Glossary

Organisation and workspace

Organisation. The account that owns people and billing. Everyone you invite joins the organisation and can see all of its workspaces.
Workspace. One goal and the data for it: its pipelines, datasets, analytics, alerts and integrations. Most teams have one workspace per data product.
Member. A person in the organisation. Members can build, run and change anything in any of its workspaces.

Jason. The agent that builds, changes and repairs pipelines and answers questions about your data.
Brief. What you tell Jason: the goal, sources, fields, schedule and any attachments.
Chat. A conversation with Jason. Each message you send starts a turn that Jason works through.
Proposal. Jason’s plan before a build: sources, fields, schedule. You approve it or refine the brief.
Statement of work. The plain-language contract on every pipeline: what it collects, from where, how often, and what it publishes.

Pipeline. The repeatable definition of a dataset: sources, steps, checks and outputs. It is built and maintained by Jason and executed by workers.
Step. One stage of a pipeline: fetch a set of pages, extract fields, match products, check rows, publish a dataset.
Version. An immutable revision of a pipeline. Every build, change and repair creates one; earlier versions are kept.
Active version. The version scheduled runs use. Only a version that passed review can be active.
Candidate. A version that has been built but not yet reviewed and activated.
Changelog. The note attached to each version saying what changed and why.
Input. Something a run needs from outside the pipeline: a dataset from another pipeline, or a parameter such as a region or a date range.
Parameter. A named input with a type and a default that can be changed per run.

Run. One execution of a pipeline version. It fetches, extracts, processes, checks and publishes.
Scheduled run. A run started by the pipeline’s schedule. Manual run and webhook run are the other two triggers.
Checkpoint. The point a run has safely reached. A failed or cancelled run can resume from its last checkpoint.
Held row. A row that failed a check. It is kept with the reason instead of being published.
Auto-heal. The automatic repair process that starts when a run fails persistently.

Dataset. The published table of rows a pipeline produces. A pipeline can publish more than one.
Dataset version. The rows as they were at one publish. Dashboards, share links and destinations read the active version unless pinned to another.
Row. One record. Every row carries provenance back to the run, step, fetch and source page that produced it.
Provenance. The chain from a value back to the page it came from, including a screenshot of that page.
Column format. How a column is displayed and exported: currency, percent, URL, date and so on.

Integration. A connection to another system. Notification integrations (Slack, email) receive alerts; destination integrations (Sheets, webhooks) receive data; service integrations (MCP, CLI) let other tools read your data.
Alert rule. A condition on a dataset or a run that fires an alert to the channels you choose.
Dashboard. Widgets generated from one or more datasets, updated on every publish.
Newsletter. A scheduled written brief generated from fresh data and emailed to recipients.
Share link. A read-only link to a dataset, dashboard or newsletter issue for people outside the workspace.
MCP. The protocol that lets Codex, Claude Code, ChatGPT, Cursor and Teams read your datasets and ask Jason for work.

Row. The unit of pricing. You pay for rows that passed checks and were delivered; retries, blocked pages and repairs are not counted.
Row limit. The monthly ceiling for a workspace. The sidebar meter shows how many rows are left this period.
Top-up. Extra rows added to the current period.