Daily operator workflow¶

Audience: operators running a sequence of experiments in one session — typically four to ten runs in a morning or afternoon. Scope: the cadence between runs. Loading, editing, applying, running, sealing, re-arming without restarting the app.

This page assumes you've already worked through your first real run once. It picks up at the end-of-day → next-morning hand-off.

Picking up where someone left off¶

Launch capa. The welcome screen's Recent list shows configs that have been opened recently:

Click the top entry to load yesterday's config. The Setup tab opens with the saved values; the connection strip reads:

◐ Config loaded — Apply & Connect to open hardware.

Hardware is not yet open — capa never auto-connects on launch. Click Apply && Connect and the strip cycles CONNECTING → CONNECTED. You're ready to run.

Editing between runs¶

The standard between-runs edit is the new sample id and any sample-specific notes:

1. Experiment → Operator & sample. Update the sample id.
2. Experiment → CAPA Profile. Update specimen mass (post-tare) and anything else that's per-sample.
3. The connection strip flips to ◐ Draft has N unsaved edit(s).
4. Click Apply && Connect. Strip cycles CONNECTING → CONNECTED.

That's the loop. Five clicks, fifteen seconds.

For setpoint changes you want during a run, don't use Apply — use the manual control dock. Apply is refused mid-run anyway: the strip reads:

🔒 Run in progress — config locked until the run completes.

What Apply & Connect actually does¶

Every Apply & Connect:

1. Validates the composed draft (Setup + Method buffers).
2. Closes the previous worker pool.
3. Builds a fresh worker pool from the new config.
4. Opens every device the new config declares — serial ports, network sockets, NI-DAQ tasks, camera handles.
5. Publishes pool_changed so the manual control dock rebuilds its cards.

This is one path, not two. There is no "hot" vs "cold" mode toggle. The practical speed difference is:

• Same hardware list, parameter-only edit: the close-then-reopen cycle takes a second or two because every port is closed and reopened. There's no shortcut today.
• Different hardware list (new device added, one removed): same close-then-reopen plus any cost from the new device's open path (NI-DAQ chassis enumeration is the slowest).

Future versions may avoid the pool rebuild for parameter-only changes; that's not in today's runtime.

What you don't need to redo¶

Within one session, after the first successful Apply:

• Hardware discovery. Once the device table matches the rig, you don't need to re-scan unless wiring changes.
• Calibration import. The calibration snapshot persists across Apply cycles. It re-snapshots on each Apply, but you don't have to re-pick the source set.
• Operator id. Stays in the field until you change it. Change it when you change.
• Procedure / profile choice. Sticky across Apply cycles. Change only when switching experiment types.

What does change on every run: the sample id, the bundle output directory (one per run), the elapsed timer.

Reading the recents list¶

~/.capa/recents.json holds up to 10 entries, most-recent first. Duplicate paths are deduped (re-opening the same config bumps it to the top instead of growing the list). Paths whose files have been deleted are filtered out at read time — the file isn't eagerly cleaned.

If the recents list shows a path that no longer exists on disk, it'll silently disappear the next time you open something — no action required.

The Open ▾ → Recent submenu in the Setup tab toolbar is the same list.

Reviewing during the day¶

The status bar's bundle: pill always points at the most-recently sealed bundle. To inspect it:

• Help → Open logs folder opens the runs root in your OS file browser.
• For a queryable view across the day's runs, use the capa catalog CLI — it reads the runs.sqlite index that every run writes into.

There's no in-app sealed-bundle browser. See reviewing a run for the external tools.

Wrapping up at end of day¶

When you're done:

1. Make sure the last run sealed (Sealed badge, (completed) or (aborted) on the run-id line).
2. Close capa (window close, or Ctrl+Q). The ShutdownCoordinator cleanly closes the worker pool — serial ports are released, NI-DAQ tasks ended, camera handles closed. No data is lost; the last run was already sealed before you got here.
3. Power down or leave running per your lab's policy. capa doesn't care either way; the next launch picks up from the recents list.

If you killed the process (Task Manager / kill -9) instead of closing cleanly, see crash recovery — the next launch's finalize-on-startup pass picks up any bundle stuck in finalizing state.

See also: Your first real run, The Setup tab, Manual controls for between-runs device commands, Reviewing a run.