Shipped: Load Existing Result Files into Attachments Dropdown

What We Built

One of the persistent frustrations with the old interface was that analysis results would vanish when you closed a plot. You’d run a calculation, examine the output, close the plot to work on something else, then reopen it later only to find all those results gone. The work was still sitting in the STAC catalog, but disconnected from the UI.

We’ve now fixed that by teaching the system to read result files from STAC item assets when a plot opens. Any results saved to a plot’s STAC item come back automatically in the attachments dropdown, exactly where you left them.

The implementation sits entirely in stacService—seven new methods that extract, identify, and deduplicate result files:

• parseViewerType() extracts the viewer type from multi-suffix filenames (file.2d.json becomes viewerType: ‘2d’)
• parseFileFormat() gets the actual file format from the extension
• assetToAssociatedFile() transforms STAC assets into AssociatedFile objects the UI understands
• isResultAsset() identifies which assets are results using roles: ["result"] as the primary signal, with fallbacks to debrief:toolId property or filename patterns for edge cases
• getResultFilesFromItem() extracts all results from a single STAC item
• loadResultFiles() loads the item and extracts results in one call
• setResultFiles() on activityPanelView handles integration with the UI

We wired this into openPlot.ts so results load automatically whenever you open a plot.

One complication: when you reload a plot, you might have both persisted results (from the STAC catalog) and runtime results (from tools you ran during this session). Deduplication logic prevents duplicates when merging them.

Lessons Learned

STAC asset metadata turned out to be more reliable than we expected. The roles field was explicitly designed for this kind of classification—marking assets as ["result"] gives a canonical signal that doesn’t require guessing. For edge cases (manually-placed files, legacy formats predating this convention), fallback patterns using debrief:toolId and filename conventions catch them cleanly without false positives.

The multi-suffix naming convention for viewer types (file.2d.json) felt fragile in theory. In practice it works well because the filename already contains necessary context, and parsing it keeps STAC metadata lean. It’s a pattern we can extend as new viewer types come online.

Twenty-four unit tests cover the full extraction pipeline—valid results, missing metadata, malformed filenames, edge cases in deduplication. They gave us confidence that the fallback logic handles messy cases without surprising behavior downstream.

What’s Next

This unblocks result archival. Once we wire up the UI for exporting or deleting results, everything downstream becomes straightforward. We’re also thinking through presentation logic—which results should load by default when reopening a plot, how to filter by tool or configuration, what happens when results come from different plot states.

→ See the spec → View the implementation