This skill governs the shape of a song over time, regardless of its notes.
Harmony, melody, and orchestration determine "what sounds at this moment"; arrangement architecture determines "how the listening experience changes from second 0 to second 220." A song can have every chord right and every instrument sounding good, yet still be so flat that people can't bear to listen—the problem lies in this layer.
This layer is the main battlefield for "not sounding like AI." The default values of generative models are almost entirely concentrated here: energy monotonically increasing, instrumentation unchanged from start to finish, sections fitting perfectly into multiples of 8, never subtracting, endings uniformly identical. These are not timbre problems—switching to a better model won't make them disappear—they arethe absence of structural decisions。
Which sections to read by task
| What you want to do | Read |
|---|---|
| Plan the structure of a song from scratch | §1 Must-fill before starting → §2 Energy curve → §4 Formal skeleton → §3 Subtraction |
| The song is "flat," lacks punch, can't bear to listen | §2 Energy curve, §3 Subtraction |
| Every section sounds the same | §5 Instrumentation as form, §6 The second pass must not be copied as-is |
| It sounds like AI when generated | §9 Diagnostic path (lookup by symptom) |
| Don't know when instruments should enter or exit | §5 Instrumentation as form |
| How long sections should be | §4 Formal skeleton |
| Can't remember this song | §7 Hook placement |
| Don't know how to end it | §8 Three kinds of endings |
| Fill in the arrangement part of ARR-SPEC | §10 Field correspondence ＋ §11 Must-pass after writing |
Boundaries: what is not covered by this skill
| Not covered here | Covered where |
|---|---|
| How specific instruments are orchestrated, how ranges are laid out | mc-orchestration |
| How frequency bands are divided, who competes with whom for frequencies | mc-texture-layering |
| How to make drums, bass, guitar, and keys play like real people | mc-rhythm-section |
| The harmonic progression within a section | mc-harmony |
| Historical forms of musical form (sonata form, etc.) | mc-form |
| How vocals are sung | mc-vocal-direction |
| EQ, compression, mastering | Mixing matters; this package only provides mix_intent this intent (mc-mix-intent） |
This skill only governsthe relationships between sectionsandevents on the timeline。
1. Must-fill before starting (four items; if one is missing, you may not draw the timeline)
These four items are not formalism. The reason things written by generative models sound like AI at first listen is not that their timbre is bad, but thatno one made a choice—and choices happen on these four questions.
| Required | Question | What failure looks like |
|---|---|---|
| ① Energy target | Which section has the highest point of the whole song? Which section has the lowest point? | "The chorus is the highest"—then what about the lowest point? Without a lowest point there is no highest point |
| ② Subtraction event | Which spot shouldbe removed? Remove what? Why remove there? | Not a single spot → see §3, this is one of the strongest AI signals |
| ③ Entry and exit arrangement | Which instrumententers midway? Which oneexits early？ | All present from beginning to end → see §5 |
| ④ Non-vocal memory point | Without relying on vocals, what can listeners remember? | Can't say → see §7 |
Self-check: If you can answer these four questions within 30 seconds, it means you haven't thought it through; if you can't answer them, it means you can't start yet.
2. Energy curve: dramatic arc
2.1 Model
DeSantis directly applies the narrative medium'sdramatic arc(dramatic arc, graphically called Freytag's pyramid) to music (p0276):
| Arc segment | In a story | In music |
|---|---|---|
| Exposition Exposition | Introduce characters and setting | Introduce the musical material to be used throughout the piece (melody, harmonic progression, rhythmic motif) (p0277) |
| Rising action Rising action | Conflict between protagonist and antagonist | Of the exposition materialvariation or sudden change（variations or mutations）(p0277) |
| Climax Climax | Turning point | Often marked bya sudden increase in texture density—"the drop" in EDM (p0277) |
| Falling action Falling action | Resolve unresolved conflict | Often mirrors the activity of rising action (p0277) |
| Dénouement Denouement | Characters return to normal | Restatement of established material (repeated chorus) or gradual dissipation (p0278) |
The scale is nestable: a single melody can be an arc (rising to a peak and then falling; the melodic contour visually is a Freytag pyramid); a one-hour DJ set is also an arc (p0278).
