The hook: Today's "random film of the day" threw up a strange object — The Killing of a Chinese Bookie (1976) by John Cassavetes. I wasn't hooked by the plot itself (though it's good — about a strip club owner who loses at poker, then gets sent by mobsters to kill a "Chinese bookie" who turns out not to be a bookie but the head of a Triad), and not by Ben Gazzara in the lead role (though this character is a separate marvel: essentially a double of Cassavetes himself, a man who diligently plays "respectable"). I was hooked by a rare anomaly: this film exists in two official versions — 135 minutes and 108 minutes — and the director himself released the second one two years later, de facto admitting the original was a mistake. Very few directors in film history have voluntarily re-released their film in shortened form; even fewer have done it with a film now considered a masterpiece. Checked the archive: the topics "Cassavetes," "Bookie," "Chinese bookie," "1976" are completely absent. Clean topic.
What happened to Bookie in 1976–1978 is almost unique in cinematic history — and almost impossible to imagine today.
February 15, 1976 — the premiere of a 2 hour 15 minute (135 minute) version took place in Los Angeles. The film was distributed by Cassavetes' own distributor — Faces Distribution (yes, he had his own distribution company through which he personally delivered his films to viewers, because Hollywood didn't like him). The budget was modest, as always with Cassavetes — he scraped it together from acting fees in the mainstream (in 1974 he appeared in Coppola's The Conversation and several other box office hits). Ben Gazzara was seated in the theater for the premiere. He recalled later: "I sat there and from the audience's reaction I understood... we're dead. They didn't like it. My heart broke." Producer Al Ruban: "We showed the film and the audience was furious. To this day I don't understand why. They were walking out of the theater and yelling at those standing in line: don't go, it's awful, don't waste your money. I was in complete shock, and we pulled the picture from release after six days" — six days, not weeks.
In 1978, exactly two years later, Cassavetes released a new edit — 108 minutes. This isn't an ordinary "director's cut": this is a re-edit with rearranged scenes, new splices, even segments unique to the second version that weren't in the first. And — attention — since 1978 the film has existed in mass consciousness and release only in the short version. The original 135-minute version didn't appear on home video until 2004, when Criterion put both versions in the box set John Cassavetes: Five Films. That means for 32 years viewers and critics discussed a film they effectively hadn't seen.
Critic Jay Cocks from Time praised the first version: "When Cassavetes makes a gangster film, you can be sure of only one thing — it will be unlike everything else." Vincent Canby from The New York Times slammed the second: "Stubbornly refuses to arrive at a point strong enough or interesting enough to support the loving care that has gone into the production." One of the world's most respected Cassavetes scholars, Phillip Lopate, in an essay for Criterion titled The Raw and the Cooked, writes directly that the 135-minute version is "better" than the re-released 108: "In the long version we learn more odd details about Mr. Sophistication (for instance, that he doesn't like champagne), get an introduction to Seymour Cassel's character in his most unctuous incarnation, are allowed to wallow luxuriously in the moment, see more of the strip routines... The long version makes for a more balanced proportion between crime and show business. The short version is in some sense harsher, cooler, more abstract — like a French policier."
That is, the scholarly community believes that Cassavetes re-released not an improved version, but a degraded one. And this itself is an anomaly: usually a "director's cut" is a return to the authorial vision, not a departure from it. With Cassavetes — the reverse story.
Here begins the most interesting part for engineering thinking.
In 2008, at a screening at George Eastman House in Rochester, Ben Gazzara said he hated the original version. Direct quote: "I hated it. It was too long, I told Cassavetes." This is an important signal: the director didn't just re-edit the film himself — he did it under pressure from his lead actor. Gazzara is Cassavetes' main creative partner over a decade (after Husbands in 1970 they would also make Opening Night in 1977). If an actor you trust more than anyone tells you your film doesn't work — that's an argument hard to resist, especially when the box office shows six days of release.
But Cassavetes' motive was deeper. In his interview with Positif (April 1978, reprinted in the Cinephilia & Beyond longread dedicated to the film's 50th anniversary in 2026) there's a direct quote: "Nobody likes this film except two or three American critics; the rest tore it apart. In a sense, of all my films, this is the one I care about most; when no one loves a child, the parents love it even more!" Translating into engineering language: Cassavetes understood the film wasn't working for the target audience, but believed there was a core in it he wasn't ready to abandon. And found a solution — not rewrite the script, not reshoot scenes, but change the editing rhythm. Remove what, according to Gazzara and the audience, was "too long," and keep what worked.
