The Investigative Journal’s weekly fact-check goes to the original records — federal statistical releases, agency data files, court dockets and working papers — rather than to other fact-checkers. This week: the FBI director’s crime testimony, the Senate minority leader’s war-cost arithmetic, a presidential forecast on fuel prices, the White House manufacturing message, and an election-fraud model circulating in progressive online spaces.
Claim 1: Kash Patel says the FBI produced “the largest reduction in the murder rate in the history of America”
The claim. Testifying before the Senate Judiciary Committee on September 15, FBI Director Kash Patel told senators the bureau and its partners had delivered “the largest reduction in the murder rate in the history of America” and described the period as “the most prolific run of crime reduction in United States history.” He cited a 2025 murder rate of 4.1 killings per 100,000 people.
The evidence. The underlying numbers come from the FBI’s own Uniform Crime Reporting Program, and they are close to what Patel described. The FBI’s final 2025 Reported Crimes in the Nation release, published August 14, 2026, states that murder and nonnegligent manslaughter offenses decreased an estimated 18.1% from 2024 to 2025, and that “the 2025 murder rate of 4.1 per 100,000 inhabitants is tied with 1955 and 1956 for the lowest murder rate.” The bureau’s preliminary May 2026 release reported the same 18.1% figure alongside a 9.3% drop in overall violent crime.
Two qualifications follow directly from those same documents. First, the FBI does not describe the decline as the largest in American history. Its August release says the 2024-to-2025 change “marked the largest year-to-year decline in violent crime rates since FBI estimations began in 1936.” That is a statement about the length of the record, not about all of American history: the UCR series simply does not extend back further, so nothing before the mid-1930s can be compared. The May release used the phrase “since 1937.” Patel’s Senate formulation dropped the qualifier that the bureau’s own press officers had attached to the number.
Second, the murder rate of 4.1 per 100,000 is not, as some accounts of the hearing described it, a 20-year low. By the FBI’s own accounting it ties the lowest rate in the entire recorded series — a roughly 70-year benchmark. On that point the testimony, or its retelling, understated the finding.
The attribution is the weaker part of the claim. Murders were already falling steeply before the current FBI leadership took office. The bureau’s 2024 Reported Crimes in the Nation release recorded an estimated 14.9% decline in murder and nonnegligent manslaughter for 2024 — a year that concluded before Patel was confirmed in February 2025. Homicide investigation in the United States is overwhelmingly the work of municipal and county police departments; the FBI’s UCR Program is a data-collection function that aggregates what those agencies report. Nothing in the published data isolates a federal contribution to the decline.
Patel also told senators the bureau had arrested 54,000 violent criminals over 18 months, a 90% increase over the prior period, and 31,000 violent gang members. TIJ could not locate a published FBI data series supporting those operational counts. We treat them as unverified pending release of the underlying records.
Verdict: Mostly accurate on the statistics, overstated on the framing and unsupported on causation. The 18.1% decline and the 4.1 rate are real and come from the government’s own final data. “In the history of America” goes beyond what the record can show, and the FBI’s role in producing the decline is asserted rather than demonstrated.
Claim 2: Chuck Schumer says the Iran war has cost $38 billion, gas is up “nearly 50 percent,” and diesel just set a record
The claim. In floor remarks on September 16, Senate Democratic Leader Chuck Schumer said the war with Iran had cost $38 billion, that “the cost of diesel just hit a record high,” and that “the price of regular gas is up nearly 50 percent since the war began.”
The evidence. The $38 billion figure traces to a Congressional Budget Office report published September 16, 2026, “Estimating the Cost of Combat Operations Against Iran.” CBO estimates that as of August 1, 2026, the conflict had cost the Department of Defense approximately $38 billion, covering munitions replacement, equipment losses, flying hours, other operations and increased fuel costs. Schumer’s characterization of the figure as a floor rather than a ceiling is consistent with the report’s own scope: it measures DoD costs only, through August 1, and excludes spending by other parts of the federal government.
On fuel, TIJ pulled the Energy Information Administration’s weekly retail price series directly. EIA’s weekly retail gasoline and diesel prices show U.S. regular gasoline at $2.937 per gallon for the week of February 23, 2026 — the last full week before prices began climbing — and $4.319 for the week of September 14, 2026. That is an increase of 47.1%. Measured instead from the week of March 2 ($3.015), the increase is 43.3%. “Nearly 50 percent” is defensible on the earlier baseline and somewhat generous on the later one.
The diesel claim checks out in nominal terms and requires a caveat in real terms. EIA’s No. 2 diesel series, which begins in March 1994, shows $6.285 per gallon for the week of September 14, 2026 — the highest weekly figure in the series, surpassing the previous peak of $5.810 set the week of June 20, 2022. Adjusting that 2022 peak for inflation using the Bureau of Labor Statistics Consumer Price Index for All Urban Consumers (296.311 in June 2022 against 334.980 in August 2026) puts it at roughly $6.57 in current dollars. In inflation-adjusted terms, diesel is about 4% below its 2022 high. A record was set; it is a nominal record.
