/docs · SkillLab · Practical
Feedback that lands — patterns from real coaches
Generic feedback is worse than no feedback. Here are the patterns that real coaches use.
A vocal coach giving feedback to a student doesn’t say “Great job!” They say “Your second verse was 8 cents flat on the chorus melody and your air support dropped on the long phrase — let’s try that line again with a tongue exercise first.”
That feedback works because it’s specific, actionable, and bounded. Most learning apps don’t come close. Here are the patterns from real coaches that translate into software.
Pattern 1: Specify the dimension before the score
Bad:
Score: 72/100. Try again!
Good:
Pitch: 88 — solid. Timing: 65 — you rushed beats 2 and 4 of the second bar. Tone: 72 — slightly nasal in the upper range.
The first version tells the user a number. The second tells the user which axis is which and where the work is. That’s an order of magnitude more useful.
SkillLab’s Challenge UI does this with a per-facet breakdown: spectral centre, loudness, voicing, tonal character, clipping. Each facet has its own score and its own note.
Pattern 2: Compare to the target, not to last time
Bad:
You scored 5 points better than your last attempt!
Good:
Target centroid was 1.2 kHz; you landed at 1.5 kHz. Slightly brighter than the goal.
The “better than last time” framing optimises for the trend; the “target vs you” framing optimises for the skill. Trend feedback is fine as a side note but should never be the primary feedback. The user is trying to land the target, not beat themselves.
Pattern 3: Name the cause, not just the symptom
Bad:
Your recording was noisy.
Good:
Background noise was high (-32 dB) — you’re probably in a live room. Try recording near a soft surface, or move closer to the mic.
The first version diagnoses a problem the user already knew about. The second names the likely cause and offers a corrective action. This is the difference between “useless” and “useful” feedback.
Even when you can’t name the cause with certainty — you’re a piece of software, not a human in the room — naming the most likely cause with the word “probably” is far more useful than the bare symptom.
Pattern 4: One correction at a time
A coach with a session full of corrections gives the student one — the most important one — per take. The student practices that one. Once it’s solid, the next session brings up the next correction.
Software that surfaces five corrections at once is software the user will struggle to act on. SkillLab orders feedback by impact and surfaces the most consequential note prominently. The rest are accessible but de-emphasised.
This sounds restrictive. It is. Restrictive is the whole point.
Pattern 5: End with what worked
Bad:
3 things to fix:
- Pitch
- Timing
- Tone
Good:
What worked: clear attack on each note. Try next: tighten timing in the second bar.
Negative-only feedback creates a relationship where the user dreads opening the app. Coaches close every session with at least one specific positive observation. Software can do the same.
This isn’t about being soft. The positive note has to be specific too. “Good job!” doesn’t count. “Clear attack on each note” works because it points to a real thing the user can keep doing.
Pattern 6: Use the user’s words
Bad (assuming you don’t know what the user is working on):
Your reverb tail was too long.
Good (if the user described their goal as “warm jazz sound”):
The reverb tail was longer than a “warm jazz” sound usually has — try halving it for the next take.
This is where personalisation actually pays off. If the user has stated their target genre, mood, or reference track, feedback should reference that. Generic feedback assumes a generic user; specific feedback respects that the user is working on a specific thing.
Pattern 7: Defer the score
When a user finishes a challenge, the score is the most visually dominant thing on the screen by default. Try the opposite: present the feedback first, then reveal the score after the user has read it.
This sounds wrong but it works. The score is a summary; the feedback is the substance. By presenting them in that order, you guide the user toward processing the substance rather than fixating on the number.
In SkillLab’s current UI we show the score and the feedback at the same time but visually weight the feedback more heavily. The score is a small badge; the feedback fills the panel. This is a deliberate choice.
What none of this means
This isn’t a license to make every interaction in your app verbose. The patterns above apply to post-attempt feedback, where the user is in a learning frame and wants substance. Quick check-ins, navigation prompts, and confirmation dialogs should still be terse.
Match the form to the moment. The user finishing a 30-second challenge wants thoughtful feedback. The user clicking “Next” wants the next thing to load.