Help

1. Triage. Open the Overview. The Needs attention queue lists every open flag, most urgent first.
2. Understand. Click a flag to see every signal that raised it, then choose Review & act for the evidence and an AI explanation.
3. Check the context. Replay the mission that raised it, look at the asset's other flags, or ask SPOTbot.
4. Act. Acknowledge or resolve the flag under your operator name. It leaves the queue, and its history is kept.

The data comes from the NextEra SPOT challenge kit: recorded missions, the actions SPOT took at each waypoint, the readings and images it captured, and the anomalies it reported.

Needs attention

• Each row is an open flag: its severity, a confidence score, what it is about (asset · waypoint · mission), and its strongest two signals. +N more signals means more rules fired; open the row to see them all.
• The coloured bar on the left of each row shows severity, so you can read the shape of the backlog at a glance.
• Use the All / Critical / High / Medium pills in the panel header to narrow the queue. The number on each pill is how many open flags it holds.
• Clicking a row opens it in place and lifts it to the top of the list. From there: Review & act, Replay mission, More on this asset, or Ask SPOTbot. Press Esc or click the row's title again to close it.

Fleet analytics

• Four charts behind the toggles in the panel header: success rate by inspection type, by battery band, action outcomes per mission, and failed actions by failure mode. Your last choice is remembered.
• With a flag open, the charts turn to it: its mission is marked, and its own numbers (in the accent colour) sit beside the fleet's.
• Click any bar to see the actions behind it. More on Analytics opens the full set of charts.

• Status line. Shows whether the flag is OPEN, ACKNOWLEDGED or RESOLVED, along with its severity and confidence.
• Explanation. Choose Load explanation. The model's draft streams in as it is written and is marked unverified. It is replaced by the checked answer once every claim has been matched to the flag's evidence. If the check fails, the draft is discarded rather than shown as fact.
• Actions. Enter your operator name, then Acknowledge (seen, being handled), Resolve (dealt with) or Reopen flag (put it back in the queue). Acknowledged and resolved flags leave the open queue; notes and action history are always kept.

A flag can be linked directly: the address bar carries ?flag=<id> while it is open. The AI review page shows the same queue without the charts.

• Choose a patrol to watch. Mission 1 runs as a mock patrol, so no robot is needed: it re-sends the kit's rows for each action, in order.
• The map shows SPOT moving through the building. A solid line is the route already walked, a dashed line is the route to the next stop.
• Each stop rolls in as a card under the map: the robot's log row, the photo, the triage decision (confidence model and rules), and SPOTbot's explanation when the stop was escalated.
• The battery chart draws as SPOT reaches each stop. The dashed line is the 25% battery cliff; dots are red when a stop needs attention and green when everything was fine.

• All missions answers "what ran". Click a card to replay the mission, or use Ask SPOTbot on it.
• Flagged missions answers "what needs a look": only missions with a failed action or an anomaly, worst first, each saying why it is listed.
• Replay: choose a mission, then play or scrub the timeline. Click a waypoint on the map (or focus it and press Enter) to see what happened there. A ! marks an error. Travel follows the hallways and stairs between stops, and time spent at each stop is compressed from the logs.

• Sections: Reliability (where actions succeed and what breaks them), Configuration, Missions, Failures & anomalies and Waypoints.
• Click any bar, slice, point or cell to open the rows behind it: the matching actions, missions, anomalies or readings, which you can drill into further.
• Shaded bands and whiskers are 95% intervals. A wide interval means few samples, so read the bar as a rough estimate.
• A bar in red marks the weak spot the chart is pointing at, such as the worst inspection type or the low-battery band.

• The gallery groups captures by inspection type, with each asset's finding and the image's source and licence. Asset details in the demo set are mock data, and each card says so.
• In the viewer, zoom and pan to any magnification. Hover a reading to light up its box on the image, and the other way round.
• The strip above the image lists every step that touched it, in order. Code computes every number; the AI models only describe and explain. A step that hasn't run yet is shown dimmed rather than hidden.

• Questions about the plant are answered from retrieved sources, which are cited in the answer. Greetings and general questions are answered without plant data.
• Every answer carries a badge saying how far to trust it:
  • Grounded · N sources: every cited id matches a source and every figure appears in the sources.
  • N figures not found in sources: numbers the check couldn't confirm. Verify them before acting.
  • No sources cited: treat the answer as unverified.
  • Cited an unknown source: the answer points at evidence that doesn't exist. Don't rely on it.
  • General answer: no plant data was needed or used.
• Hover a badge for the details. Ask SPOTbot buttons across the console open a chat with the question already written.
• Conversations are saved in the sidebar of the SPOTbot page. If a model hits its rate limit, SPOTbot offers the next one.

• Severity (Critical, High, Medium, Low) is how bad the finding would be if it were real.
• Confidence is how sure SPOTlight is that it is real: the confidence model's score, adjusted by the rules that fired. The queue's order comes from the backend's score, which weighs severity, priority and confidence.
• Signals are the rules that fired, heaviest first. The common ones:
  • SPOT reported an anomaly, or a reading failed or came back marked questionable.
  • Low battery: below 25%, far fewer actions succeed, so findings there are less reliable.
  • Repeat offender: the same asset has failed across several missions.
  • Hard environment: a high-noise or dusty area that fails more often than the rest of the plant.
  • Hidden failure: the mission reported COMPLETED even though an action inside it failed. The status field alone would never have surfaced it.

The search box in the top bar finds missions, assets and flags.

⌘K / Ctrl K

Jump to search from anywhere.

Tab or

Accept the suggested completion.

Move through the results.

Enter

Open the highlighted result, or see every match on the search page.

Esc

Close search, or close the open flag in the queue.

Settings holds your operator profile, appearance, notifications, triage and AI review preferences. The theme (light, dark or system) is saved and applies immediately. The other controls are there for the demo and are not saved yet; the page says so.

• Backend offline. Challenge-kit data (missions, replays, analytics) stays readable. Live flags, flag actions and explanations need the backend; the Overview offers Retry so you can reconnect once it is back.
• Explanation didn't finish. Its unverified draft is discarded. Choose Reload explanation to try again.
• SPOTbot unavailable. Either the assistant is switched off, its API key isn't configured, or its backend can't be reached. The chat names which one.
• Confidence model unavailable. The rules triage on their own, and the stop card says so.

Setup and service details are in docs/RUNNING.md in the repository.

Mission

One recorded run of SPOT through the plant, made of actions at waypoints.

Waypoint

A stop on the route where SPOT inspects one or more assets.

Action

One thing SPOT did at a waypoint: a visual, thermal, gauge or acoustic inspection, or navigation. Each succeeds or fails.

Asset

A piece of plant equipment SPOT inspects, such as an electrical panel or a heat exchanger.

Anomaly

Something SPOT itself reported as out of the ordinary.

Flag

SPOTlight's call that something needs an operator: an anomaly or failure, scored and explained.

Challenge kit

The NextEra SPOT dataset the console runs on.