A wooden card index drawer pulled open, packed with blank cards

INSIDE THE BOOK

Every chapter with the idea it carries, every measurement defined, and the list of things the book says it cannot establish.

Four parts, 42 chapters and three appendices over 366 pages. The line under each title is what that chapter establishes, not a teaser for it.

Part 1

THE OUTSIDE OF A FILM

A century of records, and how much of what everyone repeats survives contact with them

1
Zero Point Zero Six Eight
The number this book was built to find, and what it turned out to be
2
Five Hundred Votes
What is actually in the catalogue, and what the catalogue is not
3
One Film in Six and a Half
The denominator under every claim anyone has ever made about film money
4
The Barbell and the Silence
The best-known finding in film economics, and the correction that turns it upside down
5
Money Buys Reach, Not Respect
What a production budget purchases, and the one thing it reliably does not
6
Horror Pays and Prestige Doesn't
Genre as a financing instrument rather than a description of content
7
Ninety-Seven Minutes for Sixty Years
A claim everyone repeats, false for the entire period they were repeating it
8
They Don't Make Them Like They Used To
The most seductive artifact in the catalogue, and why the correction makes it worse
9
Someone in This Relationship Is Wrong
Two crowds, and the decade their disagreement changed sign
10
The Only Thing That Reliably Pays
The belief this catalogue most decisively overturns
11
A Dollar Is Not a Unit
Not one line of code in this system deflates a dollar, and nor does anyone else's

Part 2

THE INSTRUMENTS

What a score in this book is made of, and what it measures when nobody is checking

12
The Recipe
The master score, taken apart, exactly
13
A Rank of Who Turned Up
The popularity component is not a rating, it is a headcount
14
Two Ingredients
What most films in this catalogue are actually scored on
15
Twenty Columns, Five Things
How many independent measurements twenty features actually contain

Part 3

THE INSIDE OF A SCREENPLAY

Fifteen hundred screenplays, twenty measurements, and the answer this project was built to find

16
Fifteen Hundred and Seventy-One Screenplays
A reasonable thing to attempt, and how it was attempted
17
Nobody Read These
The grades were never craft judgements, and what that does to the question
18
Twenty Measurements and No Signal
The central result, stated plainly and then attacked four ways
19
Seventy-Four Percent Had No Dialogue
The instrument was broken, and here is exactly how
20
Three Findings, Dismantled
Everything this project published as a positive result
21
We Fixed It and Got the Same Answer
The chapter that turns a broken study into a finding
22
Budget Beats Everything You Can Read
What the text adds over what the greenlight memo already knows
23
Everything Else Anybody Knows
The director, the writer, the studio and the distributor, measured the same way
24
The Half-Life of an Honest Number
What an organisation does when it finds its own bug

Part 4

THE CASES

Thirteen films taken apart one at a time, because an aggregate tells you what is true and never what it means

25
How to Read a Case
The method, stated once, so the eight chapters after it can get on with it
26
The Film That Breaks Both Instruments
The Shawshank Redemption, measured twice, wrong twice, in opposite directions
27
The Streaming Zero
The largest failure in the catalogue is not a failure
28
Eighty-Seven Times the Line
The other tail, which is forty years older and exactly as misleading
29
The Cheapest Things That Ever Worked
One finding that is completely real, and one number that should never have been printed
30
Three Wins and No Money
CODA, and what an exception to a strong finding is usually made of
31
Two Crowds, Three Films
Where the two audiences disagree most, and what a famous flop's score is made of
32
The Film Nobody Wrote About
The middle of the catalogue, which is what the catalogue is mostly made of
33
Two Hundred and Thirty-Two Pages
A second bug, and the one case that is still wrong after the fix

Part 5

THE ADVICE

A second corpus, of nine hundred books about how to do it

34
Nine Hundred and Twenty-Six Style Guides
The other corpus, and why it is worth measuring
35
Where the Advice Disagrees With Itself
Act one ends on page thirty. Act one ends on page twenty-five.
36
What It Talks About, and What It Does Not
Twelve topics, four crafts, and some conspicuous gaps
37
The Tells
What the text says about how the text was made
38
What Could Actually Be Reprinted
Two hundred words of somebody else's writing, and where they are

Part 6

THE RECORD

The tables, each printed with the filter that produced it

39
How to Read the Record
The cleaning rules, stated once, so every table can be checked
40
The Decade Atlas
Eleven decades on one template, so the differences are differences
41
The Genre Atlas
Every genre counted twice, because there are two defensible ways
42
The Lists
Four rankings, each with its gate printed

APPENDICES

Appendix A. Method
Where every number came from, and what it is allowed to say
Appendix B. The Twenty Features
What was measured on each screenplay, exactly
Appendix C. Sources, Licences and What Is Not Reproduced
Where the data came from, and the line this book does not cross

A thick book lying open, both pages blank cream stock, lit from one side

Two hundred pages, six by nine inches, black on cream. About a thousand rows of it are tables that need paper to be usable.

