Reliability: no lesson gets lost¶

ControlAI processes dozens of lessons a day automatically. And there is one firm promise here: no lesson silently disappears. If something goes wrong — the camera is offline, the audio failed to record, processing crashed — it is always visible and flagged, never lost in silence. A silently lost lesson is a lost report and broken trust; that is why the system is built to show the problem honestly rather than pretend everything is fine.

Every lesson is always at a clear stage¶

Any lesson in the system is always at one of a set of stages — so at every moment it is known where it is and what is happening with it:

Scheduled → Recorded → Processing → Done
                              │
                              ├─→ No-show   (nobody came)
                              └─→ Failed    (something did not succeed)

"Done", "No-show" and "Failed" are final states. A lesson cannot "hang in limbo": it is either carried through to a report, or explicitly flagged as a no-show or a problem.

What can go wrong — and what the system does¶

Camera offline¶

If the system could not pull the recording (camera off, no network, broken settings) — it sees this and creates a "camera offline" task, flagging which classroom is affected. The lesson is not lost — it stays in the "Scheduled" stage waiting for its recording, and the center's team sees that a hardware audit is needed. If the recording can never be pulled (for example, the NVR archive got overwritten), the lesson is marked "Failed" with a reason — it does not hang forever.

No audio or a bad microphone¶

If a camera has a weak or non-working microphone, the system flags the classroom as "microphone needed". This is a signal to install an external microphone — before report quality actually drops.

Lesson did not take place¶

If nobody came or the classroom is empty, ControlAI flags the lesson as "no-show": it does not spend resources on analysis and does not charge for it. At the same time the system distinguishes a genuine no-show (empty classroom — confirmed by video) from a "the microphone died, but the lesson did happen" situation — the latter counts as our defect and is also not billed.

Processing failure¶

If some analysis step did not succeed, the system retries (several times with pauses) — most temporary failures fix themselves. If it still does not work after the retries, the lesson is flagged as "failed" with a reason, and a task is created for the team. The lesson stays visible and can be re-run — it does not vanish.

Self-monitoring: retries and duplicate protection¶

Retries. Every processing step has automatic retries — if a failure is temporary, the system recovers without human involvement.
No duplicates. Even with retries, the same lesson is never processed twice and is never counted toward billing twice. This protects against both double charging and confusion in reports.

Everything is visible: the Operations section and coverage¶

The entire state of the system is gathered right in the panel — on logging in, the owner or manager immediately sees:

The "Operations" section — a list of all problems at once: camera offline, microphone needed, processing failure, awaiting consent, lesson no-show — each with a clear action.
Coverage summary — how many of the day's scheduled lessons were recorded and processed on time, and a list of "gaps" with a reason (for example, "11:00, classroom No. 5 — camera offline").
Connection health — when a recording was last successfully pulled, whether the center is reachable, whether the NVR archive is about to "expire".
A "Needs attention" feed on the dashboard surfaces the most important items.

Why this matters¶

For a center, ControlAI is valuable exactly to the extent that it can be trusted. If lessons sometimes disappeared silently, the owner would not know whether the picture was complete — and all the analytics would be devalued. That is why reliability here is not a "nice to have" but a foundation: better to show the problem clearly than to lose a lesson in silence.

Key takeaways from this chapter: every lesson is always at a clear stage and cannot silently disappear. Failures, offline cameras and no-show lessons are caught by the system, flagged with a reason, and shown in the "Operations" section and the coverage summary; retries fix temporary failures, and duplicate protection rules out double counting. Reliability is the foundation of trust in the analytics.

→ Next: 08. Inside the admin panel — a tour of the panel and a role-by-role breakdown: who sees and can do what.