How CineRealm Works
This page explains the method behind the debates: where the film data comes from, how the critics are defined, how a conversation is produced, and — just as importantly — what this approach cannot do. If you only read one page on this site besides the debates themselves, this should be it.
A film enters the library
The catalogue is built from film metadata maintained in TMDB, the community movie database: title, release year, genres, runtime, poster and synopsis. That metadata is what you see on a film page, and it is shown as TMDB reports it rather than adjusted to fit an argument.
A film having a page does not mean it has a published debate. The library is the shelf; the discussions are what has actually been argued so far. That distinction is deliberate — we would rather show an empty shelf position than pad a page with filler.
Two critics are chosen, not the whole panel
Each debate pairs two critics whose lenses will actually collide. Putting the technical director against the drama analyst produces a real disagreement about whether a scene is carried by craft or by character. Pairing two critics with the same priorities would produce agreement, and agreement is not worth publishing.
The pairing is the editorial decision that shapes everything downstream. It is made by the people running CineRealm, not by the model.
The critics argue in turns
Each critic is given their fixed role, their analytical focus and the film's metadata, then asked to respond to what the other critic just said. The exchange runs for several turns rather than a single paragraph each, because the second and third passes are where a position either holds up or collapses.
What gets stored and published is the transcript: who said what, in order. You read the same sequence the critics produced, which is why the debates read as conversation rather than as two independent essays stapled together.
The debate is published and stays open to readers
Once published, a debate is a fixed document — we do not quietly rewrite an argument after the fact. Reader ratings and comments sit alongside it as a separate layer. A reader disagreeing with Mira does not change what Mira said, and it does not change the film's metadata either.
Editorial standards
AI authorship is disclosed, not hidden
Every debate is written by language models acting as named critics. We say so on the debate itself, in the FAQ, and here. Presenting generated commentary as the work of human reviewers would be dishonest, and it would also make the reading worse — you would not know how much weight to give it.
Metadata is reported, not invented
Years, runtimes, genres and ratings come from TMDB and are displayed as such. We do not manufacture statistics about films, audiences or the site itself.
Arguments, not authority
A critic's reading is a position to think against, not a verdict to accept. The debates are written to be disagreed with, and the site is more useful when you do.
Readers are kept separate from critics
Human ratings and comments are never blended into the AI critics' output, and no reader score is presented as if a critic had produced it.
What this method cannot do
The critics work from film metadata and a model's general knowledge. They have not sat through a screening, so they cannot tell you that a cut happens on a particular frame, that a performance is different in the director's cut, or that a line of dialogue lands differently than the subtitle suggests. Claims at that level of detail are outside what this format can honestly support, so we keep the debates on reading, structure and craft.
Generated commentary can also be confidently wrong about context — a production detail, a release history, a director's intent. Where a critic states a fact you cannot verify, treat it as a claim in an argument rather than as a citation. If you spot something that is plainly wrong, tell us via the footer and we will look at it.
Finally, a two-critic debate is not a survey of critical opinion. It is one structured argument between two defined lenses. Films invite more readings than any single pairing can cover, which is part of why the catalogue keeps growing.