Charts

The Charts panel turns the loaded models into a dashboard of charts that stay bound to the 3D view: click a bar or a slice and its elements are selected in the model; pick an element in the model and the chart lights up the bucket it belongs to; the selection in one chart slices every other chart on the dashboard. It lives in the bottom strip next to Lists and Schedule (Analyze → Charts, or Charts in the command palette).

What a chart is

A chart is a dataset plus a spec. The dataset has one row per thing being counted — an element, a clash pair, a topic — and every row carries the ids of the elements it stands for. The spec says how to bucket the rows:

Field	Meaning
Chart	bar, stackedBar, pie, treemap, histogram (bins a number column), timeline (buckets a date column into ISO weeks), or elementCount (one row total)
Group by	The column whose distinct values become the buckets
Stack by	For a stacked bar: a second category column, one series per value
Measure	count of rows, or the sum of an independently chosen number column; Element Count always counts
Top N	Keep the N largest buckets and fold the rest into one grey Other bucket (which still carries every element it stands for)
Order	Largest first, or by label (dates and histogram bins are always chronological / ascending)

Because a bucket keeps its element ids, the bidirectional link needs no support from the chart library: a click is bucket → ids, a 3D pick is ids → buckets. The headless half is the @ifc-lite/charts package, which the CLI and the PDF export share. Its public Bucket.isOther?: true flag identifies the synthetic top-N tail; consumers must use that flag instead of treating a literal category key such as __other__ as synthetic.

Sources and scope

Source	One row per	Columns	A bucket selects
Elements	geometry-bearing element instance across every loaded model	IFC type, Storey, Model, Name, plus exact IFC fields chosen for grouping and measuring	the elements
Clash results	clash of the current run (after exclusions), or of the saved clash report the chart is bound to	Rule, Severity, Detection (hard / clearance / touch), Review status, Type A, Type B, Type pair, Model A, Model B, Storey, Distance (m, for a penetration histogram), Group	both elements of every clash in it; rows of a saved report do not select the current model
BCF topics	topic of the loaded project	Status, Type, Priority, Assigned to, Stage, Labels, Author, Due (overdue / this week / later / none), Age (days), and the dates Created, Modified, Due date, Closed	the components of the topics' viewpoints that are loaded
Schedule tasks	task of the active schedule	Task, Status, Phase at cursor (not started / in progress / done, following the 4D playback), Task type, Critical, Milestone, Duration (days), Products, and the dates Start, Finish	the tasks' products
IDS results	(specification, entity) result of the last validation	Specification, Result (pass / fail), Entity type, Failing facet, Model	the entities
Model compare	recorded change row of a selected saved comparison	State, What changed, IFC type, Revision	recorded rows do not select the current model
Existing unbound Model compare charts	diff entry of the latest live comparison	State, What changed, IFC type, Revision	the head-side entity (base-side for a deletion)

Date columns feed the timeline chart, which buckets per ISO week — topics created or closed per week, tasks starting per week. BCF dates are the only time axis: saved clash reports let separate charts show separate runs side by side, but there is no chart of clash counts over successive runs.

The dashboard's scope applies to the elements source and decides which elements its rows cover: all loaded models, only what is visible right now (the same answer the Lists panel's "visible only" gives), or the basket.

Saved comparison charts

Complete a model comparison and choose Save comparison in Compare models. In the chart editor, choose Model compare → Saved comparison and select the completed report. Each chart retains its own choice when another pair is compared or another saved result is selected elsewhere. Newly authored comparison charts require a saved choice; existing charts without one keep their previous live-comparison behavior.

The dataset contains the saved report's exact change rows, including content matches and excluding unchanged rows. It uses the recorded state, change, IFC type and revision names. Dashboard cards, document preview and PDF export resolve the same saved report. Recorded charts ignore live cross-chart slices and do not select, frame, color or capture entities in the current 3D scene.

