Guide Track Diagnostic Review - Pass of the Shuttle

29 Jul 2026

Guide Track Diagnostic Review - Pass of the Shuttle

Date: 2026-07-29 Analysis source: Opus 5 guide track diagnostic Files reviewed:

Headline

Four of the five reported failures are the guide's fault, not Suno's. Suno reproduced this guide more faithfully than the failure list implies. It transmitted the dynamic arc from a 1.2 dB cue and inferred a bridge dropout from note length and harmony alone. Every melodic and balance failure is a property of the guide itself.

That is good news: all four are fixable in the guide, and none requires a different generator.

The Three Layers (As Measured)

| Layer | Notes | Range | Median level | Role | |-------|-------|-------|-------------|------| | vocal_marker | 105 | C3-D4 (14 st) | -5.86 dB | Melody | | room_hum | - | A2-E2 | -3.75 dB | Bass line (loudest layer) | | muted_pulse | 166 | C#6, B5, G5 | -32.43 dB | Monophonic clicks 26 dB below melody |

Critical correction: There is no muted-pulse instrument playing chord stabs. The pulse layer is monophonic single clicks 26.6 dB below the melody. It carries no harmonic information. Every chord in this guide is implied by the sustained bass line alone. The short mid-register events that looked like stabs are portamento glides (70% sit exactly one semitone below a melody note, preceding it by 23 ms).

The room_hum (ambience layer) is the loudest thing in the file. The melody sits 2.11 dB below it.

Ten Findings

1. Hook Contour - Guide's Fault

The apex is perfectly consistent - and that is the problem. All four chorus statements are the literal pitch sequence A3→G3→D3 followed by a held G3. Self-similarity r = +1.000 with 0.00 st spread. Perfect encoding of a fixed apex.

But the apex is A3 - which is also the verse's apex, and which sits 3 semitones below C4 (used 8× elsewhere) and 5 below D4. The guide gave Suno a chorus whose highest note is indistinguishable from the verse's highest note. There was no pitch signal marking the chorus as a peak.

Fix: Raise the chorus apex to D4. Keep the perfect consistency. Verse apex must stay at or below A3 so the chorus clears it by a fifth.

2. Melodic Range - Guide's Fault

The guide encodes the contraction directly. Verse spans 9, 11, 9 st. Chorus spans 7, 7, 7, 7 st. The chorus is 2-4 semitones narrower than the verse in every single instance. Whole-track range is C3-D4 = 14 st, and D4 occurs exactly once - inside a tag, not a chorus.

Fix: Widen the chorus from 7 st to ≥12 st (e.g. D3 up to D4). The chorus must be the widest melodic region.

3. Harmonic Rhythm - Guide's Fault (Inverted)

There IS differentiation - in the wrong direction. Verse: 4-bar Am pedal = 0.25 chords/bar. Chorus: F for 2 bars then G for 2 bars = 0.50 chords/bar. The chorus moves twice as fast as the verse. Backwards.

Worse, the chorus never touches the tonic. It is a plagal F→G that never resolves to Am. The most stable harmony is under the verse, and the least resolved is under the hook.

Fix: Invert. Verse: 1 chord per bar (0.50/bar). Chorus: one chord per 2 bars (0.25/bar). Put the 4-bar pedal on the chorus, not the verse. Land the chorus on Am.

4. Vocal vs Backing Balance - Guide's Fault

Melody is -2.11 dB against the bass. The chorus is the quietest melodic region every time (-7.9, -7.4, -10.1, -8.0 dB vs verses at -5.6, -4.9, -4.4). The ambience layer is the loudest thing in the file and the melody is buried under it.

Fix: Lift vocal_marker so melody sits +3 to +4 dB above room_hum. Drop room_hum by ~6 dB. Make the chorus the loudest melodic region, not the quietest.

5. Intro Length - Not the Guide

There is no intro. The first melody note is at t = 0.000 s. The vocal marker begins on the downbeat of bar 1. Intro length is zero seconds. The 24-second intro in the Suno output did not come from this guide.

6. Bridge Treatment - Partly Guide

Not sparse - density goes UP (0.48/beat vs 0.38 and 0.42 in surrounding sections). Rest is 12%, the lowest of any section. No dropout is encoded by density.

What the bridge does have: longest median note (1.68 s), narrowest span (4 st), and the guide's one moment of genuine harmonic tension (bass moves to E2, melody plays G#3 - the only raised 7th in the entire guide). A 1.89 s rest follows the phrase. So there is a caesura after it, but the dropout Suno transmitted was inferred from note length and harmony, not from density.

