06 — Review Panel¶

The Learning Review panel lets you inspect every cleaning run the integration has recorded, understand why a run was or was not used for learning, and manually exclude or restore individual jobs. This matters because the integration's time estimates are built from historical run data — if a bad run is included, it skews future estimates.

The panel has two subtabs:

Learning History — the full list of recorded runs, plus the stats, filters, and Profile Matcher (covered first, below).
External Jobs — runs started from your robot's own app rather than from this card. These need a quick review to confirm which rooms they cleaned before the integration can learn from them. When any are waiting, the subtab shows a count, e.g. External Jobs (2). See Reviewing app-started runs.

Opening the review panel¶

The Learning Review panel is a dedicated view inside the card. Navigate to it through the card's view selector, then pick the Learning History or External Jobs subtab at the top.

Overview stats¶

At the top of the panel, four counters give you a snapshot of what is currently loaded:

Stat	Meaning
Jobs	Number of jobs shown after applying the current filters.
Rooms	Number of distinct room-run pairs in the filtered set.
Profiles	Number of distinct cleaning profiles in the filtered set.
Updated	When the history snapshot was last fetched from the integration.

Filters¶

Below the stats, a row of chip-button filters lets you narrow the job list to exactly what you need. Selecting a filter chip immediately re-fetches the history from the integration with the new parameters applied — results update automatically.

Filter	Options
Room	One chip per room present in the history, plus "All Rooms".
Profile	One chip per cleaning profile present in the history, plus "All Profiles". Profile chips show a tooltip with the profile subtitle when one is available.
Status	All Statuses / Completed / Canceled / Failed / Interrupted.
Learning Use	All Learning Use / Used For Learning / Not Used For Learning.
Origin	All Origins / External / Dispatched. External runs were started from your robot's own app; Dispatched runs were started by this integration.
Sort	Newest / Highest Outlier / Suggested Exclude / Excluded Only.

Sorting is applied locally on the card — it does not trigger a new fetch.

Sort modes explained¶

Newest — most recent jobs at the top (default).
Highest Outlier — jobs sorted by outlier score, highest first. Useful for finding runs that deviated most from the average.
Suggested Exclude — shows only jobs the integration has flagged as candidates for exclusion, sorted by outlier score.
Excluded Only — shows only jobs you have already manually excluded, sorted by date.

Profile Matcher¶

The Profile Matcher lets you simulate a room configuration without editing any live room settings. You pick cleaning parameters using chip buttons, and the card tells you which saved learning profiles match exactly.

Fields¶

Field	Options
Cleaning Mode	Matches the available cleaning mode options for your vacuum.
Suction Level	Matches the available suction level options.
Water Level	Only shown when the selected Cleaning Mode involves mopping.
Cleaning Path	Matches the available cleaning intensity/path options.
Cleaning Passes	Fixed at 1 Pass or 2 Passes. (Unlike the room editor, the matcher does not expand to a vacuum's higher pass counts.)
Edge Mopping	On or Off. Only shown when the selected Cleaning Mode involves mopping.

The pass and mode choices follow your vacuum's own vocabulary, so the exact options vary by model.

Roborock (S6): no per-room cleaning mode

The S6 doesn't expose a per-room cleaning mode, so the Cleaning Mode field (and the mopping-only Water Level and Edge Mopping fields) may not appear for it.

How to use it¶

Set the fields to the combination you want to look up.
The Matched Profiles section updates immediately to show any profiles that are an exact match for those settings.
Click a matched profile chip to apply it as a filter — the job list below updates to show only runs that used that profile.
Click Reset Matcher to return all fields to their defaults.

If no profiles match, the panel says "No exact profile matches for the current settings." Adjust one or more fields until you find a match.

Runs list¶

The Runs section lists individual cleaning jobs as cards. Each card contains:

Job ID — the unique identifier for the run.
Detail line — start date and time, duration in minutes, outlier score (if available), battery used (if available), and water used in ml (if available and non-zero).

Badges¶

One or more coloured badges can appear on a job card:

Badge	Meaning
Excluded	This job has been manually excluded from learning.
External	The run was started from your robot's own app rather than from this card. This is its own neutral flag — it is not a sanity or learning verdict.
Suggested Exclude	The integration has flagged this job as a candidate for exclusion.
Non-"completed" status	The job did not finish normally (e.g. Canceled, Failed, Interrupted).
Sanity Failed	The job failed an internal data-quality check. External runs never show this badge — being externally captured is not a data-quality failure.
Recharge	A mid-job battery recharge was observed during this run.
Single Room	The job covered only one room.
Multi Room	The job covered more than one room.
Room Mismatch	On a dispatched run, the robot's live room signal disagreed with the room the run was assigned for part of the clean. The run is flagged for review — the assignment is kept, never silently overridden. Open the run to check which room was actually cleaned.

Key-value grid¶

Each card shows the following fields:

Field	Meaning
Rooms	The room slugs covered in this job.
Zones	The saved zone(s) cleaned during this run, shown when the job included a zone step. Lists the zone name(s); a multi-zone step also shows the count.
Area Cleaned	The floor area cleaned this run, in m². Shown when the run recorded a cleaned area — external (app-started) runs now carry this for both single- and multi-room cleans.
Scope	Whether the job was single-room or multi-room (formatted label).
Profile	The cleaning profile used. A subtitle line appears when the profile has extra detail.
Used For Learning	Yes or No — whether this job was included in the learning model.
Primary Room	The primary room associated with this job.