The optional ChartSpec.comparisonId stores the saved report ID. Dashboard and document files retain that ID, but do not embed the saved history. An imported file needs the matching report in this browser; a deleted or absent report shows an explicit unavailable-source message in the card, preview and PDF. Choose another saved comparison in the chart editor to replace it. The latest live run is never substituted. Saved-comparison JSON can be downloaded for evidence, but the viewer currently has no saved-history import control.

Saved clash report charts

A clash chart reads the current result unless you bind it to a saved report. Running another check replaces the current result, so every unbound clash chart follows it. To keep a run, open Saved clash reports in the Clash panel header, name the result and choose Save current result. In the chart editor, choose Clash results, then pick the report under Clash report. That chart now shows the saved run whatever is checked afterwards, and two charts bound to two reports show two runs side by side. Choose Current result in the same list to return a chart to the live run.

A saved report is a frozen copy of the run: its clashes with their rule, severity, detection status and distance, the review status, comment, group and storey each clash had when it was saved, the rules and settings of the run, and the models it ran on with their names and source identities. It holds no ids of the loaded model, so a chart of a saved report never selects, frames, colors or captures elements in 3D, ignores cross-chart slices, and cannot take an element filter (the Clash rule filter still applies and lists the report's own rules). Reviewing or regrouping clashes later does not change a saved report.

The card names its source and anything that limits it as evidence:

Label	Meaning
Saved clash report: name, saved date	The chart shows this saved report, not the current result.
Clash rules matched no elements	Every rule with recorded coverage matched zero elements on at least one side. Zero clashes do not establish a clean check.
Some clash rules matched no elements	Some rules compared no elements; the count only describes the rules that had matches.
Partial run	The run stopped at its candidate-pair limit, so some pairs were never checked. Absent clashes are not evidence that they are gone.
Models changed before saving	The model was edited or moved after the run finished and before it was saved.
N hidden by exclusions	Your enabled exclusion rules were hiding N clashes when the report was saved. The report holds the clashes the panel showed, so the chart counts a subset of the run.
Different model revision	A loaded model has the same name as a recorded one but different content. The chart still shows the recorded results.
Models not loaded	The models the report ran on are not open.
Revision not confirmed	The viewer cannot tell whether the loaded models are the recorded revision: no source identity was recorded, the recorded and the loaded model have no kind of source identity in common to compare, or the model was edited in the viewer.

The limiting labels come first on the card, before the report's name, so a narrow card cannot cut them off; hover the subtitle for the whole line. The chart editor shows the same line under Clash report, and a document prints it under the chart in the preview and the PDF. Saved results are never matched to a different revision of the model: to see what changed, run the check again and use Compare clash runs.

The optional ChartSpec.clashReportId stores the report's ID. Dashboard and document files keep that ID but do not embed the report. Reports live in this browser's saved content library, survive a reload, and travel in Storage and backup → Download library backup; importing a backup restores them, and a document imported with them keeps each chart on its own report. If a chart's report was deleted, or the file was opened in a browser that does not have it, the card, the preview and the PDF say Saved clash report unavailable and show no data. The current result is never shown in its place: choose Choose a source on the card and pick the current result or another report.

Source filter

Every chart also has its own Source filter: the same selector syntax the search Filter tab's Selector field reads (IfcWall, Pset_WallCommon.FireRating=/REI.*/), typed once per chart and run on top of the dashboard scope. It is read and evaluated through the exact same path the Filter tab uses — there is no second matcher — so a chart's filter and a search filter never disagree about what an IFC class or property means.

Choose Add rule in the chart editor to build the source filter with the same rule controls as the search Filter tab. This includes Model Tag and other rules that have no selector spelling. An existing selector is converted to editable rules when every term can be read. Switching back to Selector fills the text field only when the rules have a faithful selector representation; otherwise it starts with an empty field. Saving a rule filter stores its rule groups in filter.groups and leaves filter.selector empty. Charts saved before this option continue to use their selector text.

