Question 1

What does Datadory cover in systems software data?

Accepted Answer

Twenty-four primary datasets across five shelves: package registries (npm, PyPI, crates.io, deps.dev, Repology), vulnerability intelligence (OSV, NVD), developer-platform telemetry (GitHub, GitHut, Stack Overflow), adoption and lifecycle benchmarks (StatCounter, W3Techs, Steam, TIOBE, PYPL, endoflife.date) and hardware fleet records (Backblaze, AnandTech Bench, TOP500, Icecat, Amazon, Census, GTDC). One adjacent application-software record pools the total to 25.

Question 2

How far back do the package registries reach?

Accepted Answer

npm carries publication timestamps back to its 2010 inception and per-package download counts from 2015-01-10. crates.io holds the full publication history of roughly 320,000 crates, with surviving records from the 2014-era launches. PyPI spans 877,215 projects and 9,420,495 releases, with download aggregates available as history rather than topline.

Question 3

Can I track one vulnerability across multiple ecosystems?

Accepted Answer

Yes - that is the design of the pair. OSV aggregates 848,582 advisories across 38 ecosystems with affected ranges down to package version and commit, while NVD holds roughly 381,418 CVE records with CVSS scores and applicability matched to product version, covering 2002 forward. Alias fields such as CVE-2019-10906 let one flaw map across both corpora.

Question 4

Where does operating system market share data come from?

Accepted Answer

Three vantage points rather than one. StatCounter measures desktop, mobile, tablet and console share worldwide and by country from January 2009 onward, at daily to yearly grain. W3Techs reads the server side, splitting Unix-family against Windows across surveyed websites. Valve's Steam Hardware Survey adds monthly GPU, VRAM, CPU-core and OS-version distributions from gaming PCs.

Question 5

Which datasets track programming language popularity over time?

Accepted Answer

Four, each with a different instrument. TIOBE has rated the top 50 languages monthly since June 2001, with annual rankings back to 1986. PYPL computes monthly search-interest shares for 29 languages across six geographies back to June 2004. GitHut counts quarterly pull requests, pushes, stars and issues per language from 2012-Q2 to 2024-Q1. Octoverse adds annual reporting across 180M-plus developer accounts.

Question 6

Is there raw hardware reliability data I can model on?

Accepted Answer

Backblaze ships one row per drive per day for roughly 341,000 drives - about 530 million cumulative drive-days since April 2013 - carrying every reported SMART attribute beside failure flags and datacenter placement. AnandTech Bench complements it with archived CPU, GPU and SSD results across hundreds of standardized tests, frozen at August 2024.

Question 7

How detailed is the developer survey microdata?

Accepted Answer

Per respondent, not topline. The 2024 Stack Overflow table carries 65,437 respondents by 114 columns, with comparable editions spanning 2011 through 2025 across 185 countries - enough to run your own crosstabs on operating systems, databases, languages and salaries instead of quoting someone else's summary charts.

Question 8

How is systems software data delivered?

Accepted Answer

As API pulls, bulk files, or warehouse-native writes - daily, weekly, or hourly, your call. You pick the channel and the cadence; extraction, typing and schema stability happen upstream. Every record arrives with the same columns and join keys regardless of route, so moving a feed from weekly to hourly is configuration, not migration.

The manifest

npm Public Registry API

PyPI - Python Package Index API and Stats

crates.io - Rust Package Registry API

deps.dev - Open Source Insights API

Repology - Packaging Hub

OSV - Open Source Vulnerability Database

The lay of the water

Pick your catch

Who fishes here

Frequently asked questions

See the rows before you commit