DATA_DICTIONARY.md

# Data dictionary: pracadata/website-technologies

Four configs, all Parquet (zstd). Every row has `source`, the attribution to the HTTP Archive. Dates are the first day of the crawl month (`2026-10-01` is the October 2026 crawl).

## Shared definitions
- **origin**: the crawled root page as the HTTP Archive records it, with scheme and trailing slash (`https://www.example.com/`). `example.com` and `www.example.com` are different origins.
- **reg_domain**: registrable domain of the origin (public suffix list), for example `example.co.uk`. Empty for IP addresses and hosts without a public suffix.
- **rank_bucket**: Chrome UX Report popularity bucket of the origin: `1000`, `5000`, `10000`, `50000`, `100000`, `500000`, `1000000`, `5000000`, `10000000`, `50000000`. When desktop and mobile disagree, the smaller (more popular) one is used.
- **technology / categories**: name and categories as detected by the HTTP Archive's fork of Wappalyzer. A technology can have several categories (a sorted list).
- **versions**: sorted list of versions the page exposes (empty when none). Only values that look like a version are kept. Best effort.
- **detected_desktop / detected_mobile**: whether the technology was detected by the desktop and/or the mobile crawl. At least one is true.
- **source**: `Derived from the HTTP Archive (<Month Year> crawl)`.

## tech_by_domain (top 10,000 origins, latest crawl only)
One row per origin × technology.
| Column | Type | Meaning |
|---|---|---|
| `crawl_date` | date | Crawl month (first day) |
| `origin` | string | See above |
| `host` | string | Host of the origin |
| `reg_domain` | string | See above |
| `rank_bucket` | int | See above (here always 1000, 5000 or 10000) |
| `technology` | string | Detected technology |
| `categories` | list of string | Its categories |
| `versions` | list of string | Detected versions |
| `version_major` | int | Leading integer of the first version, or null |
| `detected_desktop` | bool | Detected by the desktop crawl |
| `detected_mobile` | bool | Detected by the mobile crawl |
| `source` | string | Attribution |

## tech_changes_monthly (top 10,000 origins, latest pair of months, `both_clients` only)
One row per origin × technology × change between two consecutive crawls. Only origins present in **both** crawls, so a technology cannot be "removed" by an origin that left the crawl.
| Column | Type | Meaning |
|---|---|---|
| `month_from`, `month_to` | date | The two crawls compared |
| `origin`, `reg_domain`, `rank_bucket` | | As above (rank of `month_to`) |
| `technology` | string | Technology that changed |
| `change_type` | string | `added`, `removed` (not detected in `month_to`; this is not proof of removal) or `version_changed` |
| `old_versions`, `new_versions` | list of string | Versions before and after |
| `confidence` | string | In this dataset always `both_clients`: the origin is in both the desktop and the mobile crawl in both months and both clients agree on the change. The Apify Actor also serves `single_client` changes, which reverse much more often |
| `source` | string | Attribution |

## tech_adoption_monthly (complete)
One row per crawl × technology × category × cumulative rank bucket.
| Column | Type | Meaning |
|---|---|---|
| `crawl_date` | date | Crawl month |
| `technology`, `category` | string | A technology appears once per category: filter before summing |
| `rank_bucket` | int | **Cumulative**: `10000` counts the origins of the 1k, 5k and 10k buckets |
| `origins` | bigint | Origins in the bucket where the technology was detected |
| `origins_in_bucket` | bigint | Origins in the crawl within the bucket (including those with no detection) |
| `share_of_origins` | double | `origins / origins_in_bucket` |
| `source` | string | Attribution |

## technology_catalog (complete)
| Column | Type | Meaning |
|---|---|---|
| `technology` | string | Name |
| `categories` | list of string | Categories |
| `first_crawl_seen`, `last_crawl_seen` | date | First and last crawl in this build with the technology |
| `origins_last_crawl` | bigint | Origins with the technology in the latest crawl |
| `source` | string | Attribution |

The catalog has names, categories and counts only. Nothing is copied from the Wappalyzer repository (no detection rules, no icons, no descriptions).