RESEARCH PROTOCOL · NOVEMBER 12, 2026

How we searched and classified the records.

Codex expanded a 12-person interface pilot into a national roster of identified newcomers after the 2026 midterms. AI assistants completed the research and review. Wayne has not validated every entry, and readers should inspect sources before relying on a claim.

We distinguish a race call from certification.

We used the new-member roster as a starting point, then searched for individual election reporting. We added Kevin Hern and Sherrod Brown after checking AP’s Senate chronology. We preserve disagreements among news organizations and keep unresolved reporting out of the default winner view. We have not independently checked certification.

We include first-time members, current House members entering the Senate, and former members returning after a break. We identify nonvoting delegates separately. We exclude incumbents returning to the same chamber. Remaining race calls and Louisiana runoffs may add newcomers after this snapshot. The unresolved profiles do not enumerate every remaining candidate.

We searched the same topic families for every person.

We combined each member’s name and state with the topic families below. We searched publicly indexed material available on November 12, 2026, without a lower date cutoff. We retained the exact queries, including follow-up searches and failed attempts, rather than claiming that every query used identical punctuation.

1. We searched addiction, opioids, fentanyl, treatment, overdose, naloxone, and harm reduction.
2. We searched Medicaid, related mental health, and recovery supports.
3. We searched methadone, buprenorphine, alcohol, stimulants, drug sentencing, and diversion.
4. We separately searched for each person’s November 2026 election result.

We screened results for attribution and followed relevant leads to official legislative pages, government releases, campaign platforms, candidate surveys, and reporting. Search engines sometimes returned another person with the same name or another candidate on a shared page. We excluded claims that we could not attribute to the correct person.

We recorded whether we reviewed a full page or only indexed text. Search completion means we executed and screened the documented queries. It does not mean we audited every roll call, archived interview, deleted page, or unindexed document. Search ranking and source access can produce uneven coverage even under a common protocol.

We keep unknown positions visible.

We use “no position found in our search” when the scoped search did not locate an attributable statement or action. We use an incomplete-review label when ambiguity or unfinished primary-source checking prevents that conclusion. Neither label means opposition or neutrality. Issue filters retain people with gaps, while the source ledger lists only located evidence.

We use “record” in the issue matrix to indicate an available action or statement. We do not encode approval, opposition, political quality, or evidentiary strength with a numerical score. The record’s wording describes what the person did, and the source-access label describes our verification.

We separate addiction commitments from related healthcare.

A broad Medicaid promise does not establish support for a particular addiction treatment. We tag those records as coverage context. We classify related mental health separately and reserve addiction-specific tags for explicit substance-use policy. We exclude institutional roles alone from policy counts and retain relevant roles as background.

A vote documents the member’s action on the whole measure. Sponsorship documents support for a proposal, without establishing enactment. Executive actions include signatures and vetoes. Campaign statements and candidate surveys document the stated position; they do not independently verify claimed accomplishments or effects. We identify claims made by parties or advocacy organizations as such.

We preserve the date of the action when a source establishes it. When we know only the report date, we label it as reported and leave the action date unknown. Undated sources remain undated. Historical statements may not describe current views.

We added opioid prescribing as a supplementary related issue when source follow-up identified pain-treatment proposals. We did not conduct a separate pain-prescribing search for every person. We keep those records separate from medications that treat addiction.

We corrected the first pass after AI review.

Three AI reviewers examined the original pilot from policy, evidence, and interface perspectives. We corrected a publication date, clarified bill status and source access, separated broad mental health records, and added direct links to individual records. We then expanded the roster and moved source-linked summaries onto the profile cards. The review documents below describe the original pilot, not independent human validation of the expanded dataset.

• Read the AI policy review.
• Read the AI evidence review.
• Read the AI interface review.
• Read the expansion interface review.

We preserve the data behind the interface.

The complete JSON uses schema 2.0 and includes profiles, policy records, sources, query logs, issue-level review status, and portrait credits. We retain the historical filename for integration compatibility. The research tables include stable identifiers and source URLs so readers can inspect or reuse the evidence without scraping the cards.

We store licensed portraits with the site and link each image to its source and attribution. We use initials when we have not verified a reusable photograph. View all portrait credits.

We welcome source-based corrections. A useful correction identifies the member, the specific claim, a source URL, and the relevant date.