Skip to content

Data map

Every table in your organization on one page, with the lines that connect them. If your organization has more than one Space, it shows the Space you’re in. Use the Data map to answer “where does this number come from?”, “what would a change here touch?” and “is anyone still using this?”.

Admins open it from Advanced → Data map in the sidebar. Anyone can open it with Show in data map on a table’s page, which picks out that table’s pipeline.

A switch at the top left chooses Graph, for tracing a pipeline, or Tables, for judging an estate. Both show the tables you have access to, and the count of tables and metrics sits under the bar.

  • Filters gathers the toggles that narrow the map, with a badge counting how many are on. Metrics only narrows either view to your metrics and what feeds them.
  • The search box covers tables and columns, so a half-remembered column name finds its table. Picking a result selects the table: it opens its panel and, in the Graph view, dims the map to what it touches. It doesn’t filter, and clearing the box doesn’t undo the selection.

Each table is a node and each input an arrow, drawn left to right in the direction the data moves.

A workflow that writes tables is a narrow grey chip between the tables it reads and the tables it writes, with a dotted arrow out to each table it writes (written by a workflow, rather than built by Ronja from its inputs). Hover the chip for its name and how many tables it reads and writes; click the name to open it. A workflow that writes no tables isn’t drawn. See Derive a table with Python.

  • Group: Feature draws one card per feature, which you open to see its tables; start there when you don’t know where something lives. Table draws one flat flow.
  • Hide singles, under Filters, drops tables connected to nothing, usually most of the noise.
  • Click a card to open it; double-click an open card to close it. Hovering a card shows Open feature, for the feature’s page, and Expand / Collapse.
  • Expand all and Collapse all, at the bottom left beside the zoom controls, act on every card and refit the canvas. Opening one card keeps it in place and never changes your zoom.
  • A line between closed cards stands for every dependency between their tables, with a count when there is more than one; hover the count for which feature feeds which.
  • Drag a node to move it; Reset layout then appears at the bottom left and puts everything back.

Lens, beside Group, colours every table with one answer: None (the default), Freshness, Usage, Cleanup, Health or Builds. Hover a lens name for its question; a legend under the control explains the colours. A lens never colours a workflow chip.

Grey means no answer, not a zero: you can’t read the table, or its history couldn’t be loaded. A collapsed feature card shows a slim bar of how its tables split (hover for counts), and its border takes the worst colour inside.

Whether a table’s data is as recent as it should be. Hover a table for the rule that judged it.

Table How it is judged
Has a freshness health check By the check’s window: Getting old once the interval has passed, Stale when the check would flag it, so a table meant to refresh daily turns red after a day and a half. Amber is a note: a run late but inside the check’s slack.
Loads from outside: synced, pushed as files, or written directly by Ronja, a workflow or the API Only by a freshness check. Without one it stays uncoloured, and the tooltip says it isn’t tracked until a refresh cadence is declared: only the people who know the table can say how often it should refresh.
Built from other tables By its inputs (below).
Foundation Not judged, unless it has a freshness check.
Marked never stale, such as a calendar Never judged. Ask Ronja to mark it, or set it over the API; this also silences its freshness check.

The legend reads Stale (red), Getting old (amber), Fresh (green), Unknown (grey: no write recorded, or the rule couldn’t be loaded) and Never stale, for every table the lens doesn’t judge; those stay uncoloured, as do metrics, which store no data. Age is all this lens measures.

A table inherits red from its inputs, never amber, even through a table marked never stale; its tooltip names the culprit, as in stale via orders_raw. A table built from others is Fresh only when at least one input is judged, because it has a freshness check, and none is stale; with no judged input it stays uncoloured (Judged by its inputs, and none of them is tracked).

Freshness and Health both look up the chain, by different rules: Freshness judges age against a refresh window someone declared, Health judges explicit failures. So a table can be red on Freshness because an input is past its window while Health says nothing about it, and the other way round.

A dashed outline marks a table with a freshness check, red while the check fails (Its declared freshness check is failing.). Add one to say “this should rebuild nightly”: it sets the table’s window on this map and raises an alert when it doesn’t.

A partial tag means part of the table’s lineage is off the map. When the tooltip names a cause, it is a real one but not necessarily the only one.

How much each table was read in the last 90 days, from No reads in 90 days through Barely read to Heavily read, with a bar for the amount. A table not read in a year is hatched (Not read in a year).

A read is a recorded query, chart render, workflow run, build or metric calculation. Opening a table’s page or its data preview doesn’t count, so a table people look at but never query reads as unread. Working out a metric counts as a read of the metric and keeps the table it reads counted as in use, without making that table more popular.

Healthy, Waste (nothing reads it, and it isn’t past any refresh window declared on it), Broken (something reads it, but its data is Stale: fix it rather than remove it) or Dead (neither). Hover for why, and whether the table is part of a dead chain. Grey Not judged covers the metrics, drafts and conversation tables the Tables view won’t let you tick.

Broken is a prompt to look, not a fault: someone declared in a freshness check how often the table should refresh, and that window has passed. A table with no freshness check of its own is never Broken or Dead; unread, it is Waste. Mark one never meant to refresh as never stale. The horizon is set by Not read in on the Tables view, for both views.

