Sophisticated algorithms. Billions of data points. And yet they all share the same blind spot: they can't tell the difference between a song that's wrong forever and one that's just wrong right now.
Spotify
The Skip Problem
A skip is the most common signal — and the least informative.
The algorithm can't tell if you skipped because you hate it, heard it too much, or it's just wrong for this moment. So it guesses.
Same 30-50 songs on loop
Feedback loop trap: Users report Discover Weekly stuck on the same rotation. Some create new accounts to escape.
Context weighting: Spotify tries to weight skips by playlist type, but still can't decode intent.
Recent fix attempt: New "smarter shuffle" scores playlists for freshness — acknowledges repetition problem but doesn't solve the core issue.
Apple Music
The Slow Learner
Skipping barely matters. Only "Love" and "Suggest Less" count.
Takes weeks or months to adapt. Initial onboarding choices disproportionately shape your profile forever.
No official reset button
Stale mixes: Users report "Chill Mix" unchanged for months. Algorithm prioritizes stability over responsiveness.
Binary tools only: "Suggest Less" can't express "not right now" vs. "never again." One button, two meanings.
No pacing controls: If a mellow track kills your workout vibe, your only option is permanent suppression.
Pandora
The Homogeneity Trap
450+ musical attributes — but only thumbs up/down to steer them.
Stations become same-tempo walls. No ebbs and flows. Great at depth within a genome, terrible at variety.
One thumbs-up = ~100 similar songs
Music Genome Project: Musicologists analyze every song across 450+ attributes. Incredibly detailed — but binary steering.
Tempo flatline: Critics note stations deliver "the same tempo throughout" — can't say "too slow right now."
Can't express nuance: "Great song, wrong moment" and "I hate this" look identical to the system.
YouTube Music
The 12% Problem
Dislike stops only 12% of bad recommendations.
Mozilla study: negative feedback tools barely work. The more signals users give, the less the recommendations seem to change.
"Not Interested" = 11% effective
Mozilla research: 7 months, 20,000+ users. Negative feedback controls have "negligible" effect on recommendations.
Rage skipping: YouTube admits they can't tell "annoyed" from "searching for something specific."
Vanishing lists: Recommendation playlists refresh aggressively — tracks disappear before users can explore them.
Across every platform, the same five problems go unsolved
Ambiguous Skips
Hate it? Heard too much? Wrong energy? Accidental tap?
No "Not Now" vs "Never"
Context-dependent dislike and permanent rejection look identical.
No Pacing Controls
"Too slow right now" becomes "remove this forever."
Repetition Blindness
Finishing a song looks the same as loving it. The system can't tell the difference.
Context Collapse
Workout, commute, focus work — all blended into one taste profile.
The Problem
Thumbs Up Tells You Nothing
One button cannot tell the difference between a track that's wrong forever and one that's wrong right now. So the system guesses — and your queue slowly drifts away from what you actually want.
The problem isn't that you don't have an opinion. It's that the app can't understand which kind of opinion it is.
Mood vs. taste — disliking a song right now is not the same as disliking it forever.
Repetition vs. rejection — heard it too many times this week is not the same as never want it.
Energy vs. identity — wrong pace for this run is not about the artist at all.
Multiple things at once — wrong speed and not in the mood are two separate corrections, not one.
The Insight
Four Buttons. Four Different Meanings.
Not stronger or weaker versions of the same feeling. Each one is a fundamentally different kind of intent.
ExpandRemove
Love
"Build my world around this"
Expansion signal. The system commits — surfacing similar artists, vibes, and energy with higher confidence. This exact song plays more often.
Signal Strength
Like
"This works — keep it around"
Inclusion signal. Stays in rotation at moderate weight. Present without becoming dominant. No obsessive chasing.
Signal Strength
Dislike
"Fix this — don't overreact"
Adjustment signal. Something about right now is wrong. Corrects without permanently discarding. This can come back.
How Long It Lasts
Never
"Stop. This is a boundary."
Removal signal. Firm and lasting. Treated as a clear instruction, not a mood. Fewer chips — finality needs simplicity.
How Long It Lasts
Now Playing
Midnight Rain
Taylor Swift · Midnights
—
Select what applies
Select what fits · tap Done when ready
Tap any reaction above to try Attune
The Flow
Tap. Clarify. Done.
One reaction opens one panel. Every chip is visible. Select what applies, combine freely, tap Done. The algorithm learns. You never leave the music.
One panel, every timeAll chips immediately visible. No branches, no second screen, no guessing.
Chips stack into richer signalsWrong speed + not in the mood is one combined instruction. Attune reads both.
Back to music in secondsDone closes everything. The queue shifts quietly. You never lose the moment.
Positive Signals
Love expands. Like sustains.
Love commits the system — it leans hard into this signal and builds around it. Like keeps things present without obsessing. Both chips add specificity so the algorithm knows what to do more of.
Love→This artist+This vibe→Done ✓
Artist + vibe anchor. The system starts surfacing similar artists and contextually matched listening situations — not just more of this exact song.
