters, not the worker's country. In Refolk's index, UK profiles with "Compensation" titles cluster at US-headquartered firms such as Google, Netflix, Apple, Citi, and Bank of America. US profiles with "Reward" titles cluster at UK and European firms such as bp, RELX, Barclays, Kerry, and Primark. A US seeker who searches only "Compensation" never sees the US offices of UK-headquartered employers that call the job "Reward". The same thing happens in other fields wherever headquarters vocabulary differs. Before you collect postings, list every title variant your target employers use.

## What to check before you use the band

Before you take the band into a recruiter call, check it against the list below. If any item fails, go back to the step it points to rather than patching the number.

#### Band ready for a recruiter call

- [ ] The SOC code is recorded, and its O*NET duties match my actual work, not just my title.
- [ ] The OEWS row shows P10 through P90, the release vintage, and whether any cell is top-coded.
- [ ] The posting sample holds 25 to 40 level-matched ranges with no open-ended or multi-level entries.
- [ ] I searched postings under every title variant my target employers use.
- [ ] Every LCA row is Certified, matched on SOC code and worksite metro, and annualized from WAGE_UNIT_OF_PAY.
- [ ] Every figure in the sheet is annual base pay on a 2,080-hour basis, with variable pay removed.
- [ ] Each candidate target names its source, and the posting candidate uses 40% penetration, not the midpoint.
- [ ] Floor, target, and stretch are written down as numbers, with the floor at 80 to 85% and the stretch at 115 to 120% of the target.
- [ ] The spread between the lowest and highest candidate is recorded, and any outlier has a written reason.
- [ ] The band carries a build date and a refresh date.

## How to keep the band current

A band goes stale from its fastest-moving source first, so set a refresh for each source on its own cycle. Postings change constantly, LCA disclosure files are released quarterly, and OEWS publishes once a year.

- **Postings: refresh before each new batch of applications.** Re-pull 10 to 15 fresh ranges and check whether your median minimum and maximum have moved. If they have, rerun the reconciliation.
- **LCA: refresh when a new quarterly file comes out.** Re-filter for your target employers. A shift in the level mix tells you more than a shift in the median.
- **OEWS: replace the row when a new release comes out.** Do not compare it with the old row. Swap it in and rerun candidate A.
- **Level changes mean a new band.** If interviews keep pricing you a level up or down, redo step 1 instead of stretching the old band.

When you apply to postings that made it into your sample, each one deserves a resume aimed at that posting's stated level and scope. Refolk tailors your resume and cover letter to each posting from your own history. That keeps your applications consistent with the level your band assumes. Whenever a recruiter asks for your range, quote the target and hold the floor in reserve. You can name each source and date behind every number.

## Frequently asked questions

### How do I figure out the market salary for my role without trusting one salary site?

Use three independent public records and reconcile them. Collect 25 to 40 posted ranges for your exact level, pull the government OEWS wage percentiles for your occupation code and metro, and pull employer-filed LCA wage records for your target companies. Each record is biased in a known direction: postings run wide, OEWS mixes all levels, and LCA is an attested offer. Where all three converge after level-matching, that is your target.

### Is a posted salary range what people actually get paid?

Often, but not reliably, and usually toward the low end. Glassdoor matched 147,568 US full-time listings to employee-reported pay. Reported pay fell inside the posted range 67% of the time, below it 22% of the time, and above it 11% of the time. Of the in-range salaries, over 60% sat below the range median. Treat a posted range as a grade, not as a promise about where you will land.

### How do I use BLS wage data for salary negotiation?

Use OEWS percentiles to anchor your level, not to quote as an offer. Look up your occupation code in your metro, record P10 through P90, and read off the percentile that matches your level using the LCA anchors: roughly the 17th, 34th, 50th, and 67th percentiles for levels I to IV. Never compare one OEWS release to another to infer pay growth, because each release pools survey panels from about three years.

### Can I use H-1B salary data for salary research if I am not on a visa?

Yes. The LCA disclosure files show the wage that specific employers attested they would pay for a role, by occupation code, worksite, and prevailing wage level. Filter to Certified rows, annualize the offered wage using the unit-of-pay field, and compare the offered wage, not the prevailing wage. Remember that the file records an attested offer, not final payroll, and Level I prevailing wage sits near the 17th percentile.

### What salary should I ask for once I have a band?

Quote the target and keep the floor to yourself. The target is the number where your level-matched sources converge. The floor, at 80% to 85% of the target, is your walk-away line. The stretch, at 115% to 120%, is the ceiling you can defend with sources if your experience sits at the top of the level. Bring the source trail so you can explain the number rather than assert it.

---

*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/market-pay-band-public-records*