Fix: Cut bridge note density by ~40% (0.48 → ~0.29/beat). Raise rest from 12% to ~45%. Keep the E-major/G#3 tension. Put a deliberate caesura on "immortal."

7. Tag Escalation - Guide's Fault

Essentially none. Chorus apex: A3, A3, A3, A3 - identical, no escalation. Verse apex: A3, B3, A3 - non-monotonic. The only escalation is +1.2 dB across verses (-5.6 → -4.9 → -4.4), correct in direction but too small to read as an arc.

Fix: Encode melodic escalation across chorus statements. Borrow B's C4→D4→G4 move.

8. Chord Vocabulary - Guide Is Right

Five chords: Am, G, F, Dm, E. All diatonic to A natural minor except the single E major at the bridge, whose raised 7th is confirmed in the melody by that lone G#3. One purposeful secondary dominant at the tension point. This is excellent and a decisive improvement from the 22-chord sprawl.

Keep: Do not change the chord vocabulary.

9. Tempo and Humanisation - Watch This

Heavily humanised - arguably past the point of usefulness. Mean |deviation| from 16th grid is 52.52 ms. The expected value if onsets were random is 52.08 ms. Ratio of 1.008 - statistically indistinguishable from random. Only 9.5% of note durations are exact grid multiples. The onsets are definitively not quantised, but the jitter is so large (~50 ms) that it may be blurring the phrase grid Suno is trying to read.

Fix: Tighten to ±15-20 ms. Keep the human feel, restore the metrical skeleton.

10. Attribution Summary

| Reported outcome | What the guide encodes | Attribution | |---|---|---| | Dynamic arc transmitted | Weak: +1.2 dB across verses, -16.5 dB outro, 6.07 dB spread | Suno amplified a faint cue | | Bridge dropout transmitted | Not by density (which rises). By long notes, narrow span, E-major tension, 1.89 s rest | Suno inferred it | | Final-chorus arrival transmitted | 1-bar caesura before every chorus entry - consistently encoded | Guide got this right | | No strong hook apex | Apex fixed at A3 = verse's apex, 3 st below C4, 5 below D4 | GUIDE | | Chorus melody not wide | Chorus span 7 st vs verse 9-11 st, in all four statements | GUIDE | | No chorus harmonic differentiation | There is differentiation - inverted. Chorus 0.50 chords/bar vs verse 0.25 | GUIDE | | Vocal not above band | Melody -2.11 dB against bass; chorus is quietest melodic region | GUIDE |

Opus 5's Fix Guide v2 - Exact Changes

1. Apex: Raise chorus apex to D4. Keep perfect consistency. Verse apex stays at/below A3. 2. Chorus span: Widen from 7 st to ≥12 st (D3 up to D4). 3. Harmonic rhythm: Invert. Verse 0.50/bar, chorus 0.25/bar. Put 4-bar pedal on chorus. 4. Resolve the chorus: Land chorus on Am. Put hook's held note on root or fifth of tonic. 5. Balance: Lift vocal_marker +3 to +4 dB above room_hum. Drop room_hum ~6 dB. Chorus must be loudest melodic region. 6. Chord stabs: Either write actual triads into muted_pulse at usable level, or delete it. Currently contributes nothing. 7. Dynamic arc: Widen from 6.07 dB to 10-12 dB. Make final chorus the peak. 8. Bridge: Cut density ~40%. Raise rest to ~45%. Keep E-major/G#3. Deliberate caesura on "immortal." 9. Keep: 5-chord vocabulary, single E major at bridge, 1-bar caesura before chorus entries, ~25 ms portamento glides, perfect apex consistency (raise pitch, keep consistency). 10. Timing: Tighten from ±50 ms to ±15-20 ms. Keep human feel, restore metrical skeleton.

What This Means

The guide is better than the Suno analysis suggested - Suno was reading it generously and extracting more than it encoded. But the guide has four encoding errors that directly map to four of the five failures:

1. Hook apex too low (A3 instead of D4) 2. Chorus melody narrower than verse (7 st vs 9-11 st) 3. Harmonic rhythm inverted (chorus faster than verse) 4. Vocal buried under ambience (-2.11 dB)

All four are fixable in the guide JSON. None require a different generator or a different approach. The guide format works - it just needs to encode the right values.

Back to studies