Methodology

Why this exists

More than a leaderboard of catastrophes

The project's founding note put it more ambitiously than a methodology page usually allows, so here it is close to verbatim:

This project aims to go beyond mere classification of disaster movies. We seek to distill meaningful lessons and insights from these films, bridging the gap between entertainment and education. By analyzing disaster narratives through the lenses of storytelling, science, and social impact, we will uncover valuable takeaways that can inform, inspire, and potentially save all of humankind.

What follows is the honest gap between that ambition and 831 rows of real data: two genre maps sketched out before a single film was logged, three rewritten attempts at getting a language model to classify against them, and the much narrower set of families, tallies, and decade breakdowns that actually made it into the chart.

Map one

The full disaster taxonomy

The genre's causes, mapped before a single film was logged — nine top-level branches, most split further into the specific hazard. The shipped site never tags a film against this full tree; instead, each branch below shows which of the eleven families in the live chart it was folded into, and why.

Turning a map into labels

Three passes at the same films

The raw DisasterType, MainCategory, and Subcategory fields didn't come from a single pass — they came from feeding film lists to a GPT model against a classification prompt, rewritten at least three times as the project's thinking changed. None of the three schemes below is what ships today; a given raw record reflects whichever pass happened to classify it, which is exactly the inconsistency the eleven-family reconciliation exists to untangle.

The relationship from planning document to live bar chart, in five steps:

Map two

Themes and elements

The second map isn't about what destroys the world — it's about what the film is actually doing with that destruction: its tone, its themes, the shape of its plot, who it follows and what they learn. Most of this framework stayed conceptual; TMDB doesn't carry fields for tone or character arc, and this project didn't hand-tag 831 films against them. Each branch says plainly whether — and how — it made it into the data.

A later, separate pass tried a different cut on the same question: instead of elements within a story, ten themes cross-mapped directly against the disaster categories most likely to carry them.

Also conceptual — no film in the dataset is tagged against these ten themes. Kept here because it's a genuinely different lens than the map above: themes tied to cause, rather than themes tied to story structure.

What actually shipped

From raw fields to eleven families

The real source data — not the planning maps above — labelled these films with 105 different free-text disaster types, 49 main categories and 191 subcategories, much of it inconsistent and some of it mangled by a bad comma-split. Those got consolidated into the eleven families in the live chart, under a three-tier matching pass: unambiguous named hazards first (Titanic's DisasterType reads "Natural Disaster," but only the iceberg fragment tells you what actually happens), then specific hand-written signals, then generic category labels only when nothing more concrete was on record.

Judgement calls were made throughout, and the rule was consistent: a film with a stated cause is filed under that cause rather than its setting, so Mad Max: Fury Road counts as Atomic and not as a post-apocalyptic film. Where the source contradicted itself, the more specific signal won. The full matching logic — every regex, in order, with the reasoning for why it's ordered that way — is public: scripts/taxonomy.mjs on GitHub ↗.

The plot-language breakdown on the main page runs the same honesty standard: it counts words in each film's own TMDB plot summary, not a hand-assigned verdict on how the story ends. Ratings, posters, and plot summaries come from TMDB; where the dataset carries box office, it's recorded for 34 of 831 films — the rest went unrecorded at the source.

The bigger ambition

The "what if" framework

The classification work was always meant to serve a larger idea: reimagine how a film's fictional catastrophe could have been prevented or mitigated, then translate that fictional fix into an actual strategy — technological innovation, social cooperation, environmental stewardship, psychological resilience — worth taking seriously off screen. The premise was to use the familiar territory of film to make daunting real-world risks more approachable, for policymakers and storytellers alike.

This is the part of the project that never left the page. Nothing here is computed or scored per film; no alternate ending has been drafted, let alone translated into a real-world strategy. It's the reason the site's own analysis leans toward what these 831 films reveal in aggregate — the plot-language tally, the shift in which countries make them, which decades got the darkest — rather than any single film's plot taken in isolation.