music/LEITMOTIF_TECHNIQUE.md
CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.
Canon served: the composition floor atdocs/spine/DECISIONS_PENDING_JOSH.md(commitda627060), and the
theme rows ofregistries/T0_Theme_Registry [DRAFT v0.1](head_cell_spec,
transformation_plan,deny_register_list).
If this document disagrees with canon, CANON WINS and this document is the defect.
CANON SUBORDINATION — this document is PROPOSAL-TIER: it serves canon and never outranks it.
The canon: the CVD · the T1 foundation docs · the T0 registries (registries/) · the spine
(docs/spine/CH_*.md) · the region pages (_source/02_Tier_2_Region_Pages/). Authority order:docs/DOC_MAP.md§ 0.
Canon served: docs/spine/DECISIONS_PENDING_JOSH.md L5482-5500 (the composition floor, Josh verbatim) ·
registries/T0_Theme_Registry(the row this craft is realised onto) · themusicwork-kind read-set
printed bypython harness/route.py music(spinemusic_moodanchor → region page Section 10 → the registry row).
READ THAT CANON FIRST — open it and derive from it before you build anything from this document.
If this document disagrees with canon, CANON WINS and this document is the defect.
Tier: RESEARCH SYNTHESIS, proposal-tier. Every substantive claim carries a source — URL, or
book/author/chapter. Where sources disagree the disagreement is stated. Where a claim is craft
consensus rather than evidence it is labelled CRAFT CONSENSUS. Nothing here ratifies canon, and
nothing here overrides docs/proposals/music/LEITMOTIF_ARCHITECTURE.md — that document owns the
roster, this one owns the craft underneath it.
LEITMOTIF_ARCHITECTURE.md answers which motifs exist, what class each one is, what it may not do,
and where it is gated. It does not answer how a motif is built so that it survives thirty years, how
it is transformed without ceasing to be itself, or how any of that is checked rather than asserted.
This document is that layer. It is written for two readers who are not the same reader:
harness/music_gen/theme_compositions.py (the notes), author_head_cell.py (the battery that checks them), to_midi.py (the notation seam),
compose_arm_b.py (the arrangement arm) — which realises and verifies those notes deterministically.
The standing constraint that shapes every recommendation below: text-to-audio generation cannot
reliably realise a specified melodic shape, which was A/B tested and recorded
(LEITMOTIF_ARCHITECTURE.md §5, the 2026-08-05 supersession block). So no principle here is allowed
to depend on a generator producing a named interval sequence. Every principle is tagged
COMPOSER-FACING (a person decides it) or PIPELINE-FACING (a script can check or produce it), and a
few are both.
The floor this serves is Josh's, verbatim, at docs/spine/DECISIONS_PENDING_JOSH.md L5482-5500:
hooks and complexities, no stacked noise, no repetitious loop past three to five cycles without a new
instrument or catch or melody or drop, and the explicit instruction to learn what someone with a
doctorate in music would focus on. The through-line of this research is that the answer to "hooks and
complexities" is not more layers. It is one small, distinctive, transformable object, deployed with
discipline across a very long form.
The most consequential correction the literature makes to the popular idea of a leitmotif is that the
recurring identity is frequently not the melody. Wagner's motifs are typically one or two bars long,
and their identity is carried by whatever combination of interval content, harmony, rhythm and
orchestral colour makes them unmistakable
(Carter, "Meaning in the Motives", UT Austin,
p. 1: motifs "often no more than one or two measures long" are described as the bricks of the
sixteen-hour cycle). Carter's own worked examples show identity living below the tune: the Ring motive
is essentially a vii°7 outline built entirely of minor thirds, and Alberich's Curse is defined by a
tritone from its first note (F-sharp to C-natural), the single most identifiable dissonant interval in
the tonal system (Carter, pp. 4-5).
This matters directly to us because two of our twelve cardinal rows are deliberately not tunes — the
wonder identity is a harmonic-transformation identity and the voice colour is a timbre
(LEITMOTIF_ARCHITECTURE.md §0). The Wagner literature says that is not an exception to leitmotif
practice; it is the mainstream of it.
Wagner did not coin "leitmotif" and did not like what was made of it. The term was popularised by Hans
von Wolzogen's 1876 thematic guide to the Ring; before 1879 Wagner used a scatter of his own terms —
melodisches Moment, thematisches Motiv, Ahnungsmotiv, Grundthema, Hauptmotiv — and preferred
Grundthemen (Grove Music Online, "Leitmotif" (Whittall)).
In his 1879 essay Über die Anwendung der Musik auf das Drama he complained that a younger friend had
treated the motifs "more from the point of view of their dramatic import and effect than as elements
of the musical structure" (quoted in
Kassof, "On the Trajectory of Leitmotif in Götterdämmerung",
p. 3).