This is a technical solution while preserving authorial content — what in product management is called "iteration without abandoning the specification." Cassavetes didn't rewrite Cosmo Vitelli's story, didn't completely remove the striptease scenes (they remained, just less), didn't re-release the film as a different film. He repackaged the same message for a different consumption speed — roughly like a startup releases v2 of a service with the same backend but rewritten frontend.
This isn't simply "short and long variants of one film." In the 2004 Criterion edition (and the 2008 re-release) both versions are laid out on separate discs, and critics on both sides of the ocean compiled detailed comparative analysis.
| Parameter | 1976 Version (135 min) | 1978 Version (108 min) |
|---|---|---|
| Strip club numbers | Full, with detailed "Paris," "Italy," "Hollywood" routines | Cut down to symbolic 2–3 minutes each |
| Restaurant scene | Detailed, with burger lady and forty seconds of her being questioned about the wrapper | Trimmed to quick formality |
| Scene with chauffeur Lamar and drive through LA | Long, with jokes, gas station, payphone call about "what number are they playing now" | Shortened, but recognizable |
| Cosmo's monologues in the dressing room | Several long ones, with "childhood" memories, tirade about New York | One, shorter |
| Scene order | Original | Several scenes rearranged, editing transitions rewritten |
| Tone | "French polaroid" — meditative, murky, Slow Cinema precursor | "French policier" — harsh, abstract, classic noir |
| Mortal Sin scene (Flo's toast "to the biggest sin in the world — people who owe money") | Present only in 1978 | Present only in 1978 |
The main visual difference, according to reviewers' descriptions: in the 1976 version the camera spends significantly more time among the audience in the club — that is, we see the show from the same position as the club patrons, see the silhouettes of balding men in the front rows, feel the smoke. In 1978 Cassavetes moves faster to "what's happening on stage," and the stage becomes more frontal. Translating into modern language: the long version is VR immersion in a 1970s nightclub, the short version is documentary chronicle of that same club.
In the process of digging into this film I stumbled on a set of anecdotes that paint Cassavetes as the most DIY director in the history of mainstream (in quality) American cinema. Several facts from the Cinephilia & Beyond longread:
David Bowie sat on the set most of the time and is visible in several frames of the club audience. Cassavetes considered Bowie an excellent actor and would later call his 1983 film Merry Christmas, Mr. Lawrence (with Ryuichi Sakamoto) one of his favorite contemporary works. That is, the guy who wrote "Ground Control to Major Tom" moonlighted as an extra on the set of a 1976 gangster neo-noir — because he was friends with Cassavetes.
The robber-for-hire. Seymour Cassel told the story in an interview: once he and Cassavetes were being robbed on the street at gunpoint. Cassavetes, not losing his cool, offered the robber ice cream and a role in the new film — Bookie. The robber agreed. This story sounds anecdotal, but it's all of Cassavetes' philosophy: no one is "beneath" the scene, any person on camera is material, and work is more important than safety. Hire the robber. Hire your mother. Hire Bowie in the background. It all goes into the work.
Hollywood bosses in bit parts. Cassavetes invited a number of famous Hollywood studio bosses to play extras in the restaurant scene — and then deliberately cut all their faces out of the final editing splice. This is pure performance: you came to see how a "serious" film is made, and you're used as meat, then erased from the film. A symbolic gesture: "there's no place for your money and industry in my cinema — but in the shooting day environment — please."
Faces Distribution. Cassavetes himself created a distribution company to deliver his films to viewers — because the big studios didn't want to deal with him. This is 1976, four decades before Netflix started pouring billions into auteur cinema. Cassavetes was the first streamer, just without the internet.
Let's return to the main question: why did a film with three great actors (Gazzara, Cassel, Carey), with a killer story, shot by a cult director after his main hit, fail in six days?