2.2 Rule: the energy curve must have a descent
★ Strong rule. This is the most strongly evidenced point in this skill.
Measured: 62 real multi-track arrangements,62/62 all have an energy descent(95% interval 94–100%). Not a single one is monotonically increasing.
And monotonic increase is precisely the most stubborn default of generative models.
Self-check action: Write the energy of each section as a sequence from 0–10. If this sequence is non-decreasing, what you've written is not a song, it's a slope.
Failure conditions：
- Forms such as dub and minimal techno that lack clear climax building—an exception DeSantis himself points out: the dramatic tension of this kind of music "is created in a completely different way throughsound designand attention to subtle changes over time in static textures" (p0278). In this case, a flat energy curve is the style itself, not a defect.But you have to be able to say that this is the path you're taking。
- Extremely short works (15-second ads, short game loops) have no room for a complete arc.
2.3 A statement that should not be used as a rule
"There should be an energy dip before the climax (leaving space before the chorus)"—thisis widely circulated, but our measured data is contradictory：
| Evidence | Result |
|---|---|
| Full measurement of 62 songs | Only 22% hit |
| 7 songs manually annotated ground truth | 5/7 = 71% hit |
The difference comes from the reliability of section boundary detection (the full-set version relies on automatic detection, while the manually annotated version is ground truth).
So this skill does not write it as a rule. It is atechnique worth trying, not a constraint that must be followed. You can use it, but don't force in an unnecessary empty beat just because "the book says to leave space."
2.4 ★ Why a climax section can be packed yet not loud: the five types of tutti and one general rule (Yang Liqing)
Tutti is divided into five types (Volume 2 p0361–0378):Unison type、Chordal dynamic type、Polyphonic type、Thematic type、Multi-factor composite type(there is also coloristic tutti). Their respective temperaments:
| Type | Temperament |
|---|---|
| Unison type | Very strong loudness when spanning a wide range,but lacking harmonic support, saturation and fullness will be insufficient（p0361） |
| Chordal dynamic type | Simple structure, concentrated sonic "focus"; when arranged properly, chord tones support each other and resonate well,the most powerful of all types（p0362） |
| Polyphonic type | Belongs to heterogeneous texture, full of inner tension and drive, mostly used for development, transition, and pushing toward the climax;pure polyphonic type is "constrained by the decomposition of force," and its sonority cannot rise, so if you want force you must follow it with a homophonic tutti to get the satisfaction of a climax (p0363–0364) |
| Thematic type | See the two counterexamples to the general rule below |
| Multi-factor composite type | Extremely wide range, many instruments in the most tense high register, with both high tension and brightness; harmony is given only to the brass and concentrated in the midrange, leaving ample space from both the bass line and the highest melody, so that resonance is not affected (p0372) |
★ General rule(p0365–0369): For a tutti aimed at displaying the orchestra's power,the clearer the textural layers, the more uniform the rhythm, and the simpler the harmonic structure, the better the sonic resonance and the stronger the effect. When the chorus is packed yet not loud, it is mostly because all three have been done in reverse: the layers are muddled together, each layer's rhythm goes its own way, and the chords are stacked in a complex way.
Three fixes (p0368–0372):
- For solidity, first strengthen the midrange: place the brass horns, trombones, and tuba in the rich midrange, in close position without omitted tones
- For brilliance, raise the register, or add a layer an octave above to reinforce overtones (divided violins repeat the melodic line in an extremely high register, "reinforcing overtones and improving resonance")
- For tension, put the main voices into a tense register, at the cost of reduced resonance
Cluster-like complex harmonic relationships not only fail to form mutual overtone support, but also affect purity due to overtone cross-interference; their sonority is inferior to a chordal tutti with a simple structure (p0363). If you want "big," don't also want "complex"; you can only choose one of the two.