The filter is applicable to Elements, Clash results, Schedule tasks and IDS results, and keeps a row if ANY of its elements matches (a clash keeps a row if either of its two elements matches; a schedule task keeps a row if any of its products matches, so a task with no matching product drops). It is not applicable to BCF topics or Model compare — a BCF row's "elements" are a viewpoint's component GUIDs (often none loaded) and a compare row straddles two revisions, so a selector has nothing well-defined to narrow; the editor disables the field for those two sources with a note, and a saved dashboard is refused if one carries a filter anyway.

Reading a chart's filter text uses the same all-or-nothing rule the rest of the selector adapter does (#4091's fix, applied here): a parse error, a selector with no filterable rule, or any construct this version cannot read yet is refused outright rather than run on the readable part — a chart that silently narrowed to what it understood would misreport its numbers with nothing on screen to say so. A chart with a filter still resolving, or one whose filter was refused, shows an empty chart with the reason in its subtitle instead of the unfiltered numbers; a successfully filtered card's subtitle names the filter (6 buckets · 12 elements · filter: IfcWall). A document's chart blocks resolve the same filters the dashboard panel does, so a printed report shows the same filtered numbers the panel shows.

IFC fields: attributes, properties, quantities and relations

For an Elements chart, Element field can stay on the built-in columns or select one exact IFC field:

Family	Picks	Value
IFC attribute	the attribute's EXPRESS name (ObjectType, OverallHeight, …)	the occurrence's own attribute; only attributes the schema declares as scalar values are offered
IFC property	exact property-set and property names (Pset_WallCommon.FireRating)	the occurrence's property, else its defining type's
IFC quantity	exact IfcElementQuantity and quantity names (Qto_WallBaseQuantities.NetVolume)	a number in the project unit, summable and histogrammable; the occurrence's quantity, else its defining type's
Material / classification / type / spatial	Material (every associated material name, joined), Type name (the defining IfcTypeObject via IfcRelDefinesByType), Classification for any system or one discovered system (the reference's identification, else its name), and the Container, Building, Site or Project the element sits in	categories

A Filter box narrows set and field names, so a model with hundreds of property sets stays pickable. Filter matching ignores case, but the selected names are the exact IFC names, including dots and slashes. For example, selecting Pset_WallCommon.FireRating makes its distinct values available to Group by; a consistently numeric property or any quantity can also drive Sum or a histogram. Measure independently offers discovered numeric IFC fields, so a chart can group by Material and sum Qto_WallBaseQuantities.NetVolume, or group by storey and sum area. Bar, stacked bar, pie, treemap, histogram and timeline charts use the same measure choice; Element Count remains a row count. A missing quantity contributes no volume and is reported in the card subtitle. Material grouping uses the element's material association and its element-level quantity: a multi-material element is grouped under the combined material label, not split into per-material volumes.

Occurrence properties take precedence. If the occurrence does not carry the selected property, its first defining type is consulted; an explicit empty/null occurrence value or a deleted property stays missing and suppresses inheritance. Attribute values are read only from the occurrence, and only attributes the loaded schema declares as scalar values are offered — a reference attribute (an IfcDirection, a placement) or a collection is not a value, whatever its STEP slot holds. 0, false, and identifiers such as "001" remain real values.

A property's shape decides what it can be. An IfcPropertySingleValue is a scalar and, when typed as a measure, a number. An enumerated, list, bounded or table value is a shape, not a scalar: its display string (A, B, 5 [1 – 10]) is a category it can be grouped by, and it is never summed — a one-member list is still a list and an upper bound alone is still a range.

