Is there a musical-scale equivalent for story structure? Clustering screenplay arcs to find out.

If AI takes the cognitive work, humanity defaults to stories and the arena - and we don't have a grammar for compelling narrative yet.

There is a pattern in how civilisations respond to labour displacement.

When conquest freed the Roman citizen from physical work, the response wasn't enlightenment. It was the Colosseum. Bread and circuses wasn't a failure of Roman civilisation - it was its rational adaptation to a newly idle population that still needed to feel something together.

The industrial revolution displaced physical labour. The 20th century was built on what followed: cinema, radio, television, recorded music, professional sport. An entire economic layer built on surplus human attention.

AI is displacing cognitive labour. The pattern will repeat.

If the machines are doing the thinking, what does humanity do with itself? My bet: the same thing it always does when freed from work. It tells stories. It watches the arena.

The production problem is already solved. The compulsion problem is not.

GenAI has effectively commoditised the production layer of storytelling. The cost of technically competent content is collapsing toward zero.

Here is the number that should concern anyone building in this space: By most audience metrics, only a small minority of produced content becomes truly loved or culturally relevant. That was true before GenAI. With AI lowering the barrier to production, the ratio of garbage to signal is about to get significantly worse - the YouTube moment for long-form narrative [1].

The bottleneck is no longer production craft. It is narrative compulsion. Not can you render it, but does it make someone feel something they will seek out again.

Music solved this problem formally, centuries ago. Storytelling has not.

[1] Before Youtube Introduced Analytics which helped see trends of what resonated and what didn't.

The Vonnegut baseline and why it's insufficient

Kurt Vonnegut proposed that all stories map to a small number of emotional shapes plotted on a good-fortune/bad-fortune axis. Man in a Hole. Cinderella. From Bad to Worse. A few years ago, a team at the University of Vermont ran NLP sentiment analysis on nearly 2,000 works of fiction and confirmed six core shapes empirically (Reagan et al., 2016).

This is real. It holds up. But it is also underspecified for what we actually need.

Vonnegut's shapes operate at the level of overall emotional trajectory. They tell you whether the protagonist ends up better or worse than they started. They don't tell you anything about the internal structure of how you get there - the sequence of emotional states, the timing of reversals, the signature patterns of specific genre grammars, or why the same "Man in a Hole" shape feels completely different in Finding Nemo versus No Country for Old Men.

The existing industry frameworks - Save the Cat, the Hero's Journey, the 12-point spine - are taxonomies of plot events, not structural grammars of emotional sequence. They tell you what needs to happen, not the shape of how it unfolds in time.

The raaga hypothesis

Indian classical music is built on the Melakarta system: 72 mathematically derived parent scales (mela kartha raagas), each with a fixed ascending and descending sequence of notes (swaras). From these 72 pure forms, musicians derive thousands of raagas - each with a precise emotional character, a rasa (emotional flavour) it reliably evokes.

The system separates the grammar of emotional structure from the act of performance. A raaga is not a performance, it is a structural truth about how a particular sequence of emotional states creates meaning in time.

Western music's equivalent is its 24 major and minor scales - a simpler, binary emotional system (major skewing toward resolution and brightness, minor toward tension and melancholy) that underlies the entire Western compositional tradition. Indian classical music is more granular: 72 parent structures encoding finer emotional distinctions.

My working hypothesis: narrative has an equivalent structure that we have not yet formally discovered.

Every film is what Carnatic music calls a janya raaga - a derived, hybrid form. Parasite starts as dark comedy, becomes thriller, ends as horror. No Country for Old Men is a Western, a thriller, and an existential meditation simultaneously. These are not genre failures. They are compositions in derived forms. The question is: what are the parent forms?

What we're building and what we're finding

We do quantitative structural analysis of screenplays at scene level: intensity score (0-100), duration, act position, peak flags, character presence. We've processed 240+ films this way.

The key methodological choice: We're working with narrative beat positions - the structural inflection points of a story. This aligns with Vonnegut's insight and gives us cleaner signal for shape comparison.

What we're finding so far:

Drama is not one shape. A grief arc (Manchester by the Sea) is structurally nothing like a moral failure arc (There Will Be Blood) or a redemption arc. Same genre label, categorically different shapes.

Genre transitions have signatures. When a film shifts genre - Parasite at the midpoint, Get Out from the opening scene - there is a specific structural pattern: a reversal that does not resolve back to the prior emotional register, followed by a crisis that forces a new key. In raaga terms, this is the pakad - the signature phrase that identifies which parent form you are in. We're starting to see these recur across films.

The six candidate arc shapes we've provisionally identified from the data: The Sawtooth (constant escalating peaks, Thriller/Action), The Double-Bounce (two major peaks with deep valley, Horror), The U-Shape (low middle, high bookends, RomCom/Drama), The Step Ladder (gradual increase with plateaus at revelations, Mystery/Procedural), The Plateau-to-Peak (long slow build, single massive climax, Epic/Western), The Chaotic (unpredictable spikes, Black Comedy/Satire).

These are provisional. The sample skews Drama-heavy (129 of 240+ titles) so we're expanding genre coverage specifically to stress-test whether these hold across Horror, Thriller, Sci-Fi, and Comedy at larger sample sizes.

Open questions I'm wrestling with

Intensity works well for describing the shape of an individual film. The problem emerges when comparing across films. Working at beat level - structural inflection points rather than absolute scores - gives us signal that is actually comparable across the corpus.

The Reagan et al. paper found six shapes in ~2,000 prose works. We're finding more at beat level in ~240 films. I don't know yet whether that's methodological, medium-specific, or sample-size artifact. Interested in views on this.

Validation is unsolved. DTW clustering will find shapes, but cluster stability across different parameter choices is unclear. Has anyone worked on structural clustering of narrative time series and landed on a validation approach they trust?

The temporal context problem: the raaga system assigns raagas to times of day and seasons - same notes, different meaning in different contexts. The narrative equivalent may be format: the same arc structure lands differently in a 90-minute film vs. a 6-episode series. I don't know how to operationalise that yet.

If there is a discoverable formal grammar of narrative structure - the music theory of storytelling - it will matter enormously in a world where production is automated and compulsion is the only scarce resource. Interested in pushback, prior work in computational narratology, and whether anyone has cracked the validation problem for time-series shape clustering.