How Song Refiners Scores Your Song
A technical, transparent look at the pipeline — what we analyze, how the seven categories are evaluated, how the Overall Score is composed, and the principles that keep results consistent across every submission.
What Goes In
Every score starts with four inputs. Each one shapes how the system evaluates the song.
The complete track in MP3 form. The system listens end-to-end — intro, verses, choruses, bridge, outro.
Your lyrics in text form. They are scored independently and cross-referenced against what the audio actually delivers.
Song title, primary genre, and revision context. Genre is used to select the appropriate craft rubric.
Production stage (demo vs. finished) and review lens (Commercial Potential vs. Songwriting Craft). These shift what the rubric prioritizes.
The Pipeline
Six stages, in order. Each stage hands off to the next; nothing is skipped.
The submission is checked for cover-song risk, copyright concerns, and basic completeness before any analysis runs. Cover songs and known commercial recordings are blocked at this stage.
Your full audio file is analyzed — not a sample, not a snippet. The system listens to the entire track and forms observations about melody, performance, arrangement, and production.
Your lyrics are parsed independently and evaluated against craft criteria appropriate for the song's genre and form.
Audio observations and lyric observations are reconciled. When the audio and the submitted lyrics disagree, the audio is treated as the source of truth.
Each of the seven categories is scored independently against the appropriate craft rubric. A weak title does not pull down a strong lyric score, and vice versa.
The Overall Score is composed last, weighing how the elements work together as a single piece of music — not by averaging the other six.
The Seven Categories
Each category is scored independently on a 1–99 scale. Independence matters: a weak score in one category does not contaminate the others.
Hook value, memorability, searchability, and how well it primes the listener for what's coming.
Melody and vocal performance — phrasing, range usage, memorability of the hook line, and emotional delivery.
Imagery, specificity, structural integrity, prosody, and whether the words do real work or fill space.
Form, dynamics, transitions, instrumental choices, and whether the bridge (or its absence) earns its place.
Mix clarity, sonic choices, vocal placement, and overall polish — calibrated to the song's genre and intent.
Market readiness, playlist fit, and industry appeal — or artistic impact, depending on the lens you chose at submission.
How the Overall Score Is Composed
The single most misunderstood number on your review.
The Overall Score is not the average of the other six. It is a holistic judgment of the song as a single piece of music — how the title, top line, lyrics, arrangement, production, and commercial potential cohere into one listening experience.
A song can have six strong category scores and a lower Overall if the pieces don't fit together. It can also have one weaker category and a higher Overall if the whole genuinely exceeds the parts. The Overall is a synthesis, not a calculation.
Weighting is contextual, not fixed. Production carries more weight on a finished pop master than on a folk demo. Lyrics carry more weight on a singer-songwriter ballad than on an EDM instrumental. Top line and hook carry more weight on a country or pop submission than on a progressive rock piece. The system selects the appropriate weighting based on the song's genre and the lens you chose at submission — Commercial Potential or Songwriting Craft.
We don't publish a fixed weighting table because there isn't one. A fixed table would either flatten genre differences or invite gaming — neither serves songwriters.
Why Scores Get Harder to Move at the Top
The "moving target" feeling is real — and it's actually the opposite of what it looks like.
Comes from fixing structural issues — form, prosody, mix balance, basic craft. Big jumps from clear, addressable problems.
Comes from craft refinements — sharper imagery, stronger hook, tighter arrangement, cleaner production. Medium jumps from disciplined editing.
Comes from tightening one or two specific weak spots — not from rewriting everything. At this altitude, a full rewrite of music and melody won't move the Overall if the lowest category (often title or top line) stays the same. The score points to where the work is, not how much work to do.
Design Principles
The non-negotiables baked into the system.
Country is not judged against EDM. A songwriter-craft submission is not judged against a commercial-potential submission. The rubric adapts; the standards do not soften.
The system is calibrated against the craft standards used by professional A&R, publishing, and sync supervisors. A 90 today means the same thing a 90 meant a year ago.
We pin the same model, prompts, and rubric for every review so the same song lands in the same scoring neighborhood — not a different score every run.
Every revision is evaluated against the same standards as the original — no bonus for effort, no penalty for trying again. Categories that genuinely did not change carry forward.
Your score is not affected by other users' scores. There is no quota of 90s, no weekly distribution to hit. Your song is measured against the craft, not against the crowd.
Scores are issued once and locked. We do not retroactively adjust scores up or down after the fact.
How the System Was Designed
The short version of a long process.
The scoring rubric was built from three decades of professional songwriting, A&R, publishing, and sync experience — the same craft standards used inside the rooms where songs get cut, signed, or passed on. It was not built by reverse-engineering chart hits or by training on user submissions.
Each category has its own internal criteria — what makes a great title different from a great top line, what makes an arrangement earn its bridge, what separates a finished master from a polished demo. Those criteria were drafted by working professionals, stress-tested against hundreds of known reference songs across every major genre, and then calibrated until the system's scores aligned with what experienced ears were already hearing.
The audio and lyric analysis layers run on current-generation multimodal AI models. The models do the listening and the parsing; the rubric does the judging. The rubric is the part that matters, and it is the part we own.
We do not train any model on your songs. Submissions are processed for the purpose of producing your review and are governed by our privacy policy. Your work stays your work — see our explicit no-ownership guarantee in the FAQ.
What We Don't Do
Just as important as what we do.
We don't grade on a curve against other users.
We don't inflate scores to keep people happy.
We don't lower scores to push you toward a paid product.
We don't change scores after they've been issued.
We don't train models on your songs.
We don't publish internal rubric thresholds — that would invite gaming instead of better songwriting.
We don't claim any ownership, rights, or royalties on anything you submit.
What the Numbers Mean
Scores range from 1 to 99. There is no 100.
Industry-grade craft. Eligible for Showcase consideration and Hit Songs Radio airplay.