The working lesson is a design lesson, not a historical footnote. A motif built to be a label — this
figure means the villain — degrades into what critics call musical Morse code. A motif built as a
musical structure first can carry meaning and still be developed, combined and reharmonised, because
it was constructed to be musically load-bearing.
Wagner's most powerful device is simultaneity: two or three motifs sounded together, in the voice and
in the orchestra, to say something no single motif and no line of text says. Carter names this
directly — one method of commentary is juxtaposing two or three motives at once to expose a
character's hidden intentions (Carter, p. 1). Kassof's analysis of the Immolation Scene closes on the
same device: the Love and Redemption motifs sound simultaneously into the final chord of the entire
cycle, which is what makes them the cadential motifs of the Ring (Kassof, p. 6).
Two numbers from Kassof worth carrying: the Ring cycle contains well over one hundred identifiable
leitmotifs across roughly sixteen hours, and the single Immolation Scene uses twenty of them, of which
Loge's Magic Fire recurs most often and the Curse of the Ring appears exactly once (Kassof, pp. 5-6).
A hundred motifs is a corpus, not a listening budget; twenty in a scene is the density that actually
reaches a listener, and one of the twenty is a single-use payoff.
The scholarly consensus is that motivic evolution and development, inside large-scale tonal structure,
is the centre of Wagner's language rather than the labelling of motifs
The operation set that this literature and the later Schoenbergian tradition share:
without changing identity.
separates basic from profound reharmonisation for exactly this reason, pp. 47-48).
remains; distinct from fragmentation, which keeps a characteristic piece
(Wikipedia, Fragmentation (music));
Wikipedia, Developing variation).
one basic motif, described in Fundamentals of Musical Composition as growth rather than repetition
(Wikipedia, Developing variation, quoting Schoenberg).
There is a real scholarly caution against treating a score as a dictionary of labels. Wagner's own
complaint is the earliest version; the modern version is that naming is interpretive and inconsistent.
Lehman states it about his own catalogue: motif labels may be "somewhat arbitrary or open to
interpretation", and there is "no one right way to analyze leitmotif", only ways more or less
self-consistent (Lehman, p. 56).
The caution is not merely aesthetic — it is measurable. An empirical study of leitmotif perception in
the Ring ran two experiments (100 participants, then 31) with surprise recognition tests over twenty
leitmotives ([Baker and Müllensiefen, Perception of Leitmotives in Richard Wagner's Der Ring des Nibelungen,
Frontiers in Psychology 8:662, 2017](https://pmc.ncbi.nlm.nih.gov/articles/PMC5415611/)). The
strongest predictor of recognition was Wagner-specific expertise (β = 0.39, p < .001); general musical
training was not significant; structural melodic complexity hurt recognition of already-known motifs;
and audio-domain chroma distance was the strongest predictor for novel items (β = 0.71), with model
accuracy around 68-69 percent. The honest reading is twofold. Motif recognition is learnable by
ordinary listeners through repeated exposure rather than training, which is exactly what a
seventy-nine-node game supplies. And complexity is measurably against you at the moment of
recognition, which is the empirical case for a short, simple head cell and a complex treatment.
Berlioz's idée fixe in the Symphonie fantastique (1830) is one theme returning in every movement,
transformed to fit each dramatic situation
(Britannica, idée fixe). Liszt generalised it into
thematic transformation: variation in which a theme becomes something new, separate and independent
while remaining audibly the same material
(Britannica, transformation of themes;
Wikipedia, Thematic transformation).
This is the technique a game score actually needs, more than Wagner's dense motif-web. One theme, many
emotional states, one identity — because a game asks the same theme to be a title screen, a quiet
camp, a loss, a chase and a final ascent. Our protagonist row already commits to precisely this shape:
the Prologue statement and the Chapter 77 statement are the same tune and the player must feel the
distance (LEITMOTIF_ARCHITECTURE.md §3.1).
Lehman's catalogue is the most explicit published methodology for this question and it is directly
portable to a registry. Four base criteria, then three that promote a motif to principal status
(Lehman, p. 56):
| criterion | test |
|---|---|
| Distinctiveness | clear, unique melody, not wholly derived from another motif |
| Recurrence | intentionally repeated in more than three discrete cues |
| Variation | the repetitions are not exact |
| Intentionality | the repetitions are compositionally intentional, not detective work |
| Abundance (principal) | occurs in more than one film, with more than ten iterations overall |
| Meaningfulness (principal) | attaches to an important subject and accrues meaning through recontextualisation |
| Development (principal) | not only varied but significantly developed and transformed across iterations |
Everything that recurs but fails these is filed as an incidental motif, and everything confined to one
cue is a set-piece theme. That three-tier split — leitmotif, incidental motif, set-piece — is the same
shape as our cardinal / Tier-B family / Tier-C region-cell split, arrived at independently, which is
mild evidence the allocation is sound.