Notes¶

If the integration provides a reason text for the job (an exclude suggestion reason, a learning blocker explanation, a sanity flag, or a cancellation reason), it appears as a plain-text note below the key-value grid.

Excluding and restoring jobs¶

Excluding a job¶

If a job card has an Exclude button, you can remove that run from learning. Before clicking Exclude, pick a reason using the Exclude Reason chips that appear on the card:

Reason	When to use it
Short Test Cancel	You cancelled the run quickly as a test, not a real clean.
Manual Test Run	You ran the vacuum manually to test something, not as a normal clean.
False Completion	The vacuum reported completion but did not actually finish.
Bad Room Attribution	The run was attributed to the wrong room.
Interrupted Run	The run was cut short by an interruption outside your control.
Custom…	Reveals a free-text field so you can record your own reason instead of a preset.

The default reason is "Manual Test Run". Selecting Custom… opens a text box — type your reason there before excluding. Click Exclude to confirm. The button shows "Working..." while the action is in progress and is disabled to prevent double-clicks.

Restoring a job¶

If a job card has a Restore button, the job was previously excluded. Click Restore to include it in learning again. The button shows "Working..." while the action is in progress.

After excluding or restoring a job, the history snapshot is automatically re-fetched and the job list updates to reflect the change.

Reviewing app-started runs¶

When you start a clean from your robot's own app instead of from this card, the integration records the run but cannot be sure which rooms it covered or what settings each room used. These runs land on the External Jobs subtab and wait for a quick review. Once you confirm them, they feed the learning model just like card-started runs.

The pending list¶

Each waiting run appears as a card showing roughly when it ran, its total cleaned area, and how many segments were detected. Two buttons sit on the card:

Review — opens the two-step review wizard for that run.
Discard — drops the run entirely. Use this for a run you don't want to keep (a quick test, a mis-detection). Discarding cannot be undone.

If nothing is waiting, the subtab shows a short message telling you to start a clean from the app and check back.

Step 1 — How many rooms?¶

The robot reports its path as a series of segments. The integration's best guess at the room boundaries is shown, but it isn't always right — a single room can look like two, or two rooms can blur into one. Step 1 lets you correct the room count before naming anything.

You have two ways to fix the split, and you can mix them freely:

Room count stepper — at the top, a Rooms counter with − and + buttons. Press + to ask for one more room or − for one fewer; the integration re-segments the run on the spot and redraws the list. You can't go below one room, and you can't go past the number of boundaries the run actually contains.
Split here / Merge up — each detected room is listed in order with a one-line summary (area, time, mode, passes). The buttons say what they do:
↥ Merge up — folds this room into the one above it (you decided the boundary was spurious).
↳ Split here — appears inside a room when the integration spotted a possible boundary it didn't act on. Click it to break the room in two at that point. A boundary the integration is unsure about is labelled · uncertain so you know it's a softer guess.

Every change is applied by re-segmenting on the server, so the room summaries always reflect the current split. While a re-segment is in flight the controls are briefly disabled. If the run can't be split any finer, a short note explains that the count was capped.

When the rooms look right, click Next: name rooms →.

Older runs: runs recorded before this feature was added don't carry the detail needed to re-split them. For those, Step 1 falls back to a simpler merge-only view — you can still merge over-split segments together, just not re-split or set an exact count.

Step 2 — Name each room¶

Step 2 shows one panel per room (in cleaning order) so you can identify it and correct its settings.

Field	What it does
Which room?	Pick the room this segment belongs to. The integration's top suggestions appear as chips (with their learned area, when known). If the right room isn't shown, use the … pick another room dropdown to choose from every room on that map.
Mode	Vacuum, Vac & Mop, or Mop — when your vacuum exposes a per-room mode. Some vacuums (e.g. Roborock S6) don't, so this field may be absent.
Passes	Choose 1× or 2×. This field is always 1×/2×; it does not vary by vacuum.
Suction	Suction level, using your vacuum's available options.
Cleaning Path	Cleaning path / intensity, using your vacuum's available options.
Water	Water level. Only shown when the mode involves mopping.
Edge mop?	On or Off. This isn't detected from an app-started run, so set it yourself.

The mode, passes, and per-setting options come straight from your vacuum's own vocabulary — the same choices you'd see in the room editor — and whatever the integration captured from the run is pre-selected, so most rooms only need a confirmation. The dropdown of all rooms is pinned to a dark, readable style so the list stays legible.

If a picked room's area looks very different from what was actually cleaned, the wizard warns you at the bottom: "N rooms don't match the picked area — re-pick, or keep anyway." Either re-pick the room or, if you're confident, click Keep anyway to confirm regardless.

Click Confirm to save. The button shows "Saving…" while it works. You must pick a room for every panel first. Use ← Back to return to Step 1, or Cancel to close without saving. Once confirmed, the run leaves the pending list and the count on the subtab drops.