3. Subtraction: Arranging is sculpting, not painting
3.1 Arranging as a Subtractive Process
The fear of facing a blank timeline is the same as the fear of facing a blank page (p0281). The conventional approach is to gradually fill it in from left to right, like painting on a blank canvas. DeSantis argues for the opposite (p0281):
Step one: Fill it up quickly
- Immediately fill every track of the entire timeline with material (p0282)
- The goal is not to make a good arrangement, but toturn "from nothing to something" into "from something to nothing" (p0282)
- ★ Hard time limit: no more than 20 seconds. Going over means you are making creative decisions, and you must resist that temptation (p0282)
- Use copy-paste, duplicate, dragging the right edge of clips (p0282)
- Do notthink about section names,no need touse all the material (p0282)
Step two: Sculptural subtraction
- Start from a whole solid block of raw material and carve into it, carving out space (p0283)
- Basis for judgment: "It is often easier to hear when something is bad than it is to imagine something good" (p0283)
- When a part goes on for too long, intuition tells you that change is needed (p0283)
Reverse variant: If the genre's characteristic is that texture density increases/decreases over the course of the song, you canwork backwards from the end to the beginning, removing more the further back you go (p0284).
Failure condition: The premise of this process is thatyou already have enough material on hand. When there is not enough material, "fill it up quickly" will fill it with a bunch of repetition, and what you carve out will also be repetition.
3.2 The power of erasure
"It seems that perfection is reached not when there is nothing left to add, but when there is nothing left to take away." (p0286)
The counterintuitive core judgment: When music feels "small," "weak," or "lacking power," the solution is oftenremovalrather than addition (p0286).
Principle: Musical space has boundaries, just like a canvas.Every time you add a part, all the other parts have to fight harder for the listener's attention (p0286). So the more you add, the weaker each thing becomes.
★ Precondition (DeSantis explicitly wrote this): The subtractive strategyis only effective when the parts you already have are as good as they can be(p0286). Delete a few from a pile of mediocre material, and what remains is still mediocre.
Clarification on "Is this minimalism?" (p0287): No.
- Music with a slow rate of change can sound minimal even with many elements
- High-density music (some commercial EDM), if itdeliberately limits the number of parts sounding at the same time, still sounds clear and open
- Core principle:add parts as necessary, but only as necessary, and ruthlessly remove anything that does not contribute
"Thinning out" — an intermediate means beyond outright removal (p0287)：
- Reduce thenumber of notes
- Shorten the sustained note'sdecay time, so it quickly gets out of the way
3.3 Rule: at least one obvious subtraction in the whole track
Measured: out of 62 tracks, 70% had obvious subtraction (energy span ≥30%).A tendency, not an iron law.
Self-check action: point out which bar your subtraction event is in, what you removed, and why you removed it there. Only if you can answer all three does it count.
Failure condition: the 30% counterexamples are mainly continuously high-density dance tracks and noise-oriented works. If that's the path you're on, explain why; don't force-remove a drum section just to comply.
4. Formal skeleton and section proportions
4.1 Steal the skeleton, not the material
Problem: you have all the material, but when you arrange it, it just doesn't satisfy;what's missing is a sense of internal proportion (p0290)。
DeSantis's method is to measure the sense of proportionout of an example (p0290–0292)：
- Find a song with an effective form that is in the same "musical universe" as your music
- Create a new empty project and place it at bar 1
- Carefully adjust the tempo until the DAW grid aligns perfectly with the song's actual beat—if you don't do this step well, everything after will be wrong
- Listen, and every time you hear a formal division (or an interesting moment within a section), stop and drop a marker
- Keep going until the end
- Give each marker adescriptive or functionalname ("Verse", "Bass starts"), so you'll understand it later without relistening
- Delete the original audio, keep only the markers — this is a known-effectiveformal skeleton
- Fill it in with your own material
The most valuable part of this method is step 7: what you take away is proportion, not content. A skeleton does not constitute plagiarism, and its source can't be heard.
DeSantis's own warning：
「once you actually start employing this in practice, you may find that your own musical materials don't naturally fit into the arrangement skeleton perfectly.」(p0292)
If your breakdown "wants" 4 more bars,absolutely trust your intuition (p0292). The skeleton is for getting started; once you're in flow, move things freely.
