A number is only useful
when you know what it measures.

Different data tells different stories.

We preserve these classifications throughout ingestion, analytics, public pages and employer exports. They are never combined into a single salary median.

How compensation becomes comparable.

We preserve original values and annualize when the source discloses a pay period or an explicitly labeled annual assumption applies. Hourly wages assume 40 hours per week and 52 weeks per year (2,080 hours); weekly wages use 52 weeks, biweekly wages use 26 periods, and monthly wages use 12 months. These conversions do not imply hours actually worked.

Missing components remain null. An unknown bonus is not a zero bonus. Currency filters use original currencies; we do not combine USD, GBP or other currencies without an explicitly dated exchange rate. Signing bonuses and deferred awards remain separate components. Total compensation is the source’s reported value, rather than an unexamined sum of potentially overlapping components.

Career research and occupation wages.

Career profiles use O*NET 31.0 descriptions, tasks, title aliases and skills. Wage tables preserve the May 2025 occupational survey percentiles distributed through O*NET OnLine. A specialization can share a broader wage group; its published group title remains visible. Annual and hourly fields are separate source measurements, and we never fill a missing annual figure by multiplying an hourly figure.

These exports do not disclose employment or survey sample counts. No denominator is invented, and occupation wages stay outside employer compensation cohorts. Exact values and source bounds remain distinct. National profiles with substantive career content can be indexed; local wage pages require an exact median and a supporting quartile in the same native unit. User-entered comparisons describe brackets between disclosed percentiles, not a predicted personal percentile.

Sample sizes, statistics and privacy.

Observation cohorts are computed separately for company, normalized role, location, currency, reporting year, measurement type and component. We use linear interpolation for percentiles: sort the available values and interpolate the rank (n − 1) × p. Mean is the arithmetic mean; median is the 50th percentile.

• Fewer than five comparable observations: financial statistics are suppressed.
• Five to nineteen observations: the cohort is flagged as a small sample.
• A source-disclosed aggregate denominator below five: numeric statistics are suppressed.
• Undisclosed source denominators: displayed as unknown coverage; they do not qualify a page for indexing.
• Overlapping source aggregates are never added or turned into synthetic individual observations.
• Only approved, consenting, full-year realized contributor reports enter current annual public cohorts. Report counts do not verify unique people; repeated submissions by one person remain possible.

Private-industry payroll averages.

Industry pay uses official QCEW annual averages for privately owned establishments in real four- and six-digit NAICS industries. Average annual pay and average weekly wages are published native means, not salary medians, occupation percentiles or individual compensation packages. Covered wages can include bonuses, overtime and certain stock-option payments; they exclude employer benefits. Full-time and part-time jobs are combined, and the reference does not disclose hours or seniority.

Employment counts covered jobs, not distinct people, and establishments are reporting locations rather than unique companies. Broader industry groups overlap their detailed subindustries and must not be added as independent populations. Same-year state comparisons preserve ownership and industry scope but do not adjust for occupation mix, hours or cost of living. Source-suppressed pay and employment remain null even where establishment counts are public. Exact redundant parent/child scopes use one canonical search page while source records remain intact.

Endowment and foundation disclosures.

Institutional pay records preserve the named role and public filing separately from anonymous contributor cohorts. A Form 990 Schedule J full total can include base, bonus, other reportable compensation, combined retirement and deferred compensation, and nontaxable benefits. Previously reported deferred amounts overlap that disclosure and are never added again. Form 990-PF compensation, combined benefits/deferred amounts and expense allowances remain separate; an undisclosed base, bonus or full total remains unknown.

Calendar compensation year, filing tax year and fiscal year end can differ and are shown separately. Former and partial-year roles retain their source context. Record counts can include the same person across years and are not counts of independent employees. Employer postings retain native ranges, one-sided starting salaries and capture dates; historical or indexed employer text does not establish a current vacancy. These public disclosures are not used to compute anonymous employee salary medians.

Dates, freshness and historical changes.

Collection date describes when we retrieved a page, not the compensation year. A source rolling window with undisclosed compensation years remains labeled as such. DOL years refer to disclosure fiscal years. A posting’s publication date is separate from a salary reporting year. Historical comparisons require consecutive years with compatible cohort definitions; missing history is not fabricated.

Entity matching and deduplication.

Curated exact aliases can resolve employers to canonical companies. Citadel and Citadel Securities remain distinct operating entities. Fuzzy similarities are suggestions for administrator review, never automatic merges. Company-specific levels are preserved. Stable source identifiers drive upserts; we do not merge independently reported records just because their amounts match. Source changes preserve prior raw payload versions.

What a benchmark can tell you.

A benchmark describes its source cohort. It does not predict an individual offer, guarantee realized compensation, or establish a winning employer. Data can reflect selection bias, undisclosed periods, geographic differences, small samples, or source estimates. We show these limits where the figure is used.

Posted pay and time periods.

Job cards preserve stated hourly, weekly and monthly ranges. Annual equivalents use 2,080 hours, 52 weeks or 12 months and are not realized annual earnings. If a salary range does not disclose its pay period, a conventional annual interpretation is labeled as assumed; small unlabelled ranges have no annual conversion. Confirm the terms with the employer.

Indexability and commercial use.

Salary pages qualify for indexing only with supported USD total-compensation or base-pay medians from comparable cohorts with at least five samples and an identified career. Offered wages remain labeled separately from realized compensation. Thin profiles remain accessible with noindex. Source statistics stay attributable. Contributor licensing requires separate consent; contributor data is currently excluded from employer CSV exports while consent-aware licensed cohort exports are being implemented.