The Own-Role Pay Benchmark, Triangulated From Public Sources

Before you name a number in a negotiation, you need one pay range you can defend line by line. This guide is the ordered procedure for building it: a single low, median, and stretch figure for your exact role, level, and metro, reconciled from three public inputs that each define the job differently, publish on different lags, and mean different things by the word "range." It is for a job seeker who is done averaging whatever a salary tool shows and wants a number traced to dated sources.

The ranking pages hand you a list of tools and tell you to average them. None of them reconcile sources that disagree on what your job even is. That reconciliation is the work, and it is what follows.

What "triangulated" means and why one source is never enough

A triangulated pay range is a single low, median, and stretch figure built by reconciling three independent public inputs - government wage percentiles, live posted ranges, and crowd-sourced self-reports - after correcting each for the lag and geography that makes it lie. No single input is trustworthy alone.

Each source fails in its own direction. Government data is authoritative on structure but old. Posted ranges are current but often gamed wide. Self-reports are timely but unverified and self-selected. Triangulation works because the three errors do not point the same way, so reconciling them cancels bias that averaging blindly would preserve.

The whole method funnels toward one number. Once you have defended the midpoint, the range around it is mechanical: the practitioner convention puts the minimum near 85% of the midpoint and the maximum near 115%, so every step before the last one exists to nail the center.

Where the numbers come from, source by source

The three inputs each have a named public home. Knowing what each one actually measures - and its refresh cadence - is what lets you correct it rather than take it at face value.

The government floor is the BLS Occupational Employment and Wage Statistics program (OEWS). It publishes annually with a May reference date and covers about 530 metropolitan and nonmetropolitan areas, reporting the 10th, 25th, 50th, 75th, and 90th wage percentiles per occupation and area. Its weakness is lag: the sample of roughly 1.1 million establishments is collected over three years, and publication adds about another year. The May 2025 reference estimates were released on May 15, 2026.

Posted ranges come from live job listings for your role, level, and metro. Their value is currency, and their trap is width. A band running $130k to $500k satisfies a transparency law and tells you nothing. Regulator guidance, such as Colorado's, requires the posted range to span the pay the employer actually believes it will offer, but enforcement is uneven, so you filter for width yourself.

Self-reported figures are crowd-sourced compensation points for the same title and level. They are timely but unverified and self-selected, so they matter most as a third bearing rather than a primary anchor.

Mapping your title to a SOC code without guessing

The Standard Occupational Classification code is the key that unlocks government wage data, and the single most common way this whole exercise goes wrong is picking the code from your job title alone. Verify against the documented duties instead.

The route is ONET OnLine. In the Occupation Keyword Search box, type your job title or an ONET-SOC code, then compare the Description and Tasks against your actual work history before you capture the code. The taxonomy is granular: in the SOC system all workers fall into one of 867 detailed occupations, rolled up into 459 broad occupations, 98 minor groups, and 23 major groups, and O*NET's version encompasses more than 55,000 job titles. One marketing title can legitimately map to several codes.

Mismatches break most often on alternate titles and on occupations that split or roll up. "Product Manager" and "Marketing Manager" land on different codes with different medians. If you pick on the label, you inflate or deflate your floor before you have read a single wage figure. The test is simple: does the Tasks list describe what you actually do most days? If not, you have the wrong code.

Record one captured code plus a note of the close alternates you rejected and why. That note is part of your audit trail, and you will want it when someone asks how you arrived at your floor.

The triangulation procedure, step by step

Here is the full method in order. It runs about three hours end to end and produces a low, median, and stretch figure with every line traced to a dated source. The sequence is midpoint-first: you defend the center, then derive the edges.

Two of these steps do the heavy correction: aging and location. Skip either one and your reconciled center inherits a bias you cannot see. The next two sections treat them on their own.

Aging every figure to a common date

Averaging figures from different years drags your median in the direction of the oldest data, so before you reconcile anything, escalate every input to the same month. This is the correction that undoes government staleness.

BLS documents the method itself: age wage-survey data forward using the Employment Cost Index. The current escalation rate is modest and defensible. For the 12 months ending June 2026, compensation costs for private industry workers rose 3.3%, with wages and salaries up 3.1%; for civilian workers, wages and salaries rose 3.2% over the year. A practitioner figure of roughly 3 to 4 percent annually holds up.

The arithmetic is a compound-growth adjustment. A median with a May 2025 reference date, brought forward roughly a year at 3.1%, gains about 3%. An OEWS figure that is effectively four years old once you count the collection window gains closer to 12%. That is not a rounding error; it is the difference between a floor that looks fresh and one that predates the last two rounds of raises.

Aging is also where posted ranges earn their keep. They are the freshest input, so they need the least adjustment and serve as the reality check on everything else. If your aged OEWS median lands far below the freshest live postings, trust the postings and investigate the gap - your SOC code or your metro may be off.

Adjusting for location without confusing labor and living costs

Location can move pay more than a promotion does, so if any input is national you must shift it onto your metro before you set the midpoint. The right tool is a cost-of-labor differential, not a cost-of-living index.

The clearest public cost-of-labor index is the federal GS locality pay schedule. It is built from BLS data comparing federal and non-federal pay in each labor market, which makes it a measure of what employers pay, not what rent costs. It applies multiplicatively: locality-adjusted salary equals base pay times (1 plus the locality rate).