Like→This energy→Done ✓
Energy reinforced. Intensity and pace noted at a lighter weight. Similar energy stays in the rotation without flooding it.
Love→This song+This energy→Done ✓
Specific song + energy. A taste-defining combination. The system plays this exact track more often in rotation, while also learning what surrounding energy should frame it.
Negative Signals
Dislike adjusts. Never removes.
Dislike has more options because there are more ways something can feel wrong. Never keeps it simple — when you're done, you're done.
Dislike→Wrong speed+Not in the mood→Done ✓
Pace + mood — both temporary. The pace recalibrates based on what you've been listening to. The artist is suppressed now, but returns when context shifts. Long-term taste is untouched.
Never→This artist→Done ✓
Hard artist boundary. Treated as a lasting instruction, not a mood. The artist is removed from rotation and stays removed.
Dislike→Not my vibe→Done ✓
Soft lane correction. Broader than a track problem, lighter than Never. The system quietly shifts the recommendation lane without drawing a hard line.
The Details
Reaction + Chip = Precise Instruction
Chips turn a generic tap into a specific signal. Toggle between states to see exactly how each combination shapes recommendations.
Love
"This needs to shape my world more."
PersistsYes — Love signals are long-lived and weigh heavily on future decisions.
ScopeExpands: opens new lanes. The system actively searches around this signal for related discoveries.
ReversalNot auto-reversed. Only overridden if you explicitly Dislike or Never something from that lane later.
Like
"Keep this in rotation, but don't overdo it."
PersistsYes — but with moderate decay. The system keeps it present without locking onto it.
ScopeMaintains: reinforces the existing lane without broadening or narrowing it.
ReversalFades naturally over time if not reinforced. Can be overridden by a later Dislike or Never.
Dislike
"Something is off. Fix it without overreacting."
PersistsContext-dependent. "Not in the mood" is temporary. "Not my vibe" persists somewhat longer.
ScopeAdjusts: corrects the current moment without damaging long-term taste signals.
ReversalDislike does not permanently close a door. The same artist or song can return later.
Never
"This is a boundary. Treat it as one."
PersistsYes — and strongly. Never signals are treated as lasting instructions unless manually changed.
ScopeRemoves: suppresses the specified artist, track, or energy type from this context.
ReversalStays in place until you decide otherwise.
What each chip tells the algorithm
This song
This exact track is what I mean
Increase this song’s recurrence in rotation directly
This artist
Give me more from this artist
Increase artist weighting in upcoming recommendations
This vibe
This feeling and listening context is right
Improve contextual matching for similar moods and sessions
This energy
This intensity level fits
Improve pace and flow matching around this intensity
What each chip tells the algorithm
This song
This specific track belongs in this rotation
Keep this exact song circulating at a moderate weight
This artist
This artist fits this rotation
Keeps this artist in the mix without flooding your queue with them
This vibe
This listening context is working
Reinforce contextual matching without broadening the lane
This energy
This pace and intensity fits right now
Gently improve flow matching without overcommitting
What each chip tells the algorithm
This artist
This artist is part of the problem right now
Reduce artist emphasis temporarily — not a permanent stop
This track
This specific track is the issue
Lower track priority without penalizing the artist broadly
Not in the mood
Not right now — maybe another session
Temporary suppression; this signal decays as context shifts
Wrong speed
The pacing is wrong for this run of songs
Looks at what you've been listening to and quietly recalibrates the pace
Not my vibe
This lane is not for me
Reduce similar lanes — longer-lasting than mood, softer than Never
What each chip tells the algorithm
This artist
I do not want more from this artist
Strong, lasting artist suppression — treated as a boundary
This track
I do not want this track in rotation
Hard track-level suppression — this specific song is out
This energy
I do not want this intensity in this context
Strongly reduce similar energy patterns — does not bleed into unrelated lanes
Why It Matters
Binary can't tell pacing from taste.
Can't separate mood from rejection.
Can't hear two reasons at once.
This system can — because every signal tells the system what you mean, not just what you tapped. The quality gap compounds with every tap.
Binary feedback
One dimension. The system guesses what you meant.
Attune signal
Multiple dimensions. The system knows exactly what you meant.
Every tap adds signal. The gap compounds.
The payoff
Fewer Skips
The queue stops surfacing what it knows is wrong for this moment.
Context Matches
Sessions shaped by mood, energy, and pace — not just a static profile.
Real Discovery
Love signals open new lanes. You start hearing things you didn't know you'd love.
No Drift
Good signals persist. Clear rejections stay clear. Your taste profile stays accurate over time.
Real Control
One tap says something precise. No menus. No maintenance.
Built by MMWB
This Is How We Think About Software.
You just watched a problem every streaming service accepts as unsolvable get taken apart and rebuilt around what users actually mean. That's not a concept. That's how MMWB works.
What you just saw
Intent captured, not just input recorded
Four signals that mean four different things
Chips that stack into richer instructions
One panel, zero branches, back to music in seconds
We find the version users didn't know was possible — then make it inevitable. Your product could work like this.
Start a conversation.
Tell us what you're building, where the experience is breaking down, or where the idea still feels fuzzy.
