# AI methods and congressional reporting review

I reviewed the first-pass data and interface on November 12, 2026, and independently checked selected records against their linked sources. I recommend the following revisions before treating the pilot as ready for wider review.

1. **Correct the proclamation publication date.** Codex assigns August 31, 2024, to `cooper-overdose.published_date`, but the governor’s office lists August 19, 2024, under First Published and Last Updated. Keep August 15 as the signature date and August 31 as the observance date. I opened the full text successfully, so the access label can also change to reflect that review. [North Carolina proclamation](https://governor.nc.gov/governor-proclaims-overdose-awareness-day-2024)

2. **Distinguish an action date from the date someone reported it.** Codex uses the alumni profile’s publication date as Colom’s `event_date` and the Assembly-passage press-release date as Wiener’s authorship date. The linked sources do not establish those as the dates when the underlying actions began. Add a `date_basis` field, or rename the field to `record_date` and display “Reported on” where appropriate. I confirmed that the Wiener page reports Assembly passage on June 30, 2022, while identifying him as the author. [Wiener’s office](https://sd11.senate.ca.gov/news/senator-wieners-safe-consumption-sites-legislation-passes-assembly)

3. **Show source access limitations alongside each summary.** Readers encounter confident member summaries before they see the distinction between full-page review and search-index text. Add a compact access label to the ledger and profile records, and keep a separate validation status for each claim. I independently confirmed the three Yea votes on S. 331 through the House Clerk and Hamawy’s veterans’ mental health statement through his survey, so those need no factual correction. The Hamawy access label can become full-text review. [House Clerk](https://clerk.house.gov/Votes/2025166), [Hamawy survey](https://ballotpedia.org/Adam_Hamawy)

4. **Make the newcomer definition visible in the headline and election metadata.** Readers could understand “New to Congress” as describing all 12 people, although Barr, Letlow, and Pappas already serve in Congress. Use “Incoming members” or “New to their chamber” for the headline, retain the entrant filter, and record the election-call date separately from the November 12 check date. Preserve the honest distinction between a reported winner and certification that nobody has checked.

5. **Make filtered summaries match the selected issue.** The current member cards retain their general summaries and all issue tags after a reader selects an issue. For example, an overdose-prevention search returns Cooper’s Medicaid-and-funding summary rather than his overdose record. Display a matching claim or explicitly state how many records match, then distinguish those records when the reader opens the member. Keep vetoes and support together only when their action verbs remain visible.

6. **Export enough information to reconstruct the review.** The CSV omits `person_id`, district, source IDs, source access labels, and source publication dates. Those omissions make stable joins and access-quality checks harder, and the CSV excludes Rabb without an explicit member manifest. Export separate member and source CSV files alongside the evidence table, add stable identifiers and date basis, and document that one member can have several records. Retain the full JSON as the authoritative package.

7. **Provide the promised research notes and an auditable search boundary.** The first-pass page links to `methodology.html`, which did not exist when I reviewed it. Add the notes with the selection rule, source hierarchy, issue definitions, and the specific remaining searches. Before expanding to national comparisons, maintain a person-by-issue search log with sources searched and completion status. A general “partial review” label cannot distinguish an unsearched topic from a completed search that found no usable record.

I could open the Maine legislative PDF, but a subsequent text-location request timed out before I independently located the veto passage. Keep its existing search-index access limitation until someone checks the cited passage directly. The pilot already avoids assigning scores and generally distinguishes historical actions from current positions; the revisions above strengthen that approach without inventing missing views.