BoxOffice Brain

The Multiverse Projector

Pick a genre and a year. Gemini reads the scenario in plain language, and the chart below proves every word with the real crowd of comparable films. Flip through parallel universes across time and across genres, and the story updates as you explore. When a cohort is too thin to call, it says so instead of inventing a shape.

checking backendLooking for boxoffice serve on this origin.

What am I looking at?

The story is Gemini reading the real ClickHouse numbers in plain language. It is told to use only the figures it is given, so it narrates the evidence rather than inventing it. If Gemini is unreachable, a deterministic grounded summary stands in, labelled as such.
The chart is the same numbers made visible: the real crowd of comparable titles placed by rating, with the mean and one spread drawn as a band.
The multiverse rail flips the scenario across years and across sibling genres. Each flip re-projects a real cohort and re-narrates the story.
Ask the data is the SQL analytics path: a plain-English question becomes read-only ClickHouse SQL, grounded in retrieved rows or an honest abstain.
Under the hood holds the load-bearing proof: the exact SQL, the grounding trace, the transport, the guardrail audit. Always one click away, never faked.

Genre Genre

Year

1999 Year

Press Project to pull a real cohort from ClickHouse.

abstained

Not enough comparable titles to project

rating axis, 0 to 10

Ask the data

A plain-English question becomes read-only ClickHouse SQL. The agent runs it, grounds the answer in the rows it retrieved, and abstains when the data cannot support a claim.

Choose a question

How it works

genre + year → guarded read-only SQL → official ClickHouse MCP server → real cohort → Gemini reads it, the chart proves it