Lehman separates the scope of a transformation (single cue, cue complex, single score, multiple
scores, single concert arrangement, multiple concert arrangements) from its technique, and splits
technique into micro-transformations (mutation) and macro-transformations (evolution), following
Bribitzer-Stull 2015 (Lehman, pp. 47-48).
extension and fragmentation, mode change, sequencing.
of thematic form (for example phrasal foreshortening), tonal motivicism (a theme's characteristic
progression spreading into non-thematic underscore), concealed repetition, developing variation,
motivic de/reconstruction, teleological genesis (a motif whose full form is the end-point of a
score-long cumulative expansion).
Two of these are worth flagging as under-used in game scoring and directly available to us. Tonal
motivicism gives a way to make an antagonist present in a region without ever stating the motif — the
progression alone travels. Teleological genesis is the formal name for the withheld theme: the
complete form arrives once, late, as the destination of everything before it.
Lehman's thematic census counts leitmotif statements per film under explicit rules — partial and
fragmented statements count when the reference is clear to a sensitive listener; exactly tracked
repeats do not; internally repeating or sequencing themes generally count once (Lehman, p. 63). Cue
counts per film run 43 to 184, with Episode IV at 107 and Episode VII at 184 (Lehman, p. 64), and the
saga is described as one of the longest sustained developments of a single set of musical materials in
film history, at nine films over forty-two years (Lehman, p. 47).
The transferable ratio: a two-hour film sustains on the order of a dozen to two dozen distinct
leitmotifs with statement counts in the low hundreds. A seventy-nine-node game is roughly forty times
that runtime. Scaling motif COUNT by runtime is the error; scaling statement count is the correct
move. This is the quantitative argument for our own recognition-budget ruling — twenty-two independent
thread themes produce mush, families do not (LEITMOTIF_ARCHITECTURE.md §2).
Lehman's most game-relevant finding is that in Star Wars, individual keys carry no consistent meaning
but PROGRESSIONS between chords do (Lehman, p. 44). The G-minor to E-flat-minor motion (the Tarnhelm
progression, neo-Riemannian LP(m)) attaches to Vader in Episode V and remains a symbol of evil through
every later film, appearing in Imperial March, Death Star, Kylo Ren B, Krennic and others. A minor
triad rising a perfect fifth reads as decisive action and war; a major-third-related pair reads as
heroism.
For a game this is the cheapest possible recurrence vehicle. A progression survives being played by an
ensemble that shares no tuning system with the last one, it costs no melodic real estate, and it can
be present in ambient underscore that contains no theme at all.
Uematsu's working method is melody-first and physical: he described going into the studio daily,
switching on the synthesiser, playing, and writing down the first two or three phrases that came into
his head (Red Bull Music Academy interview, 2014).
The compositional consequence is that the material is generated as short singable units and only then
developed into themes, battle cues and character leitmotifs.
The hardware shaped the voicing, and the constraint is instructive rather than historical. The SNES
sound system (Sony SPC700) offered eight voices of polyphony and 64 KB of audio RAM
(Wikipedia, Music of Final Fantasy VI).
Eight simultaneous voices is a hard layer budget. Under it, arpeggios were used to imply chords that
could not be sustained as blocks, samples were reused and compressed aggressively, and the melody was
given priority over everything else — which is exactly why the era's tunes are the ones people still
hum. Josh's floor calls for dozens of layers; the SNES lesson is that layer count and memorability are
independent axes, and that the discipline of a budget forces the melody to carry.
The architectural practice is theme-and-many-arrangements: the Prelude and the Main Theme reappear
across an entire title and across the series, and a character theme such as Terra's receives distinct
orchestral rearrangements within the same title and again in concert form
(Final Fantasy Wiki, Terra's Theme). One tune,
many dresses, is the JRPG equivalent of Liszt's thematic transformation.
Kondo's GDC 2007 talk, Painting an Interactive Musical Landscape, is the primary source and it names
three principles: rhythm, balance, interactivity
archived session, Internet Archive). Rhythm for Kondo does
not mean a catchy beat — it means the music's pulse matching the animation and the button presses.
His stated failure mode: music that does not reflect the game's rhythm "might as well be piped in from
a source outside the room" (Game Developer). He also named immediate recognition as a design target:
a cue should be identifiable within its opening notes so it signals location or status (Nintendo World
Report).
