Glossary

The words Graphdom uses, in the order you meet them.

View as MarkdownAsk Claude about this page

Studio, the CLI and these docs use the same words with the same meaning.

Term Meaning
Site A web portal: one or more origins, such as https://quotes.toscrape.com.
Task What an expert does on a site, and the unit a spider automates. For example, “search quotes by author”.
Expert The person who knows the task and records it. Experts use Chrome, never Studio or the CLI.
Session One recording of a task by one expert. A directory in the archive.
Action One thing the expert did in a session: a click, a typed value, a choice, a submit.
Archive The directory where sessions are written: archive/<host>/. See The archive.
Relay The local server the Chrome extension sends recordings to, when experts record in their own browser.
Sandbox The site served offline from the recorded sessions. Spiders are tested here before they touch the real site.
Level How far the sandbox may depart from the recording. Level 1 answers exactly as recorded. Level 2 tolerates values that change on every visit, such as tokens and nonces. Level 3 answers inputs no session recorded and marks those answers as generalised.
Spider The program that performs a task. It is deterministic and versioned (v1, v2…).
IR The spider’s source: a JSON document (spider.json) of states, steps, inputs and outputs. The Playwright script is compiled from it.
Input A value the spider types or picks, such as a search query. Each session recorded an example of it.
Secret An input whose value is never stored, such as a password. You pass it at run time.
Output A value the spider reads from the site and returns, such as a list of results.
Test What a session expects: the same requests in the same order and the same final page. Every spider must pass the tests of all its sessions.
Build The step that turns a task’s sessions into a spider and proves it against their tests.
Runtime Graphdom’s own engine for running a spider. The other target is a compiled Playwright script.
Effect Every step is either a read (looks, navigates) or a submit (sends data the site keeps). The runtime asks before a submit.
Drift The real site no longer matches what was recorded. The spider stops and reports it instead of guessing. In Studio, a task’s Drift lists its runs that drifted, each with the page before and after, and its repair (graphdom build --repair).