4.2 ★ A real disagreement: should sections be multiples of 8
The two sides directly conflict;both are right, and they must be considered together：
| Source | Statement |
|---|---|
| DeSantis (p0291) | In most electronic genres, formal divisions occur at multiples of 4 or 8 bars; listeners have been "primed" to expect divisions at these points |
This library's 20 self-checks #3 + AI-flavor Article 14 Item 1 (mc-ai-tell-audit §2） |
sections all being integer multiples of 8is one of the default values most easily detected in generative models |
Reconciliation：
- The primed expectation is real. It's correct for the vast majority of sections to land on multiples of 8; violating it makes people "lose count of the beat".
- The problem isn't "being a multiple of 8", it's "all of them being so". Human-written arrangements will break itat one spot: a 6-bar build phrase, a second verse 4 bars shorter than the first pass, a tag hanging 2 extra bars at the end of the chorus.
- So the rule is: default to multiples of 8, but break it at least once in the whole song, and be able to say why you broke it there.
Empirical reference: among 62 real arrangements, almost all have sections that are not multiples of 8— but this item depends on automatic section detection, and the labels
no_truth，numbers are unreliable, treat them only as corroborating evidence。
4.3 General definition of a section (p0293–0294)
| Section | Length | Function | Key points |
|---|---|---|---|
| Verse / A | Usually 16 or 32 bars | The main body of the song; when there are lyrics, it tells the "story" | Appears cyclically |
| Chorus / B | Equivalent to verse | Contrastwith the verse; usually containsa hook；the musical point of resolution(the verse creates tension) | Cyclic verses share the music but have different lyrics;cyclic choruses usually have both the same music and the same lyrics |
| Bridge / C | — | The contrast between verse and chorus | Usually appears only once in the whole song; often ina different key, uses an unusual chord progression, or has a completely different texture density and energy; in some genres it is used as an instrumental solo |
Order conventions: the first chorus appears after a verse (some songs also open with a chorus); the bridge usually comes after at least one verse + one chorus (p0294).
4.4 Where to break: conventions for section bar counts in Japanese pop (Fujitani)
§4.2 says "break the multiple of 8 at least once"; this table giveswhich positions are not 8 to begin with(Fujitani p0061–0062):
| Section | Conventional length |
|---|---|
| Intro | 4–8 bars; 32 bars is also possible, but without enough persuasiveness you get bored halfway through |
| 1A | Basically 8; 16 also occurs, in which case it is A–A'–B–chorus, with the latter 8 slightly different from the first 8 |
| 1B | Half of 1A, or the same length as 1A |
| 1 chorus | 8 or 16 |
| Interlude | The interlude after a large section tends to be short,at most 4 |
| 2A | ★ If 1A is 16, 2A takes 8; if 1A is 8, it stays unchanged |
| 2B, 2 chorus | The same length as 1B, 1 chorus |
| D melody (big chorus) or interlude | Around 8, usually made very full |
| Final chorus | The same length as the previous two times, ortwice as long; when doubled, the first half is quiet and the second half gets livelier the closer it gets to the chorus |
Three natural "break" points: 1B takes half of 1A, the interlude is at most 4, and 2A is halved. If you use one of them, the self-check in §4.2 passes, and there is no need to create an irregular section separately.
5. Layering as Form
This is the most underrated section of this skill, and also the section with the strongest empirical evidence.
The boundaries between sections are often not where the harmony changes, but wherethe instrumentation changes—an instrument is added, or one is removed. The entrances and exits of instruments are themselves the form.
