Background

The Supreme Court and Congress publish some of the most consequential documents in the country — opinions, committee reports, and Congressional Research Service analyses. Almost nobody reads them. They run to hundreds of pages, they live as PDFs on government websites, and most people end up relying entirely on what a commentator told them the document said.

True Record turns those documents into narrated audio, read in full and in their own words, so people who care what is in them can listen on a commute or a walk instead. No panel. No commentary. No spin. The document, start to finish.

The Challenge

• Documents written for the page, not the ear. Official text is full of print boilerplate, citation abbreviations, footnote rules, and paragraphs broken mid-sentence across pages. Read aloud as-is, it is unlistenable.
• Structure is meaning. A Supreme Court opinion is a syllabus, a majority, and a set of concurrences and dissents by different justices. Flatten that into one stream of audio and the listener loses who is saying what.
• Neutral means neutral. The whole promise is the document itself. Nothing can be summarized, editorialized, or trimmed without the listener knowing.
• Paid audio has to stay paid. Narrations are the product, so the audio files cannot simply sit on a public URL.

Our Approach

A document pipeline, not a text-to-speech button. Source documents are pulled from public government and court-record sources, then cleaned for the voice: boilerplate stripped, every abbreviation the voice cannot say spelled out, and broken paragraphs rejoined. Across the fifteen documents processed so far, that took 2,351 occurrences of 272 kinds of unspeakable abbreviation down to zero.

A quality gate that says no. The pipeline refuses to approve a document while any unspeakable abbreviation or mid-sentence break remains. A person then approves each document's production cost and listens before release — the first three minutes and a random minute from the middle, bound to that exact version.

Structure preserved. Each part of an opinion — the syllabus, the majority, every concurrence and dissent — becomes its own attributed chapter, and a read-along transcript highlights the sentence being spoken. Narration is synthesized from the official published text, and the source document is linked from every narration.

Generated once, served to everyone. Each narration is produced a single time and stored on infrastructure with no egress fees, so an additional listener costs next to nothing. Paid audio is protected by cryptographically signed passes that expire within a day, checked at the edge before a single byte is served.

Results

• Submitted to the App Store on September 25, 2026 — 16 days after the first commit — with its subscriptions, following a public TestFlight beta.
• Reviewed before it shipped — a dedicated security review and a usability review ran ahead of submission.
• Built with Claude end to end — all 329 commits were co-authored with Claude, across the app, the document pipeline, and the audio server.
• Private by design — no advertising, no analytics, and no tracking of any kind.

Why It Matters

True Record is document intelligence pointed at the public instead of a single firm. The hard part is the same one our legal and construction work solves every day: take long, messy, high-stakes documents and make them usable without losing a word of what they actually say.