The interactive techniques he demonstrated are a deployment catalogue in themselves — instrument
substitution by player location in Mario 64 (shore, swimming, cavern), percussion tracks gated on
riding Yoshi in Super Mario World and Sunshine, sparse-to-full arrangement as the player approaches
Castle Town's centre in Twilight Princess, and enemies animated to chord stabs in New Super Mario Bros
(Game Developer). The Hyrule Field daytime music is the landmark case: twelve phrases are selected and
stitched into an eight-phrase melody, and enemy proximity transitions the material toward ominous
phrases while the theme's integrity holds (Game Developer; Nintendo World Report).
On melodic construction, his own account in Iwata Asks is that the Famicom's three sound channels
pushed him toward material that would be distinctive and would still read as music, and that he
deliberately reused the same themes — above-ground, underground — in different arrangements across
levels so the melodies would stick with the listener
(Iwata Asks, Super Mario All-Stars, Music Commentary by Koji Kondo).
That is theme-and-arrangement stated as a memorability strategy rather than an economy measure.
The diegetic layer is the other half. Ocarina of Time's songs are motifs the PLAYER physically
performs, which is the strongest possible form of motif learning: the player rehearses the interval
sequence with their hands. Academic treatment exists (Gervais 2015 on the teleportation pieces; Tim
Summers, The Legend of Zelda: Ocarina of Time — A Game Music Companion, Intellect 2021,
publisher page), and our
own refusal fingerprint set already carries twelve of these ocarina sequences as interval-only
signatures (harness/music_gen/exemplar_hook_signatures.json).
Mitsuda is over-represented in our corpus for a reason: seven of the tracks Josh named as staying loved
for decades are his Chrono Trigger rows (EX_006 Main Theme, EX_007 Frog's Theme, EX_022 Wind
Scene, EX_033 Corridors of Time, EX_102 Magus Castle, EX_113 To Good Friends, EX_162 Schala's
Theme, verified in build/audio/exemplars/CORPUS.json).
His own account of melodic construction is modal: he has said the melodies of his songs are often
created using modes
(MuseScore, How Yasunori Mitsuda Composed Corridors of Time).
The Corridors of Time model is an ostinato plus a melody whose tonal centre is deliberately ambiguous:
the piece is in F-sharp minor but progresses mainly in D major, which is what produces its floating,
melancholy quality (MuseScore). Schala's Theme is in B-flat minor with the sequence pattern avoiding
D-flat entirely (MuseScore). Both pieces use tabla, and that shared instrumentation is what makes them
read as one place — the Kingdom of Zeal — rather than two cues; the two also share a four-note-per-bar
bell figure and an eight-note melodic cell of similar shape
(Video Game Music Daily analysis).
Frog's Theme shows the opposite trick: a G-sharp minor base with almost every chord major except the
root, which makes a minor-key tune read as heroic rather than sad, achieved through substitute chords
(Video Game Music Academy analysis;
That is reharmonisation used to set emotional register while the melody is untouched — the single most
efficient transformation available to a game score, because the tune stays recognisable at zero cost.
Chrono Cross extends the practice to whole-score recomposition. Its soundtrack is built substantially
from material first written for Radical Dreamers, adapted at Masato Kato's suggestion, with several
themes carried over largely unchanged except for new instrumentation — Gale, Frozen Flame, Viper
Manor, Far Promise ~ Dream Shore, The Girl Who Stole the Stars
(Wikipedia, Music of Chrono Cross). The lesson
for us is that re-instrumentation alone is a legitimate and sufficient transformation when the source
material is strong; the cheapest variation is the carrier change, and our familiar-bond row already
banks on exactly that across twenty-two species carriers (LEITMOTIF_ARCHITECTURE.md §3.3).
The non-Western instrumentation in Mitsuda is structural rather than decorative — tabla is not colour
on Zeal, it is Zeal's identity condition. This is the craft standard our region cells are held to:
instrumentation copied from the region page's own Section 10, never composed generically
(LEITMOTIF_ARCHITECTURE.md §3.6).
Sakuraba came from a keyboard-driven progressive rock band (Deja-vu) before game music, and his
approach is performance-first: he plays everything by hand and avoids sequenced music, which is his
stated reason his tracks do not sound mechanical
(shmuplations, Motoi Sakuraba 2002 composer interview). His
stated priority is battle music specifically, because players hear it far more than anything else, and
his design target is that it stay engaging over many repetitions rather than becoming tedious
(shmuplations). He also described a pragmatic workflow — make something quickly and revise later
rather than chase perfection up front (shmuplations).
EX_147 Fighting of the Spirit (Tales of Phantasia) is our corpus row for this class. The
prog-derived traits that make his battle music read as complex are the fast harmonic rhythm (chords
changing at or faster than the bar, so the ear cannot settle), keyboard-idiomatic figuration that a
human is actually playing, and an extended form with real contrasting sections rather than an
eight-bar loop. This is the direct answer to Josh's three-to-five-cycle law: Sakuraba's solution to
repetition fatigue is not a longer loop, it is a battle theme with the internal sectional variety of a
prog song.