5.1 Two strong rules
Measured(62 real multitracks, with true per-instrument stems, not estimates):
Rule Hits Range There must be an instrumententering midway 59/59 94–100% At least one instrumentexiting early 59/59 94–100% The opening should not lay out all the instruments at once (proportion present at the opening ≤60%) 54/59 82–96% All three arestrong rules. 59/59 means that in this corpusthere is not a single exception。
Compare with the default of generative models:the instrumentation stays unchanged from beginning to end. This is item 3 of the fourteen signs of AI flavor (mc-ai-tell-audit §2),
and the measurements tell us that real arrangersnever do this in a single song。
5.2 How to implement it: write an entry/exit event table
Every instrument must be able to answer three questions:
| Which frequency band does it occupy | If you cannot say, then you will not know when it is fighting someone else for frequencies (details go to mc-texture-layering） |
| When does it enter | Down to the section |
| When does it exit | ★ An instrument for which you cannot say "when it exits" is usually the one that makes the arrangement flat |
Self-check action: sort the roster by entry and read it from top to bottom— you should be able to read how the song's instrumentation piles up layer by layer. If you cannot read the layers, there are no layers.
A real sample(an alt-rock track from Cambridge-MT, measured):
I1 Drums / Bass / ElecGtr1–6 / LeadVox ← 开场六把吉他
I3 +ElecGtr7,8 +BackingVox1
V1 +Organ1,2
C1 +HandClaps +Organ3 +BackingVox2 +VocalSFX1
V2 +Tambourine +BackingVox4,5
C2 +ElecGtr9 +BackingVox3 ← 到第二段副歌堆到九把吉他、五组和声
Note that itis not"everything that should enter enters at the opening"; instead, something new enters in almost every section.
5.3 Failure conditions
- Minimal instrumentation(a vocal plus one guitar demo, solo piano) does not have enough layers for entrances and exits. In this case "form" must be carried by register, dynamics, and texture density, not by adding and removing instruments.
- Certain electronic genresget their sense of section from filter sweeps and effect amounts, not from adding or removing instruments.
5.4 ★ The rhythm of timbre: the frequency of timbre changes is also a parameter to plan (Yang Liqing)
Harmony has "harmonic rhythm" (the number of root changes and the rhythmic character), and timbre has it too:the color rhythm formed by the frequency of timbre changes(Yang Liqing, volume 2, p0381). It sometimes aligns with and sometimes diverges from the rhythms of melody and harmony, and is usuallymore relaxedthan they are (p0381). The differences between styles of different eras are to a large extent embodied in these three points: frequency of timbre change, manner of combination, and manner of contrast (p0383).
How to set the frequencydepends on the structural function of the section within the whole piece (p0397, p0400):
| Section | Timbre rhythm |
|---|---|
| Theme presentation, summary and conclusion | Relaxed and unhurried. Example from the book: a 10-bar melody is carried through in one go by 3 flutes, and the accompaniment texture hardly changes |
| Development, connection, transition | Frequent, with large jumps. Example: each time the motive splits and shifts, the timbre configuration changes bar by bar, with not a single repeated combination |
| Fast-tempo music | Change with caution; changing too frequently can cause auditory and even psychological fatigue |
| Slow-tempo music | Moderately increasing the frequency of change is an effective means of reducing monotony |
| Long-form | Conserve; leave room for what comes later |
| Short-form | Increase variation appropriately |
It can also be donedeliberately lowering the frequency, treating a large passage as a "color block" (Boléro, Atmosphères); that is expressive intent, not oversight (p0397).
Usage: When writing the entrance/exit event table (§5.2), count along the way how many times the timbre changes in each section. The frequency in the exposition section and the development section should be clearly different; a single frequency for the whole piece is the version of the "flatness" described in §2.2 in the color dimension.
6. The second pass must not be copied exactly
Symptom: The second verse is exactly the same as the first verse, and the second chorus is exactly the same as the first chorus. When listeners hear it the second time, their attention drops.
DeSantis gave three types of variation techniques (Creating Variation 1/2/3): Mutation after cloning、Generational mutation、Note transformation。
The least effortful and most effective approach is not to change the notes, but to change the instrumentation—the second verseshould be emptier than the first verse, not fuller. This is counterintuitive: many people add things on the second pass, but in fact they should subtract.
