How scoring works
Every answer is scored 0–100 on each rubric dimension against the fixed anchors below — the same instructions the AI coach is given, published here in full. A score is a practice signal that shows you what to fix and lets you watch yourself improve; it is not a verdict on you.
How we score spoken answers
A spoken answer is transcribed and scored on five delivery dimensions — the overall spoken score is their average. The substance of what you said is also scored, on four of the written dimensions, so a confident delivery of a vague answer can't hide. Lead-in hesitation, a pause or false start before your answer begins, and background noise are ignored — scoring starts where your substantive answer starts.
Pace
Steady and unhurried, with pauses that aid emphasis — not rushed or draggy.
- Full marks (100)
- Comfortable, varied pace; pauses used for emphasis; never rushed or dragging.
- Low
- Words tumbling out, or so slow it drags; no useful pausing.
Filler-free
Few 'um', 'uh', 'like', 'sort of', or false starts breaking the flow.
- Full marks (100)
- Clean delivery; pauses instead of fillers; no false starts breaking the flow.
- Low
- Frequent fillers or false starts that interrupt the message.
Clarity
Easy to follow spoken — leads with the point, well-organized, not rambling.
- Full marks (100)
- The spoken point lands early and the answer is easy to track from start to finish.
- Low
- Rambles or circles; the listener struggles to find the point.
Enunciation
Words clearly articulated and easy to understand (general intelligibility, not accent).
- Full marks (100)
- Words are crisp and fully formed; easy to understand throughout.
- Low
- Mumbling or swallowed words that genuinely impede understanding.
- Fairness guardrail
- Judge general intelligibility only. NEVER penalize an accent, dialect, or non-native pronunciation — only genuine articulation problems that impede understanding.
Tone
Confident and appropriately expressive — engaged and human, matched to what the message calls for.
- Full marks (100)
- Warm, self-assured, and naturally varied; the emotional register fits the message; sounds like someone worth listening to — neither flat nor forced.
- Low
- Flat/monotone, robotic, or anxious/apologetic; or a register that clashes with the message (breezy for serious news).
- Fairness guardrail
- Judge whether the tone SERVES the message, not whether it matches one 'correct' style. A calm, measured speaker delivering calm content should score high — do NOT equate 'confident' with loud, fast, or high-energy, and never penalize a quiet or culturally different style that fits the message.
Substance dimensions applied to the transcript: Structure, Precision, Audience, Impact — scored against the same written anchors above.
How we score written answers
Each dimension is scored independently against the anchors below; the overall score weighs the dimensions an exercise emphasizes.
Clarity
Could a busy executive grasp the point on one read, with no re-reading?
- Full marks (100)
- The main point is unmistakable on the first read — no ambiguity about what's meant or why it matters.
- Low
- The reader must re-read to find the point, or could reasonably walk away with the wrong one.
Concision
Is every word load-bearing? Filler, hedging, and throat-clearing removed?
- Full marks (100)
- Nothing could be cut without losing meaning. No hedges ("just", "I think"), filler, or throat-clearing.
- Low
- Padded with wind-up, repetition, or hedging; the same message would be clearly stronger at half the length.
Structure
Does it lead with the point, then support it in a logical order?
- Full marks (100)
- The conclusion or ask comes first; everything after supports it in the order the reader needs. Nothing important is buried.
- Low
- Buries the point (backstory-first or point-last); support is out of order or hard to follow.
Precision
Concrete and specific rather than abstract, vague, or jargon-laden?
- Full marks (100)
- Every claim is concrete — real numbers, names, dates, or examples where they'd help. No vague words ("a lot", "soon", "things") where a specific was available; no unexplained jargon.
- Low
- Leans on vague abstractions or jargon; the reader can't tell what actually happened or what's meant.
Audience
Written for this reader — translates jargon into their terms and leads with what they care about (their stake, the 'so what').
- Full marks (100)
- Opens with the reader's stake (their 'so what'); any jargon is translated into their terms; includes exactly what they need to act and nothing they don't.
- Low
- Written from the writer's point of view; assumes context the reader lacks, or buries what the reader actually cares about.
Impact
Executive presence — confident, decisive, and owns a clear recommendation.
- Full marks (100)
- Confident and decisive; owns a clear recommendation or ask the reader can act on immediately. No unnecessary hedging or permission-seeking.
- Low
- Tentative or wishy-washy; no clear recommendation, or hides behind "maybe we could possibly consider…".
How scores stay consistent
- Deterministic grading. The model runs on a fixed, pinned scoring configuration with fixed score-band anchors, so the same answer earns the same score.
- An independent judge pass. Your displayed score comes from a separate scoring call made independently of the response that writes your coaching — so the score is never a model grading its own rewrite.
- Weekly calibration. An automated run replays a fixed set of reference answers through the scoring system every week — and on every scoring change — and flags any drift from the baseline.
- A relevance gate. An answer with no relevance to the scenario scores 0 across the board, with an invitation to read the scenario again and retry — a weak but genuine attempt is always scored normally.
What we deliberately do not score
- Your accent. Accents, dialects, and non-native pronunciation are never penalized — enunciation is judged only on whether you can be understood.
- Phoneme-level pronunciation. Spoken feedback judges general intelligibility, not individual sounds.
- Personality or style. A calm, quiet delivery that fits the message is not marked down; only a delivery that undercuts the message is.
- Whether your specifics are real. Scenarios set up a situation, not a data set — inventing a realistic number, client, or date for your practice answer is part of the exercise, exactly as you would draw on real details at work. The coach scores how you deploy specifics, never their factual accuracy.