CRAFT CONSENSUS, flagged as such: the specific odd-meter reading of individual Sakuraba battle themes
circulates widely in fan analysis but I found no peer-reviewed or composer-confirmed metric analysis of
Fighting of the Spirit. Treat metre claims about specific tracks as unverified until measured off our
own card.
Magic Orchestra named as influences, and for Symphony of the Night she analysed the earlier
Castlevania scores to find what made them work before writing
(Game Developer, classic interview;
VGMO interview). The published technical study of
Dracula's Castle finds two melodic germs — a four-note fragment and a six-note triplet figure —
developed recursively across the track, G minor destabilised by a foreign F-sharp against C (a
tritone), and orchestration used to mark the emotional turn
Two germs per track, developed, is a usable density target for a Tier-C region cell.
series constant by being re-arranged for every entry — hazier and more ephemeral for Chain of
Memories, sorrowful with violin for Birth by Sleep
This is the single-theme-many-arrangements practice run over decades and it is the clearest evidence
that a title-screen theme is the most valuable motif real estate in a series.
monster and altered their melodies to reflect relationships, feelings and antagonism
(Vagrant Story Wiki, Music). Director
Matsuno pushed him off the bright Tactics register toward deeper, heavier writing (same source). His
signature density is contrapuntal — many independently moving lines rather than melody-plus-pad —
which is the craft answer to "there cannot just be stacked noise": the cure for stacked noise is
counterpoint, where every layer has its own line.
an orchestral and choral arrangement of the earlier theme rather than a variation on it, with Todd
Howard's brief being the series theme as if sung by a barbarian choir
(Wikipedia, Dragonborn (song))). A franchise-level
identity carried by re-scoring rather than re-composition.
the whole game, with the cello as the player's voice and the rest of the orchestra as the world —
effectively a cello concerto in which the player is the soloist
Motif-as-progression: the theme's state at any moment is a readout of the player's arc.
Games, covering how characters and biomes are portrayed and specifically how interest is sustained
across repeated runs
Repeat-play tolerance is a first-class design problem in a roguelike and in a seventy-nine-node RPG
alike.
parents are Fox's own earlier work (Metal Crusher and The Door from Cucumber Quest, Patient from
Homestuck) and external sources, and the title track Undertale is itself built from two character
themes — Memory (Asriel) and Once Upon a Time (Chara)
(Leitmotif Index, origins analysis;
Undertale Wiki, Leitmotifs). There is a
scholarly treatment of the practice
(Queen's University, Connecting Unspoken Elements through the Use of Leitmotif in Undertale).
The honest caveat that the community itself records: players keep finding new connections and a
conclusive list may not be possible — which is a warning about over-labelling in exactly Wagner's
sense, now reproduced in a game score.
Synthesising Lehman's tiering, Kassof's density figures, the perception study, and the practice of the
JRPG masters, the defensible shape for a very long game is:
between the two statements as the score's whole argument (Journey; Chrono Cross; our
PROTAGONIST_THEME).
the film-score number and matches the perception study's finding that recognition is exposure-driven
and complexity-averse. Our twelve cardinal rows sit exactly here.
and a shared harmonic signature, with each member differing by one substituted interval and a
distinct carrier voice. Lehman's leitmotivic family networks name the same structure in Williams
(Vader, Emperor, Kylo Ren, Psalm of the Sith share melodic and harmonic affinity, Lehman p. 48).
coherent with the parent and built from the place's own attested instruments.
felt in music that states no theme at all (Lehman p. 44).
The relation rules that make this a system rather than a collection:
changes most often — but it is the strongest bond WITHIN a region, which is why Mitsuda's tabla makes
Zeal one place.
theme cleanly once before fragmenting it is visible throughout Lehman's catalogue, where the initial
statement of each motif is recorded as a specific timecode and cue.
statement reads as recognition instead of introduction. This is what our companion row does for ten
chapters (LEITMOTIF_ARCHITECTURE.md §3.2).
anything. Lehman's fragmentation entry is the technique; the reveal discipline is ours. Our
antagonist pre-gate row permits fragmentation and nothing else before Chapter 38, because
augmentation, combination or a mode shift would each announce that the score has an opinion
(LEITMOTIF_ARCHITECTURE.md §3.4).
moment the relation between two subjects is the point. Wagner ends the Ring on it.
destination of everything that preceded it.
cadence for Chapter 77 and forbids the three endings any shared arrival gesture
(LEITMOTIF_ARCHITECTURE.md §3.1).