★ This one is conditional; the two schools sit side by side: Fujitani says the opposite—if 1A has only piano and lead vocal, 2A should add drums or bass, or backing vocals (p0059), because one section has already come to a close, and if the arrangement is the same, listeners will feel "here we go again." Both are right;the difference lies in the starting density of 1A: if 1A is already full instrumentation, then 2A subtracts; if 1A is a minimalist opening, then 2A adds.The criterion is not add or subtract, but that 2A must be different from 1A. Fujitani gives two other labor-saving approaches: in an orchestral arrangement, change only the drum groove of 2A; in the first measure of 2A, do a break with only vocals, and enter only in the second measure (p0059).
See the sample in §5.2: A2 (the second verse) in the actual measurement isstrings and half the guitars drop out, thinner than A1.
Self-check action: List the differences between A1 and A2. If you cannot list a single one, go back and revise.
Failure condition: In some forms (such as loop-based house), repeating the second pass exactly is part of the style; but even so, there is usually variation on some layer (change a hat pattern, add a pad layer).
7. Hook placement: non-vocal memory points
Required item：arrangement_hook — something listeners can remember without relying on vocals.
It can be: a riff, a signature timbre, a recurring sonic event (a sample hit, a riser).
Why it is required: Missing this item is the main reason something "sounds okay but is not memorable." The vocal hook relies on words and melody, butinstrumental sections, intros, interludes, and outros have no vocals—what holds people during those times?
The three decisions of placement：
| First appearance | The earlier the better. Listeners' attention is most concentrated in the first 20 seconds |
| Recurrence position | At least 3 times, anddo not distribute evenly—even = mechanical |
| Variation | On the last occurrence, change it a little (switch timbre, shift octave, extend) |
Self-check action: Without looking at the score, can you hum this hook? If you can't hum it, it's not a hook.
8. Three kinds of endings
The default for generative models is a "3:30 fade-out" or just an abrupt stop. DeSantis gives three paths (p0308–0309):
① Mirror
Play the layered accumulation from the beginning in reverse.
仅底鼓 (8 bars) → +其他打击乐 (8) → +贝斯 (8) → +和声与旋律 (8)
→ −和声与旋律 (8) → −贝斯 (8) → −其他打击乐 (8) → 仅底鼓 (8)
Applicable: Songs that open with layered accumulation. Symmetry itself is the structure.
② Loop and add/subtract
Find a loop and repeat it, each passadding a layerorsubtracting a layer / thinning (p0309). The two directions give two kinds of endings:
| Direction | What you get |
|---|---|
| Add a layer each pass | Density and tension pile up all the way, ending at the fullest point |
| Subtract a layer or thin each pass | Let the song end weakly rather than strongly |
Difference from ①: ① is symmetrical addition and subtraction, piling up to the top first and then retreating along the same path; symmetry itself is the structure. ② does not require symmetry, and can add without subtracting, subtract without adding, or stop halfway through adding.
How to choose a loop: Choose asection that can loop indefinitely without seeming cut off. This is the same judgment as the premise of ③ fade-out; jagged, asymmetrical, rough material suits neither.
What to change each pass: You don't have to move only one layer each pass. The means of thinning are not limited to removing an entire part; reducing the number of notes of an instrument and shortening the decay time of sustained notes also count (see the subtraction section in §3).
★ It and the maximum-density method in the composition stage are a pair(see mc-texture-layering reference §6.2):
The maximum-density method first vertically piles up the fullest section and then subtracts downward, producing several sections of different densitysections;
loop and add/subtract is withina single sectionchanging density pass by pass. The former defines the section table, the latter defines how to move within a section.
Self-check: Does your addition and subtraction move every pass, or once every other pass? Moving every pass will make people hear a mechanical feel, the pass left open by moving once every other pass is exactly the time for listeners to confirm this loop。
③ Fade out
There is a premise; it is not just any fade：
- Electronic musicians are generally negative about fade-outs, because songs with fade-outsare hard to use in a DJ mix (p0309)
- The philosophical meaning is "this song may never end,"so a fade-out is only suitable when you are already stable on ending material suitable for infinite looping (p0309)
- Material not suitable for fade-out: jagged, asymmetrical, "rough" sections (p0309)
- Timing: If you are going to master,the fade-out should be done at the mastering stage—the perceived sound of a fade-out differs enormously before and after mastering compression (p0309)
Self-check action: Which one did you choose? If you can't say, then it defaults to fade-out, which is the AI's choice.