The object

THE TWENTY MEASUREMENTS, DEFINED

These are the entire empirical content of Part Two, so they are printed here in full, including the ones whose definition does not survive close reading. Three things are worth noticing on the way down: several measure formatting rather than storytelling, several are near-duplicates of each other, and every one of them is computed from a document rather than a film — so none can see casting, direction, editing, music, marketing or release date.

Precision measuring instruments laid out in a row on dark felt — Twenty instruments, each one exact about something, and none of them able to see a film.

01Scene countscene_count · scenes: The number of scene headings detected. A scene heading is a line the parser reads as a slugline, so this counts formatting, not narrative units.
02Average scene lengthavg_scene_length · lines: Mean of action lines plus dialogue lines per scene. Measured in lines rather than pages or seconds.
03Interior to exterior ratioint_ext_ratio · ratio: Interior scenes divided by exterior scenes, from the INT. and EXT. markers in the sluglines. A screenplay with no exterior scenes divides by one rather than by zero, so the value is the interior count.
04Scene length variancescene_length_variance · lines squared: Sample variance of the scene lengths above. High values mean a mix of very long and very short scenes.
05Total pagestotal_pages · pages: Total lines divided by fifty-six, the standard lines per screenplay page. An estimate from the text, not a page count from a PDF.
06Transition densitytransition_density · per page: Count of lines beginning FADE IN, FADE OUT, FADE TO, CUT TO, DISSOLVE, SMASH CUT or MATCH CUT, divided by the estimated page count.
07Distinct speaking charactersunique_character_count · characters: The number of distinct character cues the parser found. A character who appears under two spellings counts twice; one who never speaks does not count at all.
08Dialogue ratiodialogue_ratio · share of lines: Dialogue lines divided by total lines. This is the feature that read exactly zero for three quarters of the corpus before the parser was repaired, and chapter [!ch:seventy-four-percent] is what that did.
09Top three character dominancetop3_character_dominance · share: Words spoken by the three largest parts, divided by all spoken words. A two-hander scores near one; an ensemble scores low.
10Average speech lengthavg_dialogue_length · words: Spoken words divided by the number of dialogue blocks. The average length of one uninterrupted speech.
11Character introduction ratecharacter_intro_rate · per page: Distinct speaking characters divided by the estimated page count. Named as a rate, computed as a density: it does not measure when characters arrive, only how many there are per page.
12Vocabulary richnessvocabulary_richness · ratio: Distinct words divided by total words, over the first five thousand words. A type-token ratio, sampled at a fixed length because the raw ratio falls as a document gets longer and would otherwise measure length.
13Average word lengthavg_word_length · characters: Mean length in characters of every alphabetic word in the document, action and dialogue together.
14Sentiment, meansentiment_mean · score: Mean VADER compound sentiment over the dialogue blocks, on a scale from minus one to plus one. VADER is a lexicon tuned on social media text, and applying it to screen dialogue is an assumption rather than a measurement.
15Sentiment, variancesentiment_variance · score squared: Sample variance of the same block scores. Read as emotional range.
16Sentiment arc slopesentiment_arc_slope · score per tenth: The document is cut into ten equal chunks, each chunk scored, and a straight line fitted across them. Positive means the dialogue gets more positive from start to end. A straight line through ten points is a crude summary of an emotional arc and is treated as one.
17Action ratioaction_ratio · share of lines: Non-blank lines that are not dialogue, divided by all non-blank lines. Roughly the inverse of the dialogue ratio, and it fails in the same way when the parser fails.
18Capitals densitycaps_density · share of words: Words of two or more capital letters, divided by total words. Intended to capture emphasis and sound cues; it also captures character cues and sluglines, so it is partly a measure of formatting.
19Exclamation densityexclamation_density · share of words: Exclamation marks divided by total words, across the whole document.
20Question densityquestion_density · share of words: Question marks divided by total words, across the whole document.

WHAT IT CANNOT ESTABLISH

Every figure in the fact base carries a caveat field, and these are the caveats attached to the claims the book leans on hardest. They are not a disclaimer. They are the reason to believe the rest: a project that will tell you where its own numbers are weak is one whose strong numbers mean something.

A wide flat drawer standing mostly empty, a few objects at one end

83,606

Of: 158,178 rows in the master file, of which these fell below the gate
How: count of rows with fewer than 500 IMDb votes in hollywood_analysis.csv; the build script keeps only rows at or above 500
But: This single filter is the most consequential decision in the whole dataset. It removes the majority of the source rows and it removes them unevenly.

