# AI reviewer examined the first-pass interface

I reviewed the HTML, CSS, JavaScript, and pilot data on November 12, 2026. I also inspected the desktop screenshot supplied in the parent context. I could not open a separate browser tab because this child session could not access IAB, so I do not claim live interaction or mobile testing.

1. **P1. Readers need the matching record immediately after choosing an issue.** I found that `filteredPeople()` selects matching people, while `profile()` and `openProfile()` still display every record, the general summary, and total counts. A reader selecting mental health can therefore see an unrelated addiction summary first. Render a dated matching action on each filtered card, distinguish matching counts from all-record counts, and place matching records first in the dialog with a clear option to inspect the rest.

2. **P1. Readers need historical dates and action labels on the cards.** I found a 2013 LePage veto behind a card that provides no year, while current campaign statements occupy cards with the same visual treatment. Add each summarized action’s year and evidence type beside the summary, label undated statements explicitly, and explain that a historical action may not describe a current position. Preserve opposing actions under the same neutral topic label, but make the verb visible so an overdose-prevention tag cannot look like an endorsement.

3. **P1. The headline should describe newcomers to a chamber accurately.** I found three current House members under the headline “New to Congress.” Replace that headline with wording about incoming members or new faces in each chamber, and explain “new to the Senate” as “moving from the House” in the filter. Keep the pilot limitation close to the headline because the colored squares resemble an election-results chart, even though the current caption correctly limits them to this selected sample.

4. **P2. Readers need larger evidence text and easier controls.** I found source metadata and topic tags at 9–10 pixels, controls at 11–12 pixels, and several short buttons with very small click areas. Raise body and evidence text toward 14–16 pixels, keep supporting labels comfortably readable, and provide roughly 44-pixel touch targets for frequent actions. Do not claim a specific accessibility conformance failure solely from these font sizes; validate mobile overflow, zoom, keyboard focus, and dialog closure in the parent’s browser.

5. **P2. Analysts need stable links to members and complete source trails.** I found that filters persist in the URL, but opening a member does not add an identifier, and ledger links expose only the first supporting source. Add a shareable member or record parameter, display all sources supporting a ledger record, and provide a way to copy that record’s link. Source-backed review becomes harder when a reader cannot direct a colleague to the same evidence without repeating the navigation.

6. **P2. Readers need clear download scope and a working research-notes destination.** I confirmed that `methodology.html` returned HTTP 404 at review time, and the CSV omits source-access status even though the JSON preserves it. Create the notes page before delivery, label downloads as the entire pilot rather than the current filter, and include a source table or source metadata in the CSV package. Add a filtered export only if needed; the crucial requirement is that readers can tell what they downloaded and reproduce its provenance.

7. **P2. Clinicians need brief explanations of congressional terms at their point of use.** I found chamber labels and district numbers without an explanation of statewide Senate representation versus district-based House representation. Add a short expandable guide near the chamber filter and describe what a vote on a whole bill establishes. Keep technical research details in the methods section so novice readers can reach the people quickly.