The 2026 rates run from 17.06% in Rest of U.S. to 46.34% in San Jose-San Francisco-Oakland, averaging 23.89% across 58 areas. The SF-versus-Rest-of-U.S. gap is 25%, which rivals a full pay-grade jump. That is why location is adjusted before the midpoint, not after: get the geography wrong and you are anchoring to the wrong market entirely.

The trap here is using a cost-of-living index instead. Rent-based COL overstates pay in expensive-but-low-wage metros, because a place can be costly to live in without employers paying a premium. Move wages with the labor differential.

Turning the reconciled midpoint into low, median, and stretch

Once you have a single reconciled midpoint - all inputs aged to one month and sitting on one metro's labor basis - the range is mechanical. Set the minimum near 85% of the midpoint and the maximum near 115%, and widen only if your level warrants it.

The convention is well documented. Minimum sits at roughly 85% of the midpoint and maximum at 115%, giving about a 30% spread. Most sources aim for a 30 to 50 percent spread from minimum to maximum for typical roles; under 20% is too restrictive to justify pay differences, and over 60% makes those differences hard to defend. Mercer reports spreads typically 40 to 60 percent wide.

Spread widens by level. Use the band that matches your role.

There is a structural reason the stretch end is harder to defend than it used to be. A 2026 NBER working paper finds that posting transparency reduces pay dispersion but has no effect on average wages: transparency raised posted and realized salaries without cutting vacancies or raising requirements. In plain terms, the number you were hoping to widen is being compressed. Your defensible upside now comes from level and scope, not from betting on a wide posted band.

A quick check before you commit: your compa-ratio is your target salary divided by the range midpoint, where 1.0 means paid exactly at the midpoint. If you are reading a below-1.0 ratio as evidence you are underpaid, contextualize it by tenure and time in role first - a new hire below the midpoint is normal, not a grievance.

Role / level / metro spec: ____________________
SOC code (verified against Tasks): ____________________
OEWS median (dated ____, aged to ____): $______
Posted midpoints, filtered <60% spread (n=__): $______
Self-reported center (aged to ____): $______
Location basis applied: ______ (multiplier ______)
--- reconciled center ---
Median (midpoint): $______
Low  (0.85 x midpoint): $______
Stretch (1.15 x midpoint, or level spread): $______

How this goes wrong, and the check for each

Most bad benchmarks fail in one of seven predictable ways. Each has a signal that looks like signal but lies, and a specific check that catches it.

• OEWS staleness read as current. A "fresh May 2025" median can be up to four years old once you count the three-year collection window plus the one-year publication lag. It looks current because of the recent release date. Check: age it forward with the ECI before comparing to anything.
• SOC mismatch on title alone. Picking the code from the label maps "Product Manager" and "Marketing Manager" to different medians and inflates or deflates your floor. Check: confirm the Tasks list matches your actual duties, not just the title.
• Posted-range gaming. A very wide band, such as $130k to $500k, satisfies the law but has no informative center. It looks like a data point. Check: discard postings whose min-to-max spread exceeds about 60% before averaging.
• Locality versus cost-of-living confusion. A rent-based COL adjustment overstates pay in expensive-but-low-wage metros because it measures living cost, not labor price. Check: move wages with the cost-of-labor differential.
• Blending vintages. Averaging a 2022 self-report with a 2026 posting drags the median down invisibly. Check: age every input to the same month first.
• Midpoint drift. If your blended midpoint lands 15% below what competitors currently post, the range is broken before you build the edges. Check: sanity-check the midpoint against the freshest live postings.
• Compa-ratio misuse. Reading a below-1.0 ratio as underpayment ignores tenure and performance. Check: contextualize the ratio by experience and time in role.

Reading supply so you know when to push past the median

Anchoring to the government median leaves money on the table where local talent supply is thin, because in scarce markets the same title commands a higher percentile. Supply is a negotiating lever, and you can read it before you name a number.

The signal is how many people hold your title in your market. In Refolk's index there are 348,384 profiles with the title Software Engineer in the United States against 22,178 in Germany - a 15.7x difference for the same title. Where supply is that much thinner, the market clears higher up the percentile band, so aiming at the median undersells you.

The top US employers for that title in Refolk's index include Google, Figma, Microsoft, and Glean, concentrated in San Francisco; in Germany the top employers include Zalando, TeamViewer, and Helsing, concentrated in the Berlin Metropolitan Area. That employer concentration tells you which live postings are worth weighting most in your posted-range pull.

To read your own supply, build a peer set narrow enough to match one SOC code and level cleanly. That is where a people search removes the friction of hand-counting profiles across sites.

Refolk builds that peer set in one query, and because it also writes and tailors your resume to each posting and scores how well you fit, the pay range you build here plugs straight into the applications where you will use it. Start from Refolk when you want the supply read and the application work in one place.

Keeping the benchmark current

A pay range decays the moment you file it, so treat it as a document with a shelf life rather than a one-time answer. Two things go stale: the vintage of your inputs and the market itself.

Re-pull when either moves. OEWS refreshes once a year on a May reference cycle, so a new release is a natural prompt to redo steps 3 and 6. Between releases, the live postings drift faster, and 77% of companies with formal salary structures review them annually while 94% use market data - so the bands you are benchmarking against move on roughly that cadence too. If more than a quarter has passed, re-run the posted-range pull and re-age everything.

Before you rely on the range in any negotiation, run the final check that it is still internally consistent.

The point of the audit trail is not neatness. It is that when someone across the table asks how you got to your number, you can walk them from the SOC code to the aged median to the location factor to the 85/115 edges, and every step has a date on it. That is what makes the range defensible, and defensible is the only kind of range worth building.