Between these, the runtime layer is horizontal resequencing (cue A loops until state changes, a
transition plays, cue B begins) and vertical remixing (layers added or removed live), which are the two
standard adaptive techniques
(Writing Interactive Music for Video Games, ch. 8;
Kondo's Hyrule Field phrase-stitching is horizontal resequencing at the phrase level, a decade before
the vocabulary existed.
Our canon carries a hard rule: no region's own register is used as the threat signal, systematically
darkened, inverted or menace-coded; the darkening operations belong to the antagonist material and
travel with it (LEITMOTIF_ARCHITECTURE.md §4). The craft literature that supports this is the
musical-exoticism scholarship. The core finding is that exotic scoring conventions essentialise their
subjects and reference other exotic musics rather than the music of the people depicted — drones,
jingling ornaments, and the handful of stock riffs that function as caricature
(The Key Debates of Musical Exoticism and Orientalism in Historical Musicology, MUSIC.OLOGY.ECA).
The specifically game-music treatment names the mechanism: the repetition of stock melodic motifs such
as the snake-charmer or oriental riff functions as a racist caricature carried forward from older
traditions (Unwinnable, On Orientalism and Exoticism in Videogame Music).
The craft consequence is the one our canon already states and is worth restating in technical terms:
menace is routed through SEPARATE MATERIAL that travels with the antagonist, not through a treatment
applied to a place's own register. Williams supplies the working model — the Tarnhelm progression
follows Vader from film to film and into other villains, and is never a property of a planet.
The best-powered study is Jakubowski, Finkel, Stewart and Müllensiefen, Dissecting an Earworm,
Psychology of Aesthetics, Creativity, and the Arts 11(2), 2017, 122-135
open PDF). Method: 3,000 survey
respondents named earworm tunes; 100 frequently named tunes were matched against 100 never-named tunes
on popularity, recency, chart position and style using a genetic matching algorithm; 83 melodic
features from the FANTASTIC toolkit plus tempo were fed to a random forest.
The findings, stated precisely because the popular summaries get them wrong:
are arch-shaped phrases, consistent with the melodic arch being the commonest contour in Western
traditions (Huron 1996, cited by the authors; see
The Melodic Arch in Western Folksongs).
intervals between direction changes are unusual, typically because the tune contains a few large or
unusually large leaps.
only marginally significant in the follow-up logistic regression (p = .10).
dens.int.cont.grad.mean, and dens.step.cont.glob.dir.
The honest limits, which the authors state themselves:
2012 that earworm tunes use longer note durations and smaller intervals. Sources disagree here, and
the more recent, better-powered study is the one that failed to replicate. Any rule of the form
"keep the intervals small" is therefore unsupported.
not a result.
HOW OFTEN. Exposure does most of the work, which for us is an argument for statement count.
A second empirical line worth carrying: Burgoyne et al., Hooked: A Game for Discovering What Makes
Music Catchy (ISMIR 2013, PDF) defines
catchiness as long-term musical salience and finds that different sections of the SAME song differ
significantly in recognition speed — some sections are simply better hooks. And Müllensiefen and
Halpern 2014, summarised inside the Jakubowski paper, found explicit memory for tunes enhanced by
motives that are RARE relative to a corpus combined with FREQUENT repetition of all motives.
Combining the evidence with the practice:
melodic prescription available, and it is precisely what our head-cell law already encodes: a
declared contour class with exactly one deviant.
three to four notes. CRAFT CONSENSUS on the exact number; the supporting evidence is the perception
study's finding that structural melodic complexity hurts recognition.
March's characteristic rhythms as one of the three carriers of the Anakin-Vader connection alongside
melodic contour and mode (Lehman p. 45) — but not isolated as a predictor in the earworm work.
easier to sing, per the earworm authors' own discussion). Label as CRAFT CONSENSUS.
survive three Famicom channels, Shimomura's survives four re-arrangements, Soule's survives a
barbarian choir. If a motif needs its orchestration to be recognisable, the orchestration is the
motif and it will not travel.
build/audio/exemplars/formula_cards/EX_124.json measures, per exemplar, the following fields that
bear directly on this document:
| card field | what it lets us check |
|---|---|
the_hook.onset_s / onset_bars | how soon the identity arrives (EX_124: 0.104 s) |
the_hook.pitch_contour | the U/D/- string — direction changes and turning points |
the_hook.interval_vocabulary_semitones | how many distinct interval sizes the head uses |
the_hook.intervals | the interval sequence itself, which is what the fingerprint screen consumes |
the_hook.range_semitones | ambitus, against the singable-ninth bound |
the_hook.distinctive_deviation | largest interval, its note index, and whether it is a licensed leap |
motif_economy.self_similarity_peak / baseline / lift | whether anything actually recurs above the track's own texture |
motif_economy.recurrence_lag_s | how long between a statement and its return (EX_124: 46.07 s) |
motif_economy.interval_3gram_repetition / 4gram | internal motivic economy |
duration_and_structure.repeat_map | which sections are repeats of which, with a similarity number |
duration_and_structure.novelty_peak_strength | where the score actually introduces something |
tonality.windowed_key_agreement | the modulation signal (below about 0.7 reads as modulation) |
opening_and_flow_law.time_to_hook_s / _bars | the identity-arrival law |
The card's distinctive_deviation block is an unusually direct implementation of the earworm finding:
common contour plus one uncommon gradient is exactly a declared contour class plus a single licensed
leap, and the card reports both halves.