15.46%

Of: 74,571 films in the catalogue
How: share carrying both a reported budget and a reported worldwide gross
But: Every profitability statement in this book, and in every book like it, describes this fraction of the catalogue and not the catalogue. The films that never reported are not a random sample of the ones that did.

R-squared = 0.92

Of: 1,229 graded screenplays carrying a grade and the reception measures
How: R-squared of a linear fit predicting the screenplay grade from the finished film's reception: imdb_rating, rt_tomatometer, oscar_wins
But: The grade on these screenplays is almost entirely recoverable from how the finished film was received. Nobody read them. Correlating the text against the grade is therefore partly correlating the text against the film, which is not the same question.

12.7%

Of: 1,289 films used to fit the shipped box office model
How: the R-squared reported in the shipped model file
But: This model reads twenty features of the screenplay text. It explains less than a linear fit on the budget alone.

11 rows

Of: 11,427 films with a reported budget of at least $10,000 and a reported worldwide gross of at least $1,000, which is 11,427 of the 74,571 in the catalogue, grouped by release decade
How: Each film's budget and gross deflated individually by the CPI-U annual average for its release year, rebased to 2025, then the median taken. Author's adjustment; the index is printed in craft.table.cpi_index.
But: The real columns are the author's external assumption, not a repository figure, and nothing in the repository is inflation adjusted. They are also unsafe before about 1970, where a lifetime gross accumulated across re-releases is being deflated at the price level of the release year. The comparison the table supports is the recent one: in real terms the median reporting film's budget peaked in the 1990s and has fallen by more than half since.

60 of 107 years

Of: 107 the 107 release years, in how many the highest rated film changes if the 20,000-vote floor is removed
How: Compared the per-year winner over all films against the per-year winner over films with at least 20,000 votes.
But: The unfiltered winners include a 2026 release rated 10.0 on 512 votes. This number is the argument for the floor, and it is why no "best film of the year" claim in this book is made without one.

n/a

Of: 0 the 926 craft guides and their front matter
How: Front matter keys inspected; no author or date field exists.

n/a

Of: 0 the 926 craft guides and the third-party screenplay text inside them
How: Deliberately not computed.

All of that is free and always will be. The paperback is the argument in order, plus about a thousand rows of reference tables that need paper to be usable.

The paperback, direct — 19.99 USD

CONTENTS

Four parts, 42 chapters and three appendices over 366 pages. The line under each title is what that chapter establishes, not a teaser for it.

Part 1

THE OUTSIDE OF A FILM

A century of records, and how much of what everyone repeats survives contact with them

1
Zero Point Zero Six Eight
The number this book was built to find, and what it turned out to be
2
Five Hundred Votes
What is actually in the catalogue, and what the catalogue is not
3
One Film in Six and a Half
The denominator under every claim anyone has ever made about film money
4
The Barbell and the Silence
The best-known finding in film economics, and the correction that turns it upside down
5
Money Buys Reach, Not Respect
What a production budget purchases, and the one thing it reliably does not
6
Horror Pays and Prestige Doesn't
Genre as a financing instrument rather than a description of content
7
Ninety-Seven Minutes for Sixty Years
A claim everyone repeats, false for the entire period they were repeating it
8
They Don't Make Them Like They Used To
The most seductive artifact in the catalogue, and why the correction makes it worse
9
Someone in This Relationship Is Wrong
Two crowds, and the decade their disagreement changed sign
10
The Only Thing That Reliably Pays
The belief this catalogue most decisively overturns
11
A Dollar Is Not a Unit
Not one line of code in this system deflates a dollar, and nor does anyone else's

Part 2

THE INSTRUMENTS

What a score in this book is made of, and what it measures when nobody is checking

12
The Recipe
The master score, taken apart, exactly
13
A Rank of Who Turned Up
The popularity component is not a rating, it is a headcount
14
Two Ingredients
What most films in this catalogue are actually scored on
15
Twenty Columns, Five Things
How many independent measurements twenty features actually contain

Part 3

THE INSIDE OF A SCREENPLAY

Fifteen hundred screenplays, twenty measurements, and the answer this project was built to find

16
Fifteen Hundred and Seventy-One Screenplays
A reasonable thing to attempt, and how it was attempted
17
Nobody Read These
The grades were never craft judgements, and what that does to the question
18
Twenty Measurements and No Signal
The central result, stated plainly and then attacked four ways
19
Seventy-Four Percent Had No Dialogue
The instrument was broken, and here is exactly how
20
Three Findings, Dismantled
Everything this project published as a positive result
21
We Fixed It and Got the Same Answer
The chapter that turns a broken study into a finding
22
Budget Beats Everything You Can Read
What the text adds over what the greenlight memo already knows
23
Everything Else Anybody Knows
The director, the writer, the studio and the distributor, measured the same way
24
The Half-Life of an Honest Number
What an organisation does when it finds its own bug

Part 4

THE CASES

Thirteen films taken apart one at a time, because an aggregate tells you what is true and never what it means

