Skip to content
Canaria

Research-Grade Data Quality

Every record in Canaria's dataset passes through a multi-stage enrichment and validation pipeline. This page documents the coverage benchmarks, accuracy metrics, and known limitations that quant funds, enterprise teams, and academic researchers need to evaluate before using the data.

At a Glance

1B+

Unique jobs, by URL and source job key

10B+

Observations behind them

Near-real-time

Data acquisition cadence

100+

Enriched fields per record

Enrichment Field Accuracy

Coverage is measured on the delivered Job Postings files (October 2026); accuracy is measured on held-out gold sets.

FieldCoverageAccuracyNotes
SOC code (6-digit)99.9% of postings93% top-5, 73% top-1Uses title plus description context. Measured on a held-out gold set across 867 BLS 2018 codes.
Salary (stated or estimated)90.5% of postings42K+ cells, 95% CI42.6% of salaries are employer-stated; salarySource marks model estimates. Benchmarks fuse three sources: employer-posted pay, employee-reported pay and BLS OES. Estimates require valid state, zip code, and SOC.
Skills extraction89.8% of postingsNot published39,000+ technical skills in a 43,000+ entry taxonomy. Two-step extraction: a high-recall dictionary scan, then a title-aware relevance filter that removes off-role mentions. Tuned conservatively to keep legitimate skills: an independent multi-judge audit confirms about 99% of removed tags are false positives, with fewer than 1 in 500 legitimate skills affected. Concentrates the average from roughly 10 to roughly 8 skill tags per posting.
Seniority level99.94% populated94% primary accuracyThe classifier never declines: every posting it scores receives a band (Entry, Mid, Senior, Lead, Manager, Director, Executive). The 0.06% with no value were never scored.
Work mode (remote/hybrid/on-site)>85% (2022+)93% primary accuracyNULL means the posting states no work arrangement. Structured work mode fields became common on job boards in 2021-2022.
Location (city/state)97.8% geocodedNot publishedGeocoded to lat/lng; 89.1% also carry a metro area. Parse failures are tracked and reported separately.
Title normalization99.8% of postings84.58% top-5, 69.94% top-1Canonicalizes abbreviations, Roman numerals, and level indicators across ~100K canonical roles. Confidence scores included for downstream filtering.

See the classifier accuracy breakdown for top-1 and top-5 numbers per model, held-out gold set sizes, and class counts.

Deduplication Methodology

Deduplication runs in two stages. First, 10B+ observations (every sighting of a posting on its source) resolve to 1B+ unique jobs, unique by URL and source job key. Second, a cross-source step groups copies of the same job posted to several boards: vector similarity matching on job descriptions, near-duplicate detection for company name variants, and graph-based transitive closure across sources and reposting cycles.

Every posting keeps its own row. The isUniqueJobPosting flag marks one representative per duplicate group, and representativeJobID points each duplicate at it, so summing the flag counts each job once while a plain row count gives gross volume. No duplicate job ID appears in any delivery file.

Acquisition and Delivery Cadence

Sources are crawled continuously: near-real-time data acquisition across ATS portals and major job boards. ATS portal postings are captured within hours of going live on the employer career page. Job Postings Data is then delivered at least daily (several times a day on request), Company Data monthly or quarterly, and Salary and Skills data monthly. Snapshots used in the interactive explorer are refreshed weekly. Full-history flat file deliveries are available on request.

Data Vintage and Coverage Notes

The archive starts in 2022, and field coverage still varies by vintage:

  • Work mode (remote / hybrid / on-site)

    Structured work mode fields became common on job boards in 2021-2022, so early-2022 postings state an arrangement less often than recent ones. A NULL means the posting stated none.

  • Stated salary

    Salary transparency laws (Colorado 2021, NYC 2022, California and Washington 2023) drove a step-change improvement in stated salary coverage, so older postings carry fewer employer-stated salaries. Where a posting states no pay, a model estimate is provided when state, zip code, and SOC code are valid, and salarySource marks it.

  • SOC code and skills

    Enrichment runs over the whole archive, so a 2022 posting carries the same fields as one collected today. Skills coverage depends on description length: short descriptions yield fewer extracted skills.

Researcher note: Comparing null rates across vintages without accounting for structural breaks (remote work adoption, salary transparency laws) will produce misleading conclusions.

Evaluate quality firsthand

Request a sample dataset to run your own validation before committing. We provide sample files with full enrichment fields so you can test against your own benchmarks.