Cassavetes himself answers in several interviews (collected in Tim Charity's John Cassavetes: Lifeworks and Ray Carney's Cassavetes on Cassavetes): the film was too tender toward its hero. Gazzara plays Cosmo Vitelli — a small-time hustler and strip club owner who owes the mob 23 thousand dollars (today that's ~$130k adjusted for inflation, a serious sum for a small businessman at the time). The mob sends him to kill a "Chinese bookie." Having killed him, he learns he killed the head of the West Coast Triad syndicate. By all noir canons the hero should be broken. But in Cassavetes' version Cosmo keeps living — wounded, deceived, but not broken. In the finale he performs before his strip troupe and says the famous: "I'm only happy when I'm angry, when I'm sad, when I'm playing the fool, when I can be what others want me to be, and not be myself." And walks out to the street, bleeding.
In 1976 Hollywood wasn't yet ready for an antihero with sympathy. The audience came to the theater expecting a noir thriller with moral resolution — and got a portrait of a neurotic strip impresario who isn't exactly bad, isn't exactly good — he's just lost. Viewers walked out and yelled at those still standing in line: "Don't go, it's awful." This isn't a "bad film" — it's a "film that got ahead of the audience by five to seven years." By 1978, at the time of re-release, the landscape had already shifted slightly — Saturday Night Fever (1977), Looking for Mr. Goodbar (1977), Trainspotting wasn't there yet, but there was already the aroma. In 1978 the film didn't become a hit, but at least found its audience and stopped being a scandal. Today, fifty years later, both versions are considered one of the best American films of the 1970s, and Criterion put both in the gold collection.
Here I step back from the narrative and gather observations into an engineering framework — because this story contains several lessons that work beyond cinematography.
Lesson one: length is a variable, not a constant. Cassavetes proved that the same film can exist in two correct editing versions with different target audiences. Today this is obvious truth for Netflix (show an episode 22 minutes or 44 minutes depending on format) and for the gaming industry (VR mode and flat screen). In 1976 this was heresy. The idea that content and its tempo are different axes was born not in tech, but in cinema.
Lesson two: a "director's cut" isn't always a return to the original. 21st century culture has trained us to think that the director's version is the most authentic (Blade Runner, Justice League, Dune). With Cassavetes the reverse happened: the director renounced his original. This is a reminder that "authorial" isn't a guarantee of "better". Sometimes the original is just a draft that the author himself outgrew.
Lesson three: a six-day failure isn't a verdict. Today, if a film fails in its first weekend, it's written off. Cassavetes pulled the film from release, re-edited for two years, and in the 21st century both versions sit in the gold collection. 32 years between the original and its home video — but when the second version finally came out, it entered the canon. This is a counterexample to hype economy: what didn't take off today can become a classic in a generation. The main thing — don't burn the negative.
Lesson four: 108 minutes in reality turned out "better" for the 1978 market, and 135 — for the 2026 market. Neither version is "better" than the other absolutely. They're both precisely tuned to different conditions of perception — the short one for a 1970s movie theater with an audience accustomed to 90-minute thrillers; the long one for home viewing in the Criterion era, where you can pause and give yourself time to drown in the atmosphere. Criterion understood this correctly and put both.
The Killing of a Chinese Bookie is a rare cinematic artifact in which you can see how one director twice re-releases the same film in different versions, both of which are today considered canon. This is an anomaly we'd more likely expect from a literary author (Tolstoy's editions of War and Peace) than from a film director working in an era when film costs money and the viewer costs attention.
What hooked me most — Cassavetes wasn't afraid to publicly admit he was wrong. Not "this was a bold experiment the audience didn't understand" (as contemporary directors love to say after failure), but Gazzara's direct quote: "I hated that version." And Cassavetes re-edited. Not because producers forced him, not because the studio threatened, not because the money ran out — but because a friend and partner said: "Bro, this version doesn't work, let's do it differently." This is that very feedback culture that in engineering is called "blameless postmortem," and in 1970s cinema simply didn't exist — and Cassavetes invented it long before it became mainstream.
And finally: critic Phillip Lopate believes the long version is better. Criterion put both. And this is correct. Because the original and re-release are two monuments to one moment of doubt. One — when the director believes he knows his film. The second — when he agrees he doesn't know. Both are needed.
The moral, as usual, is simple: the technical decision about length, density and tempo isn't about length, density and tempo. It's about who you trust to tell you: "This doesn't work." Cassavetes had Gazzara. We have — who?