How the teacher rating is calculated¶
The teacher rating is the most sensitive number in the product. That is why the main principle here is transparency. The rating in ControlAI is not a subjective opinion and not a "black box": it is a clear formula built from measurable indicators that can be broken down to the last digit and explained in plain words.
Not to be confused with the lesson method-compliance score (1–10) from the "What it can do" chapter — that one is about compliance with the center's rubric. Here we are talking about the teacher rating — an overall assessment of their work (on a scale up to 5), built from the metrics of their lessons.
What the rating is made of¶
The rating is a weighted sum of four factors. Each factor is taken from lesson metrics (see the «Metrics» chapter) and has its own weight:
| Factor | Weight | What it reflects |
|---|---|---|
| Talk-time balance | 35% | how much the students talk during the lesson, rather than the teacher |
| Questions | 25% | how actively the teacher engages with questions |
| Target-language ratio | 20% | how much of the lesson is conducted in the target language |
| Punctuality | 20% | whether lessons start and end on time |
Why exactly this way: talk-time balance carries the most weight (35%), because it is the strongest sign of whether students are actually learning or just listening. The weights are configurable: a center can change them to fit its own methodology (the values above are the current example).
The rating can always be "unfolded"¶
This is exactly what sets it apart from a "black box": the rating does not arrive as one mysterious number. It can be unfolded to see what exactly it is made of. Each factor is shown with:
- its value (0–100),
- its contribution to the change in the rating,
- a plain-language explanation,
- a color — what is good, what is dragging it down.
Example breakdown (a teacher whose rating is falling):
| Factor | Value | Contribution | Explanation |
|---|---|---|---|
| Talk-time balance | 28 / 100 | ▼ −0.42 | The teacher started talking more (71% → 89%). The main reason. |
| Questions | 30 / 100 | ▼ −0.28 | On average dropped from 19 questions to 8. |
| Target-language ratio | 84 / 100 | — 0.0 | Stable, no problems. |
| Punctuality | 94 / 100 | — 0.0 | 94% of lessons on time. |
From a breakdown like this you can immediately see not only "the rating dropped", but also why — the teacher slipped into a monologue and stopped asking questions. And, most importantly, what to do: bring back the questions and let the students talk. This turns the assessment from a label into a hint for growth.
Skills radar¶
The same data can be viewed as a radar — a five-axis "star" of strengths and weaknesses. Each axis is computed from the metrics by a clear rule:
| Axis | How the axis is computed |
|---|---|
| Questions | full marks — around 28 questions per lesson |
| Talk-time balance | full marks — students talk around 55% of the time |
| Target-language ratio | the higher the target-language ratio, the higher the axis |
| Engagement | the less silence in the lesson, the higher the axis |
| Punctuality | the more lessons on time, the higher the axis |
Two clarifications to avoid confusion. The radar is slightly wider than the rating: a fifth axis — "Engagement" (silence) — is added to the four factors; it is visible on the radar but does not enter the rating formula. And "full marks" on the axes is the level of an outstanding lesson, not the norm: the "healthy" 15–20 questions from the Metrics chapter land solidly mid-scale on the "Questions" axis.
One glance at the radar and you can see where the teacher is strong and what is worth improving.
A trend, not a single snapshot¶
The rating is tracked by week (usually 8 weeks), so you see a trend rather than a random spike. Any change is explainable: the system shows which factor shifted the rating and by how much over the period. "The rating has been slipping three weeks in a row because of rising talk-time" is a concrete, verifiable conclusion, not a gut feeling.
Why this is fair¶
- The same for everyone. The same calculation for every teacher — no "favorites".
- From facts, not from opinion. All factors are measurable lesson metrics.
- Fully disclosed. The teacher sees what the assessment is made of and what to improve — nothing is hidden.
- Configurable for the center. The weights and benchmarks are tuned to the center's methodology, not imposed.
- Context-aware. The course type and the group's level affect which values count as good.
Rating and salary¶
The rating can (at the center's discretion) be tied to a bonus: the higher the rating, the larger the salary supplement. Because the rating is transparent, the bonus is transparent too — the teacher sees which specific improvements in their lessons will raise both the assessment and their income. More on salary calculation in the «What it can do» chapter and in the role instructions.
Important: the rating is for growth, not a label¶
- The purpose of the rating is to help the teacher get better, not to punish. It is tied to goals, coaching and hints, not just to "an assessment from above".
- There is no public "leaderboard". The rating is a working tool for growth, not a reason to set up a competition that demotivates part of the team.
- Any disputed number can be checked by opening the lesson transcript (and, within the first ~7 days, the recording itself) — the final word always belongs to a human.
Key takeaway from this chapter: the teacher rating is a transparent weighted sum of four measurable factors (talk-time balance, questions, language, punctuality) that can always be broken down to a concrete cause and a clear "what to improve". It is a tool for growth, not a label, and the final word stays with a human.
→ Next: 07. Reliability: not a single lesson is lost — what happens when a camera goes offline, a lesson did not record, or a failure occurred.