harness/music_gen/exemplar_hook_signatures.json is a REFUSAL fingerprint set, not a generation
input: thirteen interval-only signatures (one Super Mario Bros opening, twelve ocarina sequences), with
refusal on a contiguous interval match of six or more or on a complete match of a reference of length
four or more, and with inversion and retrograde hits printed but never refusing. It declares its own
gaps honestly — fourteen named exemplars have no fingerprint because no reliable transcription was
available, and the file states that CLEAR means no seed signature matched rather than clean.
harness/music_gen/motif_compare.py answers whether a claimed variation actually carries the motif,
in three stages: V1 held-region identity by waveform correlation, V2 pitch-class profile cosine
(explicitly reported and never relied on), V3 interval n-gram Jaccard over the extracted melodic line,
and V3b the same comparison restricted to the newly generated region, which is the deciding block. Its
thresholds: 4-gram Jaccard at or above 0.15 or 3-gram at or above 0.30 means motif material survives;
3-gram at or above 0.12 is weak and unsafe; below that the word variation is refused.
What these two can see: whether a candidate reproduces a known hook; whether two audio artifacts share
interval content; whether a head cell holds its declared contour, deviant, response operation,
continuation-by-fragmentation, harmonic signature, closure and ambitus (the eighteen-check battery in
author_head_cell.py, C01 through C16, including negative controls that must fail when the deviant is
removed).
What they cannot see:
motif_compare V3 uses intervals, so it is transposition-invariant by construction — but only for the melodic line the rig extracts. V2 is pitch-class based and is not.
diminished. Interval n-grams ignore duration entirely, so an augmented statement and a diminished one
are the same vector, which is convenient for detecting them as related and useless for CHECKING that
an augmentation was actually performed.
motif_economy.self_similarity_peak is computed on achroma-plus-MFCC self-similarity matrix, so a theme returning a minor third higher scores as less
similar than the same theme returning in the same key. The card therefore under-reports exactly the
transformation the literature calls most characteristic.
to_midi.py extracts the predominant melodic line only, one voice, andsays so. Two motifs sounded simultaneously is invisible to every measure we have.
tonality.mode is a single global estimate; there is no per-statementmode readout, so a darkened variant cannot be verified as a mode change rather than a different tune.
confidence as low for roster purposes; it cannot say the flute statement and the shakuhachi statement
are the same motif.
circular rotations of the chroma vector and take the per-cell maximum, recording the winning rotation.
This turns "the theme returns in a new key" from invisible into a measured event with a transposition
interval attached. Highest-value gap by a distance.
audio-domain extraction to compare them. The MIR literature's best-performing family for this is the
geometric spline model with local alignment over n-gram spans, which is transposition- and
time-scale-invariant and produced the best MIREX results reported for the 2005 collections
(Symbolic Melodic Similarity: Local Alignment with Geometric Representations).
Edit distance over interval sequences is the cheap first version.
(interval, log-duration) sequence with a warp path whose average slope is reported gives a direct
readout: a warp slope near 2.0 is an augmentation, near 0.5 a diminution. This is the check that makes
the transformation catalogue enforceable rather than declarative.
and report tonic plus quality per statement. That makes "the darkened variant" a measurable claim.
the audio domain — but trivial on the authored side, where the score exists as data. Recommendation:
do not attempt it in audio; assert it from the authored MIDI and treat the render as evidence only.
candidate that reproduces a famous rhythm with different pitches, which is a real reproduction risk
for anything as distinctive as the Imperial March's dotted figure.
and decide which of the four is the load-bearing one. COMPOSER-FACING. PIPELINE HOOK: the
motif_class column already routes tune, harmonic-transformation and timbral identities to different
acceptance tests; author_head_cell.py C02 and C07 check contour class and harmonic signature
separately, so a harmonic identity is not failed for lacking a hummable line.
