Methods

What we’re measuring

After a disaster makes national news, we score each Core Democratic Official for public statements only, in a five-day observation window — 120 hours from when the story went national. We do not infer private beliefs or why someone stayed silent.

For each official–disaster review, we code three checks. Tallies count only yes (partial is documented but does not increment the tallies).

Disaster gate. If a Core Democratic Official does not mention the current disaster in the five-day observation window, Climate Connection and Mentioned Republican Obstruction of Climate Action are not assessed.

That gate decides who can score a climate or obstruction yes. Findings still report those rates as shares of all official–disaster reviews (1,261) — each review is an opportunity to name the disaster and make the link.

1. Disaster Mention — Did they publicly speak about this specific disaster inside the window? Public-safety or air-quality advisories tied to this plume or storm count as yes. Generic climate talk with no link to this disaster is no.
2. Climate Connection — Did they link this disaster to climate change, global warming, or the climate crisis? Clean-energy advocacy without a causation link is scored no.
3. Mentioned Republican Obstruction of Climate Action — Did they name Republican Obstruction of Climate Action in connection with this disaster (not disaster-response politics alone)?

Yes means an in-window statement about this disaster names Republican opposition to addressing climate change and connects it to this disaster — for example a blocked climate bill, IRA rollback, Paris withdrawal, or a specific GOP vote against emissions or clean-energy policy while discussing this disaster.

Partial (documented, not counted in tallies) means the speaker criticizes Republicans about the disaster, but not for blocking climate policy. These are not near-hits on yes. Typical cases:

• Blaming disaster-response failures or agency cuts alone (for example NWS, FEMA, or NOAA staffing)
• Disaster-aid fights that never name blocked climate policy
• Citing fossil-fuel money without naming a blocked climate bill, rule, or treaty

No means they did not name Republican Obstruction of Climate Action in the window.

To be coded “Yes” or “Partial,” the statement must be in the window and about the disaster in question. The opposition cited can be from any year — a 1990s vote named while discussing this flood counts; a 2019 floor speech that never mentions this disaster does not.

The ChatGPT verification found Sen. Ed Markey’s in-window X posts on the January 2025 Los Angeles fires (Palisades and Eaton). Obstruction went from 0 to 2.

How we pick disasters

From 2023 onward, every disaster that meets all four rules is included:

• Climate-linked. Wildfire, smoke, flash flooding, hurricane, extreme heat, or a comparable hazard with climate context on the record.
• National news. The story broke through nationally — not just local or regional coverage.
• Clear start date. We record when it became a national story. The five-day observation window begins there. Each disaster page explains why we picked that date.
• Distinct national moment. One nationally saturated episode — not the same news cycle twice, and not one storm split into arbitrary pieces.

If you lived through a disaster that is not here, that is not because it did not hurt enough. The four rules are not a death or dollar test. Local floods, regional wildfires, and storms that never became one national news cycle can still kill people and destroy communities. They stay out when they miss a rule, so the reviews we do publish stay comparable. We apply these four rules first, then we ask what Core Democratic Officials said about that event.

What keeps a second event off the list as its own record is the same news cycle, not overlapping days in the window.

Examples: the February 2024 Texas Panhandle outbreak is one wildfire record (Smokehouse Creek), not a second record for Windy Deuce. The June 2024 Ruidoso fires are one record for South Fork and Salt together. The April 2025 Mid-South / Ohio Valley floods are one flood record, not a second record for that week’s tornadoes.

Later 2023 Canadian smoke reloads are not separate records — they never became a second national moment. Palisades and Eaton (January 2025) had a distinct national day, so each is a separate record. Beryl, Francine, Helene, and Milton in 2024 each have their own record for the same reason.

Edge case: Hurricane Debby (August 2024) caused serious flooding. It is not in this project because it fails the clear-start-date test — the impacts unfolded across several states without one national-wire saturation point we could pre-register. Deaths and dollars are not how we decide. Debby was deadlier and costlier than Hurricane Francine, which is on the list. Leaving Debby out does not mean it was a minor storm.

A ChatGPT verification pass found four disasters that met these inclusion rules and had been missed in the first Cursor coding: the April 2023 Fort Lauderdale floods, Hurricane Lee (September 2023), Tropical Storm Ophelia (September 2023), and the June 2025 eastern U.S. heat dome. They are in the project now.

The five-day observation window

The window runs 120 hours from when the disaster became national news through the same clock time five days later. Messaging tallies use only in-window public statements. Remarks from outside the window may appear as context on a disaster page but never change the score. Each disaster page prints the window’s start and end timestamps.

Heat waves, fire seasons, and multi-week smoke episodes do not start on first ignition or the first bad air day. The window starts when the disaster becomes national news — for example when coverage saturates or NWS issues its highest-tier alerts. Each disaster page states why we picked that start date.

One published window end date was later corrected by one day so the span was a full 120 hours.

Core Democratic Officials

Tallies cover Core Democratic Officials in office on the disaster’s start date: all Democratic senators, House Democratic leadership, and the DNC chair. House rank-and-file, candidates, and other speakers are not in the tallies. The denominators for each disaster vary based on how many Democratic officials were in office on the first day of the five-day window. For example:

Senator counts include all in-office Senate Democrats plus Independents who caucus with Democrats (King, Sanders), not Sinema. Each disaster uses who held the seat on that start date.

Official–disaster reviews

Every Core Democratic Official gets one review per disaster — including silence. Findings pools those reviews across 24 disasters (1261 official–disaster reviews). That is not a count of published statements. Full tallies: Findings.

Before the September 2026 verification, Findings were 1,045 reviews, 39 disaster mentions, 11 climate connections, and 0 obstruction.

They are now 1,261, 49, 15, and 2 across 24 disasters. The overall pattern — and the conclusion — did not change.

Deaths and dollars

Messaging scores are fixed to the five-day observation window. Fatalities and economic damage are assessed separately in six-month and two-year passes, labeled on each disaster page. Each pass uses only sources available on or before its cutoff date — not later literature retrofitted into an earlier figure.

Until the six-month pass exists, we do not print headline death or dollar figures — early counts stay as “at least” in the summary. Headlines are not final until the two-year pass is complete. Smoke and air-quality events often use modeled premature mortality rather than confirmed on-scene death counts; each pass documents scope and sources on the disaster page.

How we code

Cursor assisted in developing the original coding rules and coded the first 20 disasters. First-pass scores still come from Cursor. A second pass runs in ChatGPT.

Both use the same codebook and procedures. Differences are resolved against the codebook before a record is closed.

That check is two AI passes, not two human coders. Each new disaster follows the same process.

Republican framing

The dated timeline on Republican Obstruction records Republican blocks and delays of federal climate action.

Disaster pages may also show how Republicans talked about that disaster in the same five-day observation window. Those quotes use a separate code set (for example blaming Canada or spreading disinformation). They are context on the disaster record — they do not enter Democratic Findings tallies.