View Results#

The Results viewer pairs the master measurements parquet with each plate’s overlay tiles. You filter the measurements pane on the left, pick an image from the dropdown, and the OpenSeadragon canvas renders the colony map with detection overlays.

Hub viewer (empty state)#

Open the Viewer tab in the hub:

Viewer tab in empty state.

The hub viewer starts in empty state. Pick a CLI output directory in the sidebar and use the hand-off banner to bind it. The shared binding panel reports queued/discovery/publication progress and permits cancellation. Results and Analysis receive one coherent, read-only snapshot, so a failed, cancelled, or superseded bind keeps the previous output visible rather than mixing two runs. If the output-consistency report finds contradictory or incomplete terminal evidence, you may inspect it but mutation controls for QC, Error, curation, Analysis, rebuild, and publication stay disabled. Two other ways to get a populated viewer are:

  1. Standalone launch (recommended for now). Run phenotypic.gui.results_viewer directly with --output-root pointing at the CLI output:

    uv run python -m phenotypic.gui.results_viewer \
    --output-root gui_tutorial_dataset/results --port 8051

  2. Open deliverables/dashboard.html in a browser. The dashboard is a self-contained HTML report under the output directory’s deliverables/ subfolder; you don’t need a server to view it.

The remaining screenshots come from the standalone launcher pointed at the output the Run Locally page produced.

Hub viewer after an asynchronous bind has atomically published the Results and Analysis snapshot.

Loaded viewer#

Note

The screenshots below are the standalone results viewer (no top bar / sidebar) so the page header reads “Results Viewer” instead of “PhenoTypic GUI”. The body is identical to a bound hub snapshot.

Loaded viewer with filter pane and image selector.

The viewer is split into three regions:

Region

Contents

Header

The output’s pipeline name (here gui_tutorial) and an SSH-tunnel reminder. Use the cache-nuke command if you ever see stale tile thumbnails.

Filter pane (left)

A query builder over deliverables/measurements.parquet (the post-applied mirror; falls back to deliverables/master_measurements.parquet on legacy outputs). The header shows how many images currently match. + Add filter adds a clause (column / operator / value); the dropdown is populated from the parquet’s columns.

Plate / Colony tabs (right)

Plate shows the full-plate overlay with OpenSeadragon. Colony shows per-colony crops aligned to the grid layout.

To render a plate, pick a filename from the Select image… dropdown. The dropdown is populated from the master measurements; once you pick one, the OpenSeadragon canvas tiles the overlay and the > Details link reveals per-image stats. Clicking individual colonies in the canvas highlights the matching row in the measurements.

Measurement table view.

Scrolling further shows the full measurements table.

Memory note#

The viewer loads the master measurements parquet and per-image overlays into memory on first access. If the parquet is large (tens of MB+), expect the first navigation to take a moment. Subsequent navigation between plates reuses the in-memory state. When the viewer is mounted inside the hub, the hub chrome’s Release control (not visible in the standalone screenshots above) drops the in-memory state; the next access reloads from disk. Standalone viewers intentionally do not subscribe to hub refresh events.

Process RSS may not return to the OS after release — the CPython allocator pools freed pages rather than returning them immediately. “Release” is honest about freeing the Python object graph; it is not a promise about RSS.

Where to next#