Passing, Failing, Check error or No checks, from the table’s health checks; the tooltip says how many are failing where it can. A small amber dot marks a table whose column scan flagged possible personal data.

Input failing, hatched amber, marks a table built on something failing, however many steps back: an input whose updates keep failing, an input with a failing Admin-protected health check, or a connection, the table’s own included, that needs re-authorization or has stalled (When something a table is built from is failing). The tooltip names the oldest problem you can see and counts the rest.

It wins over the table’s own Passing, Check error, No checks and checks that have not run yet, because the tables built on a broken input usually have no checks of their own. A failing check of the table’s own still reads Failing.

Ready, Building (it pulses), Invalidated or Build failed. The bar under the name is how long the last build took, and the tooltip gives the outcome and duration.

Waiting on an input, hatched amber and in the legend only when some table is in that state, means an upstream change has reached the table while one of its inputs is still catching up with it, or is held by a sender still sending several tables over the API, about 15–20 minutes at most. The table then rebuilds once, over everything new. The tooltip says how long it has waited, and names the input when Ronja recorded it.

Click a table and the rest of the map fades, leaving its blast radius: everything it is built from and everything built on it, however many hops away. A strip at the bottom reads, for example:

orders_mart · Fed by 3 tables · Feeds 7 tables · Also directly used by 1 app, 2 automations (2 structural)

Fed by and Feeds count the whole chain. Also directly used by counts only the apps, automations and notes that name this table; select a table further down to see its own. Structural consumers stop working if the table changes; the rest only mention it.

With the Freshness lens on, selecting a table that is red because of an input draws the path back to it in red and fades every other connection; a hop through a collapsed card isn’t drawn, but the strip names the cause. Clear, a click on empty canvas, closing the panel or Escape brings the whole map back.

The same estate as a sortable list: faster when you know roughly what you want, and the view for deciding whether a table still earns its keep.

Each row shows Name, Feature, Kind, Level (hops from a source), Status, Inputs and Updated. Not updating in Status means the table’s rebuilds keep failing while it serves its earlier data; hover for since when and why (When a table stops updating). The same marker shows on the table’s card in the Graph view. Five more columns:

Column What it says
Last read When something last queried the table (what counts); > 1 year means not within a year.
Last built When data was last written: unknown if never recorded, building while a build runs. A warning marker means the table is past its Freshness window; tables that rule doesn’t judge never get one, so no marker doesn’t prove the data is recent.
Health The table’s health checks, or Input failing when something it is built from is failing.
Used by How many other tables read it, hidden ones included. Struck through: all of those are unused too. A table a metric reads is never a cleanup candidate; hover to see it.
Size The storage it takes.

A label at the end of a row reads Waste, Broken or Dead (Cleanup); a healthy table has none. Sort by clicking Name, Level, Last read, Last built or Updated. Group here is two switches, Feature and Source, which nest the list. Show hidden tables, under Filters, adds hidden tables. Show pipeline on a row opens the Graph view on that pipeline.

Preset, above the list, narrows it with live counts: All; Cleanup candidates, unused tables with no live dependent still in use, the shortlist to start from (judged by recorded queries; other consumers are checked when you preview a removal); Broken, past the window its own freshness check declares, which can’t be ticked for removal (if the window no longer fits how the table loads, change the health check); Waste; and Not read in a year. Not read in sets the horizon: 30, 90 (the default) or 180 days.

Under Cleanup candidates, tables that feed each other are one Dead chain row, so raw → staging → mart is one decision. Grouping is ignored under this preset, since a chain can cross features.

  1. Tick the tables to remove, or a Dead chain to take the whole chain. A footer shows the count and the storage they hold.
  2. Click Preview removal →. Ronja checks what depends on each table.
  3. Read the preview. A table something else depends on is Blocked and unticked, naming what depends on it: add that too if it is dead, or rewrite it first. A table mentioned in a note enables only after you click Mentioned in N places — show before removing and read them, since Ronja can’t track a mention.
  4. Click Move N to trash. Ronja removes them one at a time, dependents first, and records why, for example “not read in 104 d (horizon 90 d), no live dependents”, naming only what she checked, where colleagues can see it later.
  5. Read the result for each table; anything refused stays listed with the reason. The dialog can’t be cancelled or dismissed while it runs: closing it would hide the run, not stop it.

Three kinds of row can’t be ticked; hover the box to see which:

The box says Rows
Read recently — not a cleanup candidate Read recently, Broken ones included.
No read history available No read history: that isn’t evidence nothing uses it.
Not judged by the cleanup lens A metric, a draft copy of a table, a table that belongs to one conversation, or anything in the Trash.

Clicking a table, in the graph or the list, opens a read-only panel: Metadata (with Reads and Recent builds), Schema, How it’s calculated for a derived table, Lineage (the tables around it, by hops) and Used by (the workflows, apps and other things built on it that aren’t tables). Open opens the table’s own page in a new tab, where you edit, share, move or check it; the map changes nothing itself. In a conversation, a table’s side panel has a Map tab with the same neighbourhood (The Map).

The page’s address carries the lens, preset and horizon, so a link shows a colleague the same view. An unrecognised value opens the plain map.