Topic
Data quality
Coverage, freshness, accuracy and how to evaluate a dataset before buying.
- How Fokals intent scores are built, and how to read oneEvery Fokals intent score is built from dated things a company did in public, and carries them as evidence. The signals, the scale, the surge flag and how to read a row.
- How to evaluate a company data vendor in ten questionsTen questions for any vendor of company data, from coverage to change management, each with a document to request and a test to run on a sample. Our own answers are included, each with the public document behind it.
- How we detect the technologies on a company websiteWhat stands behind a row of technographic data: the evidence a detection rests on, how a technology is dated, and how an adoption, a confirmed removal and a platform migration are recorded.
- Why we read job postings from the company's own boardA job posting can be read where the employer published it or where it was copied to. The source decides how many records a role produces, whose role it is and what its dates mean.
- Why we write each table once: point-in-time data by designA table corrected in place shows what is known today, not what was known then. The five rules that keep Fokals tables as first written, what they guarantee and how to check them.
- Building a point-in-time company dataset for back-testsA back-test is only as honest as the dates in its data. This guide shows how a dated, write-once record lets you key company signals by when they were known and query as of a date.
- Training and evaluating models on dated company signalsA model on company signals is only as honest as its feature table. This guide builds one as of each date, splits by time and names the properties of the record that keep it safe.
- Datarade: how the data marketplace works for buyersA buyer's reading of Datarade's own pages: free search, listing fields, sample previews, the request form and the contract, with a way to cut a long list to three providers.
- How to evaluate company data providers on DataradeSix rows, each tied to a field on a Datarade listing and to a check you can run: how to score providers on coverage, freshness, sourcing, delivery, licence and a sample for the companies you track.
- Modelling licensed company data with dbtA working dbt layout for licensed company data: sources, staging models, tests on keys and versions, freshness set from each table's closing rule, and the one table worth a snapshot.
- What alternative data scouts look for in a new datasetWhat Neudata and Eagle Alpha publish about how funds judge a new dataset, set beside five checks you can run on a company-level dataset, with the Fokals tables as the example.
- Data coverageData coverage is how many of the companies you track a dataset contains and how fully each record is filled in. How to measure it on your own list, and how Fokals states coverage dataset by dataset.
- Data dictionaryA data dictionary says what every table and column of a dataset means. This entry lists what a useful one states and how to test it against a sample.
- Data freshnessData freshness is the age of a record's observation when you use it. How it differs from refresh cadence, how to measure it in your own warehouse, and the cadence of each Fokals dataset.
- Data provenanceData provenance records where data came from, when it was observed and what was done to it. What a record should hold, and how source, observation time and label version appear in Fokals tables.
- First-party dataFirst-party data is what an organisation collects directly from its own customers and operations. Two senses of the term, and why the second matters in due diligence on a data vendor.
- Label versionA label version names the frozen rules and model behind a dataset's labels. Why it matters for comparing periods and for products built on labels, and the versions used in Fokals data.
- Look-ahead biasA back-test that uses information from after the date it simulates looks better than any real decision could have been. How it enters a test and how to find it.
- Point-in-time dataData stored as it stood on each past date, with no later corrections. How the two dates of a record work, an as-of query and the mistakes that break a back-test.
- Survivorship biasStudying only the companies that lasted makes past results look better than they were. Where the error appears, how to test for it and how to keep delisted names.