(Jakubowski et al. 2017). COMPOSER-FACING and PIPELINE-FACING. PIPELINE HOOK:
the_hook.pitch_contour plus the_hook.distinctive_deviation.is_licensed_leap on the card;
author_head_cell.py C03 enforces exactly one deviant of the declared class and C15 proves the
detector reports its absence when the deviant is removed.
later fragment is then recognition rather than introduction. COMPOSER-FACING. PIPELINE HOOK: the
registry's plant_node column exists and is deliberately unpopulated
(LEITMOTIF_ARCHITECTURE.md §5); the card's opening_and_flow_law.time_to_hook_bars is the
arrival-time measure for the plant cue itself.
continuation, cadence) with the continuation built from the head cell. COMPOSER-FACING.
PIPELINE HOOK: theme_compositions.py carries an explicit continuation_fragment interval list per
theme and author_head_cell.py C06 fails a continuation that introduces new material.
reharmonisation, fragmentation, carrier migration, contrapuntal combination, closure change; six to
ten per cardinal theme across the whole game, each spent at a named node. COMPOSER-FACING.
PIPELINE HOOK: author_head_cell.py TRANSFORMATIONS currently realises two of them
(COMPANION_HEALERS_CHILD_HEAD_CELL_ALONE, FAM_THE_OTHERWORLD_MINOR_RECURRENCE); the remaining
operations need the DTW and per-statement-mode measures of §8.3 before they can be checked rather
than asserted.
one substituted interval and a distinct carrier per member. PIPELINE-FACING. PIPELINE HOOK:
parent_theme_id is a declared self-edge in docs/fk_spec.json with zero dangling tokens, and
harmonic_signature_holds() in author_head_cell.py is the family-audibility check.
motif count; the perception evidence says exposure, not training, is what makes a motif land.
COMPOSER-FACING and PIPELINE-FACING. PIPELINE HOOK: the recurrence_anchors column is the statement
ledger and is deliberately populated per node by the lane that read that node; the card's
motif_economy.recurrence_lag_s and self_similarity_lift measure whether a recurrence is audible
within a single track.
restated as a scoring rule, with an associative progression as the carrier so the presence can be felt
without a theme. COMPOSER-FACING. PIPELINE HOOK: every theme row's deny_register_list already
carries the universal clause (LEITMOTIF_ARCHITECTURE.md §4); the missing measure is a per-cue check
that a region's own instrumentation set never co-occurs with the antagonist material's darkening
operations, which is checkable from the cue table once cues carry both fields.
the late arrival, contrapuntal combination for the one moment two subjects are the point.
COMPOSER-FACING. PIPELINE HOOK: reveal_gate plus emit_music_cues.py's fail-closed refusal is the
mechanism for withholding; combination is currently unmeasurable in audio and should be asserted from
the authored score, with the render as evidence only.
run before anything else is spent on it. COMPOSER-FACING and PIPELINE-FACING. PIPELINE HOOK: render
the head cell through at least three carriers from sfz_palette.py and require
motif_compare.py V3b 3-gram Jaccard at or above 0.30 between every pair; the tooling for this exists
today and the check has not been wired.
Not every citation above carries the same weight, and pretending otherwise would be the same defect as
a proof number that no longer re-derives.
al. 2013; Huron 1996; Bribitzer-Stull 2015 (Cambridge); Summers 2021 (Intellect); Lehman's catalogue
(adapted from his chapter in John Williams: Music for Films, Television, and Concert Stage, Brepols
2018); the JSMG Korb interview; the exoticism review in MUSIC.OLOGY.ECA.
the shmuplations Sakuraba interview; the Red Bull Uematsu interview; the Square Enix Shimomura
interview; the Wintory interviews.
the Kassof honours thesis and the Carter course paper (the latter is a 2002 undergraduate essay
hosted on the UT Austin Wagner site; its Deryck Cooke framework is contested in the literature and is
cited here only for its uncontroversial descriptions of motif construction).
Game Music Daily and Video Game Music Academy readings of Mitsuda, and the MuseScore article. These
are the best available sources for their subjects and none of them is refereed.
Everything in §§1-7 derives from published sources, cited inline. Everything in §§8-9 derives from
reading our own stack: build/audio/exemplars/formula_cards/EX_124.json,
harness/music_gen/exemplar_hook_signatures.json, harness/music_gen/motif_compare.py,
harness/music_gen/to_midi.py, harness/music_gen/theme_specs.json,
harness/music_gen/theme_compositions.py and harness/music_gen/author_head_cell.py. The canon this
document serves and never overrides is the composition floor at
docs/spine/DECISIONS_PENDING_JOSH.md L5482-5500 and the roster in
docs/proposals/music/LEITMOTIF_ARCHITECTURE.md. Where this document names an outstanding
measurement it is a FINDING for a later lane, not a change to canon and not a change to any harness
file — this pass wrote exactly one file.