9. Diagnostic path: look up by symptom
If what comes back is wrong, start here.First look at what's measurable, then listen.
| Symptom | Most likely cause | Which section | Measurable? |
|---|---|---|---|
| Flat, lifeless, can't keep listening | Energy curve is monotonic | §2.2 | ✅ Segmented RMS |
| Gets noisier toward the end but never peaks | Only additions, no subtractions | §3 | ✅ Segmented RMS |
| Every section sounds the same | Instrumentation never changes from start to finish | §5 | ✅ Per-track onset density |
| The second section is boring | The second pass is a verbatim copy | §6 | ✅ Section similarity |
| Can't remember it after listening | No arrangement hook | §7 | ⚠ Semi-automatic |
| Ending is abrupt or drags | No ending method chosen | §8 | ✅ Final-segment energy slope |
| "Can't put my finger on what's wrong, it just feels like AI" | Most likely several of the above hit at once | §9 full table | — |
Check the measurable items one by one against the spec first, don't rely on repeated listening to find the problem—save your ears for the last step.
10. Correspondence with ARR-SPEC fields
This skill is responsible for filling in these ARR-SPEC fields:
| Field | Which section of this skill | lint |
|---|---|---|
arrangement.energy_curve |
§2 | #5 Must have a decrease |
arrangement.subtraction_events |
§3 | #6 At least one |
arrangement.roster[].entry/exit |
§5 | #7 Must have both entries and exits |
arrangement.hooks.arrangement_hook |
§7 | #8 Required |
form[].bars 's asymmetric design |
§4.2 | #3 Not all allowed to be multiples of 8 |
form[].events |
§5.2 | — |
Ending treatment written into form the final segment's note |
§8 | #14 Duration self-consistency |
Not filled in by this skill:spectrum_budget（→ mc-texture-layering）、
groove（→ mc-rhythm-section）、vocal（→ mc-vocal-direction）、
mix_intent（→ mc-mix-intent）。
11. Must pass after writing
Before passing mc-workflow §3.0's 20 items, go through them yourself first:
- The energy sequence is written out,and it is not non-decreasing
- The subtraction event can be statedin which bar, what is removed, and why there
- Read the roster in entry order; the layering of the instrumentation should come through
- At least one instrument's exit is not end
- arrangement_hook can be hummed without looking at the score, and its recurrence positions are uneven
- A2 and A1 can list at least one difference
- Section lengths are mostly multiples of 8, butbreak it at least once, and you can say why
- The ending is one of the three types, and you can say which
- Any place where "not followed," state thefailure condition(which exception was taken)
The last one is the most important: every rule in this skill comes with a failure condition, not following it is allowed; not being able to give a reason is not.
Appendix: Explanation of evidence levels
The rules in this skill are divided into three tiers, marked one by one in the text:
| Tier | Meaning | In this skill, |
|---|---|---|
| Hard rules | Tested on 62 real arrangements, hit-rate lower bound ≥80% | Energy must have a decrease, an instrument must enter midway, at least one instrument must exit early, the opening should not be fully filled |
| Tendencies | Hit rate 50–80% | At least one obvious subtraction (70%) |
| Techniques | Advocated by textbooks, but not supported by testing as rules | The drop before the climax (§2.3) |
Test corpus: 60 Cambridge-MT multitracks (2,906 independent instrument tracks) + 2 finished Chinese pop songs, 62 total, each reverse-engineered into ARR-SPEC and then hit statistics computed.
This "hit rate" is the fundamental difference between this library and ordinary arrangement tutorials: A rule is not "some teacher said so", but "how many of 62 real arrangements do this". If the hit rate is low, this skill does not write it as a rule.
Main source: Dennis DeSantis, Making Music: 74 Creative Strategies for Electronic Music Producers (Ableton), page numbers refer to that book. For the rest, see reference.md。