This is literally what each party said today, compressed to one voice, with receipts.
Compression, not parody. OnScript ingests what elected U.S. officials publicly say and distills each party's real talking points into one composite voice. The comedy, where there is any, is emergent: the source material is real, and we never editorialize. We are analysis of speech, not a substitute for it.
The data is the joke. When dozens of members converge on the same phrase within a day, that convergence is the story — the public output of private coordination. We track first appearances, plot adoption curves, and score how on-script each party's language runs, day over day.
Citation-backed. Every distilled talking point links to at least three real source statements — member, date, source. If a claim can't be cited, it doesn't ship.
A symmetric instrument. The identical pipeline, prompts, and thresholds run for both parties, audited nightly in public on the Methodology page. Asymmetric findings are allowed; an asymmetric instrument is not.
Three layers, in order: (1) a deterministic phrase engine that measures which exact phrases spread across members and when each first appeared; (2) an LLM voice that renders the day's measured clusters into one composite paragraph, permitted to use only the code-computed clusters and code-computed numbers; and (3) a deterministic verifier that blocks any line whose quotes or counts don't check out against the source.