Field kind and unit interpretation are saved in the dashboard document (version 2: adds the per-chart source filter above and drops a list scope kind a version-1 dashboard could carry; an older file is migrated on load), so temporarily unloading a model cannot reinterpret the chart. A saved field that is unavailable in the currently loaded federation remains selected and is labelled unavailable instead of being silently replaced. Property edits and deletions invalidate the data even when element ids and row counts do not change. The chart editor's own draft binds to a column of its own and never rebuilds the dashboard's datasets, so browsing fields cannot disturb another chart's live selection. Discovery merges what every loaded model exposes before deciding a field's kind, so a property that is numeric in one model and text in another is a category whichever model loaded first.

A numeric column is summed in one unit. Each value is converted into it from the unit it is actually in — the property's own explicit Unit when it declares one (converted by that unit's parsed scale, so a decimetre or a derived unit converts even without a curated display alternative), else the model's project unit assignment. A value whose unit cannot be established is reported as unsupported rather than guessed: a typed measure in a model that declares no unit for it, an explicit unit reference the file does not let the parser read, or a monetary amount in a currency other than the column's (there is no exchange rate to fold euros into dollars; the column takes the first model's currency). A sum subtitle reports rows without a measure and unsupported rows separately, so an all-missing field cannot look like a measured zero. Categories are unit-qualified instead (1 mm and 1 m are different labels).

Chart ↔ 3D

• On click sets what a bucket click does to the model: Ghost others (the default — the bucket's elements stay solid, everything else turns translucent), Isolate (hide everything else) or Highlight (selection outline only). The panel claims the isolate/ghost channel it writes and releases only what it installed, so an isolation another feature set up survives a chart click, and vice versa.
• A pick in the viewport highlights the matching bucket; a bucket that is only partly selected is emphasised, not marked selected. A foreign pick drops the dashboard slice.
• Colour in 3D pushes the first chart's bucket colours onto the model as an overlay layer (priority 75 — above the lens, below a running 4D playback). The chart that produced the selection keeps its full aggregation while the other cards cross-filter. Its selected elements retain the exact clicked bucket colour across overlap, reordering, and named ↔ Other folding; in Ghost others mode, unselected context keeps its authored model colours. Colours are assigned per label and kept when a bucket changes rank, so the legend stays a key.
• The frame button on a card frames the selected buckets (or the whole chart) in the camera.

Coordination report (PDF)

Report in the panel header prints the active dashboard: choose A4 or A3, portrait or landscape, fill the title-block fields (seeded from the 2D sheet's title block when one exists — project, drawn by, revision — plus report title and date), and optionally include a 3D snapshot of each chart's largest bucket, ghosting everything else. Every chart is drawn as a vector (ECharts → SVG → PDF), followed by its bucket table (label, element count, measure); a table longer than the page space is cut with an "… n more" row. Pages break between chart blocks and carry a running header, a footer with the generation time, and page numbers. A snapshot that cannot be captured (no WebGPU renderer, a lost canvas) is reported in its place; the rest of the report still renders. The page setup and fields are remembered on the dashboard, so the next export is one click — a dashboard exported to a file with a page setup is a report template.

Dashboards

Cards sit on a 12-column grid: drag a card by its title bar, resize it from the corner; positions are part of the dashboard and saved with it. The ⋯ menu next to the dashboard picker renames, duplicates, deletes, exports the dashboard as an .ifclite-dashboard.json file, or imports one — an imported dashboard gets fresh ids so it never collides with the copy it came from, and its layout comes along.

Dashboards persist in the browser like saved lists. The first open seeds Model overview (elements by type, by storey, and types per storey); the dashboard picker's New from preset group adds Coordination (clashes by type pair, severity, review status per storey, penetration depth; topics by status, open topics by assignee and due, topics created and closed per week), Delivery (IDS result per specification, failures by facet and entity type, elements treemap) and Schedule (tasks by phase at the cursor and by type, tasks starting per week, products per phase). Add, edit and remove charts from the panel header; every dashboard is a DashboardSpec JSON document validated on load, so a hand-edited or stale entry is dropped with a warning rather than crashing the panel.