Disclosed · 披露了

Members of the U.S. House must disclose their stock trades, but "public" doesn't mean usable. In 2026 alone there are 446 filings with 9,205 trades, spread across PDFs. 56 of them are scanned paper forms with stamps, hand-ticked boxes and no ticker symbols. Amounts are ranges, not share counts. And there was no tool at all for Traditional-Chinese readers. Disclosed (披露了) is a free, non-commercial, bilingual (繁體中文 / English) site that makes these records readable. Meta's Muse Spark model reads every scanned form (56 of 56, zero failures) and independently re-parses every text PDF. It agrees with our rule-based parser on 99.44% of rows, and the disagreements exposed a real parser bug. A three-role "MAGI" review then checks the data: a verifier, a risk reviewer and an explainer, all Meta-model roles. Each votes, and plain code (not AI) turns the votes into a confidence badge shown on 1,790 reviewed rows. On top of the data: - A Congress Consensus board syncs each member's buys and sells and estimates net amounts from the disclosed ranges. - Daily picks come in two switchable methods, with an "In both" badge for overlaps. - Every row links to its original PDF. We're strict about honesty. Live picks are frozen into an append-only, hash-chained log that can't be back-dated, and backtests are labelled reconstructed. Weak results are shown: the average disclosed purchase did not beat the S&P 500. Not investment advice. Live site: https://jo-peace.github.io/disclosed/