Films are shown to recruited audiences long before release, and endings, characters and entire sequences are altered on the strength of what those rooms report. The practice is old, frequently criticized and still standard.
The screening measures reaction, not opinion
Cards collected afterwards ask blunt questions about whether a viewer would recommend the film and which moments lost them, rather than asking for critical judgment.
What the studio is buying is a sense of where attention drops and where confusion sets in, and those are things a room reveals whether or not anybody can articulate them.
Filmmakers who defend the process usually defend this part of it, since a scene that consistently loses an audience is a problem regardless of how it reads on the page.
Recruited audiences are not a neutral sample
Attendees are recruited to a demographic profile, and people who accept an invitation to an unnamed film differ from people who would buy a ticket to a specific one.
The sample is also small and local, and it watches a version without finished effects, music or color, all of which change how a film plays.
Reading a rough cut's scores as a prediction of a finished film's reception therefore involves more interpretation than the numbers suggest.
Endings are the cheapest thing to change
A different ending can often be reached through reshoots of a few days, while restructuring a middle act touches everything around it.
Because the ending is also what an exit survey captures most strongly, it draws a disproportionate share of the changes made after a screening.
This is why films with troubled previews so often arrive with an ending that feels detached from what preceded it.
Contracts determine who gets to decide
Final cut authority is a negotiated term, and most directors do not have it, so the studio can act on screening results over an objection.
Directors with sufficient standing negotiate the right to control the final version, and those films go through previews as information rather than instruction.
The same data therefore produces very different outcomes depending on a contract signed long before anybody watched anything.
Alternate versions now reach audiences later
Home releases and streaming windows have made it practical to publish an earlier cut, so versions that lost the argument sometimes surface years afterwards.
Audiences campaigning for an unreleased version have occasionally succeeded, which changed how studios think about what a discarded cut is worth.
That possibility has made the preview process more visible, since the existence of alternatives is now something fans expect to be told about.