Results — trait emergence over time#

Once a run is finished, the colony overlays under deliverables/overlays/<dataset>/ capture how each plate looked at every timepoint. The Timeline tab of the results viewer lays those overlays out as a matrix — one row per grouping (e.g. plate number), one column per timepoint — so you can watch a trait emerge across a single plate’s time-course and compare plates side by side without leaving the page.

It reuses the same focus-and-navigate engine as the Browse Timeline, but draws its axes from the run’s post-applied measurement mirror (OutputRoot.master_df, which already carries the joined Metadata_* columns) and renders overlay thumbnails instead of raw source images. Pop-outs reuse the viewer’s existing /tiles deep-zoom route, so zoom and pan are GPU-smooth and need no extra wiring.

Step 1 - Open the Timeline tab and choose the axes#

Open a finished run in the results viewer (see View Results) and switch to the Timeline tab — the 6th tab in the strip. Two dropdowns drive the matrix:

  • Y (group) — the row axis. It offers every grouping column with no cardinality cap, so a high-cardinality grouping like Metadata_PlateNum (one row per plate) is selectable.

  • X (time) — the time axis. It offers any column whose name looks like a time field or whose dtype is numeric/temporal, with no cap. Prefer a monotonic column such as Metadata_ImageNumber; a time-of-day column mis-orders across days.

The Results Viewer Timeline tab: the Y (group) and X (time) dropdowns, the focus window with one focused overlay cell, the four edge navigation buttons, the position readout, and the tile-size stepper.

If the run has no eligible time column, the tab shows a guided empty state instead of a matrix, nudging you to re-run with a --metadata <csv> that carries a time field.

Note

The Timeline honors the active filter offcanvas — it renders the same master_df slice as the Plate and Colony tabs, so narrowing the filter narrows the matrix.

Step 2 - Navigate the matrix#

The matrix is not scrollable. A centered, no-scroll window renders around one focused cell (highlighted), and only the focused neighborhood plus a margin ring of just-off-screen cells mount their thumbnails — everything further out is offloaded, so the DOM stays light no matter how large the matrix is.

To move the focus:

  • Arrow keys — click the matrix viewport, then press ←/→ to walk one plate’s time-course and ↑/↓ to compare across plates. Focus clamps at the matrix edges (no wrap), and arrow keys are ignored while a dropdown holds focus.

  • Edge buttons (◀▶▲▼) — the four on-edge directional buttons move the focus the same way for pointer-only use.

  • Position readout — a row N/M · time N/M caption reports where the focused cell sits in the matrix.

The Results Timeline after stepping the focus a few columns to the right along one plate's overlay time-course, showing trait emergence across timepoints.

Each populated tile carries an N=k badge for the number of images that resolved to that (row, time) cell, and a tile-size − / + stepper scales the whole matrix between a configured min and max.

Step 3 - Deep-zoom a single plate#

To inspect one overlay at full resolution, open the deep-zoom pop-out:

  • Press Enter (or Space) on the focused cell, or

  • hover any visible tile and click the revealed ⤢ button.

Either path opens the same pop-out modal and mounts an OpenSeadragon viewer that reuses the viewer’s overlay DZI deep-zoom route. Close the modal to return to the matrix with your focus unchanged.

Note

Overlay thumbnails are cached on disk under the run’s .viewer_cache/timeline_thumbs, a sibling of the DZI pyramid cache, so the matrix is warm on tab re-entry regardless of the controller’s re-attach.

Where to next#