Methodology

Last updated: April 23, 2026 · Written by Dan Cohen

This page documents exactly how every song in the Music I Want library gets rated. It is long on purpose. If you're going to trust a rating system with your nervous system, you should be able to read the rubric.

The five dimensions

Every song receives a rating on each of the following dimensions. Ratings are produced through a combination of automated acoustic analysis and human review. A "top-line" level — Safe, Moderate, or Intense — is then derived from the five scores using the aggregation rule described further down.

1. Dynamic range (1–10)

How much the song's loudness varies from its quietest moment to its loudest. The raw signal is crest factor and short-term LUFS variance across the full track. A song that sits at a steady volume (a lullaby, a drone piece, most ambient music) scores 1–3. A song that opens whispered and hits a wall-of-sound climax (many rock ballads, post-rock crescendos, classical symphonies) scores 8–10. A 4 is a song with normal pop dynamics. A 7 means "you will feel this in your chest if your headphones are at a reasonable volume."

Why it matters: dynamic range is the single most reliable predictor of startle response. A song at 9/10 that catches you at low volume will make you physically flinch when the chorus lands.

2. Sudden changes (none / mild / present)

Abrupt, unsignaled transitions: a drop, a key change without warning, a drum hit that re-enters after silence, a vocal that screams after being whispered. This is a separate axis from dynamic range because a song can have huge dynamic range without sudden changes (a classical piece that builds for three minutes) or modest dynamic range with many sudden changes (a clicky electronic track with snare hits on unexpected beats).

"None" means the song telegraphs every change — nothing will surprise you if you're half-listening. "Mild" means there are one or two moments a sensory-sensitive listener should know about, but they are musically resolved. "Present" means the song uses surprise as a feature and will reward you for paying close attention, which is the opposite of what most sensory-sensitive listeners want from background music.

3. Texture (sparse / layered / dense / harsh)

How many distinct sound sources are happening at once, and how they sit against each other. Sparse is one voice and one instrument, or a single synth line. Layered is a full band arrangement with clean separation. Dense is a wall of overlapping elements — shoegaze, some metal, busy orchestral sections. Harsh is any combination where the timbres clash: distortion against bright cymbals, close-mic'd breath against compressed drums, unresolved dissonances.

Many sensory-sensitive listeners can handle Dense textures if everything is tonally resolved, but find even Sparse textures difficult if there's a Harsh element in the mix — close-mic'd mouth sounds, for instance, are a common misophonia trigger regardless of how few other instruments are playing.

4. Predictability (high / medium / low)

How well the song telegraphs what it is going to do next. A blues-structured song with clear verse/chorus/verse is highly predictable. A jazz piece with extended improvisation is low predictability. A song that switches time signatures mid-way, or introduces a new instrument in the last thirty seconds, is low predictability even if each individual section is simple.

For focus work and sleep, high predictability is nearly always preferred. For active listening and discovery, lower predictability is often what makes a song rewarding.

5. Vocal style

The character of the vocal performance, rated on a spectrum from "clean / intelligible / emotionally neutral" to "raw / strained / high-affect." The reference axis is not quality — we love raw vocals — but what the vocal does to a listener who is already dysregulated. A whispered, close-mic'd vocal can be as difficult to tolerate as a screamed one if the listener has ASMR-adjacent misophonia. A heavily compressed, bright vocal in the 2–4 kHz range (common on modern pop) fatigues ears faster than a softer mix.

The top-line Safe / Moderate / Intense rating

The top-line level is derived from the five component scores by the following rule, which is deliberately conservative:

• Safe if dynamic range ≤ 5 AND sudden changes = none AND texture ∈ {sparse, layered} AND vocal style is clean or soft.
• Intense if dynamic range ≥ 8 OR sudden changes = present OR texture = harsh OR the vocal style is screamed / wailing.
• Moderate is everything else.

The rule is OR-heavy on the Intense side and AND-heavy on the Safe side. That's intentional: we would rather a song that could be Safe get rated Moderate than a song that could blindside you get rated Safe. False negatives on Intense would undermine the whole premise of the site.

Worked example — "Black" by Pearl Jam

Dynamic range: 8/10 (the outro crescendo is substantial). Sudden changes: mild (the shift at the outro is musically signaled but emotionally large). Texture: layered. Predictability: medium — the verse structure is predictable, the outro is not. Vocal style: raw, with the vocal strain increasing toward the end. Top-line: Intense — triggered by the dynamic range threshold alone.

Worked example — "Holocene" by Bon Iver

Dynamic range: 4/10. Sudden changes: none. Texture: layered but tonally resolved throughout. Predictability: high. Vocal style: soft, falsetto, consistent. Top-line: Safe.

Moods, traditions, and recommendations

On top of the five-dimension sensory profile, each song carries:

• Moods — tonal descriptors (melancholy, hopeful, tender, etc.), 1–4 per song.
• Traditions — genre/subgenre tags, 1–3 per song, drawn from a controlled vocabulary.
• Recommended for — use-case tags (sleep, focus, bedtime, meltdown recovery, etc.), assigned only where the sensory profile genuinely supports the use case.
• BPM — beats per minute, where determinable.

What we don't do

We do not rate songs by "vibe." We do not use Spotify's internal "energy" or "valence" scores as ground truth — they're useful inputs but they collapse the five dimensions we care about into two that can mislead. We do not rate songs by popularity — a universally beloved song can still be Intense, and an obscure song can still be perfectly Safe. We do not AI-generate sensory commentary for the song pages. The first batch of ~2,900 songs received AI-written "arc descriptions" in an experiment early in 2026; the output was templated and we pulled most of it. Human review, slower, wins.

Limits and honest disclosure

The ratings are guidance, not clinical verdict. Sensory sensitivity is personal — what Safe feels like to one listener with ADHD is different from Safe for a listener with PTSD or misophonia. If a song we rated Safe unsettles you, trust yourself. We try to err conservative on the Intense side precisely because a false-Safe is much costlier than a false-Intense.

Acoustic analysis is fallible. Live recordings, bootlegs, and remasters can have very different dynamic ranges from the canonical studio version. When two versions differ meaningfully, we try to rate the version most listeners will encounter on a major streaming service.

Feedback

If you think a rating is wrong, we want to know. Email dan@thehivemakes.com with the song, the rating, and what you experienced. Ratings get revised. This is a living methodology — it will change as we learn.

See also: About · Guides · Analyze a Song