25
How to Read a Case
The method, stated once, so the eight chapters after it can get on with it
26
The Film That Breaks Both Instruments
The Shawshank Redemption, measured twice, wrong twice, in opposite directions
27
The Streaming Zero
The largest failure in the catalogue is not a failure
28
Eighty-Seven Times the Line
The other tail, which is forty years older and exactly as misleading
29
The Cheapest Things That Ever Worked
One finding that is completely real, and one number that should never have been printed
30
Three Wins and No Money
CODA, and what an exception to a strong finding is usually made of
31
Two Crowds, Three Films
Where the two audiences disagree most, and what a famous flop's score is made of
32
The Film Nobody Wrote About
The middle of the catalogue, which is what the catalogue is mostly made of
33
Two Hundred and Thirty-Two Pages
A second bug, and the one case that is still wrong after the fix

Part 5

THE ADVICE

A second corpus, of nine hundred books about how to do it

34
Nine Hundred and Twenty-Six Style Guides
The other corpus, and why it is worth measuring
35
Where the Advice Disagrees With Itself
Act one ends on page thirty. Act one ends on page twenty-five.
36
What It Talks About, and What It Does Not
Twelve topics, four crafts, and some conspicuous gaps
37
The Tells
What the text says about how the text was made
38
What Could Actually Be Reprinted
Two hundred words of somebody else's writing, and where they are

Part 6

THE RECORD

The tables, each printed with the filter that produced it

39
How to Read the Record
The cleaning rules, stated once, so every table can be checked
40
The Decade Atlas
Eleven decades on one template, so the differences are differences
41
The Genre Atlas
Every genre counted twice, because there are two defensible ways
42
The Lists
Four rankings, each with its gate printed

APPENDICES

Appendix A. Method
Where every number came from, and what it is allowed to say
Appendix B. The Twenty Features
What was measured on each screenplay, exactly
Appendix C. Sources, Licences and What Is Not Reproduced
Where the data came from, and the line this book does not cross

THE TWENTY MEASUREMENTS, DEFINED

WHAT IT CANNOT ESTABLISH

INSIDE THE BOOK

CONTENTS

Zero Point Zero Six Eight

Five Hundred Votes

One Film in Six and a Half

The Barbell and the Silence

Money Buys Reach, Not Respect

Horror Pays and Prestige Doesn't

Ninety-Seven Minutes for Sixty Years

They Don't Make Them Like They Used To

Someone in This Relationship Is Wrong

The Only Thing That Reliably Pays

A Dollar Is Not a Unit

The Recipe

A Rank of Who Turned Up

Two Ingredients

Twenty Columns, Five Things

Fifteen Hundred and Seventy-One Screenplays

Nobody Read These

Twenty Measurements and No Signal

Seventy-Four Percent Had No Dialogue

Three Findings, Dismantled

We Fixed It and Got the Same Answer

Budget Beats Everything You Can Read

Everything Else Anybody Knows

The Half-Life of an Honest Number

How to Read a Case

The Film That Breaks Both Instruments

The Streaming Zero

Eighty-Seven Times the Line

The Cheapest Things That Ever Worked

Three Wins and No Money

Two Crowds, Three Films

The Film Nobody Wrote About

Two Hundred and Thirty-Two Pages

Nine Hundred and Twenty-Six Style Guides

Where the Advice Disagrees With Itself

What It Talks About, and What It Does Not

The Tells

What Could Actually Be Reprinted

How to Read the Record

The Decade Atlas

The Genre Atlas

The Lists

Appendix A. Method

Appendix B. The Twenty Features

Appendix C. Sources, Licences and What Is Not Reproduced

THE TWENTY MEASUREMENTS, DEFINED

WHAT IT CANNOT ESTABLISH

INSIDE THE BOOK

CONTENTS

Zero Point Zero Six Eight

Five Hundred Votes

One Film in Six and a Half

The Barbell and the Silence

Money Buys Reach, Not Respect

Horror Pays and Prestige Doesn't

Ninety-Seven Minutes for Sixty Years

They Don't Make Them Like They Used To

Someone in This Relationship Is Wrong

The Only Thing That Reliably Pays

A Dollar Is Not a Unit

The Recipe

A Rank of Who Turned Up

Two Ingredients

Twenty Columns, Five Things

Fifteen Hundred and Seventy-One Screenplays

Nobody Read These

Twenty Measurements and No Signal

Seventy-Four Percent Had No Dialogue

Three Findings, Dismantled

We Fixed It and Got the Same Answer

Budget Beats Everything You Can Read

Everything Else Anybody Knows

The Half-Life of an Honest Number

How to Read a Case

The Film That Breaks Both Instruments

The Streaming Zero

Eighty-Seven Times the Line

The Cheapest Things That Ever Worked