Schumer also cited a press account that “Americans have spent $100 billion more on fuel during the Iran War.” TIJ ran an independent approximation using EIA weekly prices and typical U.S. product-supplied volumes of roughly 8.9 million barrels per day of motor gasoline and 3.9 million barrels per day of distillate. Holding consumption constant and measuring the excess over pre-escalation prices from March 2 through September 14 yields roughly $84 billion in additional gasoline outlays and about $49 billion in additional distillate outlays. Together that is roughly $133 billion in additional fuel outlays, which makes the $100 billion figure conservative on these assumptions rather than inflated. The estimate is sensitive to those assumptions: it does not account for reduced consumption in response to higher prices, which would lower the true figure, and it attributes the entire price increase to the conflict.
Verdict: Accurate. The $38 billion figure matches CBO’s published estimate and is correctly characterized as partial. The gasoline and diesel claims hold up against EIA’s primary data, with the diesel record being nominal rather than inflation-adjusted.
Claim 3: President Trump says fuel prices are “going to come tumbling down” like “rocket ships in reverse”
The claim. At a September 16 rally in Gastonia, North Carolina, President Trump addressed fuel costs: “It’s a very inexpensive price to pay for what we’ve done… But that’s going to come tumbling down. That’s going to be like rocket ships in reverse.” He has separately said prices will fall “precipitously” once the conflict concludes.
The evidence. The most direct test is the federal government’s own energy forecast. The Energy Information Administration’s September 2026 Short-Term Energy Outlook, released September 9, projects Brent crude averaging about $90 per barrel through the second half of 2026 and falling “gradually” to an average of $74 per barrel in 2027. Retail gasoline is forecast to average $3.84 per gallon in 2026 and $3.35 in 2027. Retail diesel is forecast at $5.07 in 2026 and $4.40 in 2027.
Those figures point to a decline, but a measured one spread over more than a year — and one that leaves prices above where they were before the conflict. EIA’s 2027 gasoline projection of $3.35 remains above the 2025 annual average of $3.10. The agency also states that it assumes “some constraints to exporting oil from the Middle East will persist through the end of the year,” keeping regional crude production below pre-conflict averages until the second quarter of 2027. On diesel specifically, EIA forecasts U.S. distillate inventories remaining below the five-year low “through much of 2027.”
Predictions are not facts, and no forecast is binding. But when a claim about future prices can be tested against the responsible federal agency’s published projection, that projection is the relevant benchmark.
Verdict: Not supported by the administration’s own energy forecast. EIA projects a gradual, partial decline extending into 2027, not a rapid collapse.
Claim 4: The White House says a “factory comeback” is under way
The claim. A White House release circulated through mid-September asserts that “factories are hiring, construction is booming,” that manufacturers added 16,000 jobs in August — “the strongest monthly gain in three years” — and 58,000 this year, and that “more than one million private-sector jobs have been created in President Trump’s second term.”
The evidence. TIJ queried the BLS Current Employment Statistics series directly. Manufacturing payrolls stood at 12,638,000 in August 2026 against 12,622,000 in July — a gain of 16,000, exactly as claimed. Against December 2025 (12,580,000), the year-to-date gain is 58,000, also exact. Construction added 22,000 jobs in August (8,359,000 against 8,337,000). Total private employment was 135,752,000 in August 2026 against 134,711,000 in January 2025, a gain of 1,041,000 — above one million, as claimed.
Two points of context the release omits. The “strongest monthly gain in three years” is marginally off: November 2023 recorded a gain of 22,000 manufacturing jobs, which falls inside a three-year window from August 2026. The August figure is the strongest in nearly three years.
The more consequential omission concerns the baseline. Manufacturing employment in August 2026 (12,638,000) remains below the January 2025 level of 12,673,000 — a net decline of about 35,000 over the term to date. The 2026 gains are real and are recovering ground lost during 2025; they have not yet restored the sector to where it started. Separately, the one-million private-sector figure covers 19 months, an average near 55,000 per month, which is modest by historical standards even though the total is accurate.
Verdict: Accurate on the payroll figures, wrong on one superlative, incomplete on the trend. The employment counts check out against BLS data; the “strongest monthly gain in three years” claim does not. The framing omits that manufacturing payrolls are still below where they stood at the start of the term.
Claim 5: An election group says a professor’s model was “validated in professional scientific publications”
The claim. The Election Truth Alliance, a Nevada nonprofit with an audience in anti-Trump online spaces, has said that University of Michigan political scientist Walter Mebane Jr. found “evidence of potential election fraud, confirming ETA’s analysis” of Pennsylvania’s 2024 results, and stated in a press release that his “eforensics” model “has been validated in professional scientific publications.”
The evidence. Pennsylvania’s certified 2024 returns record 3,543,308 votes for Trump and 3,423,042 for Harris — a margin of 120,266. Mebane’s working paper estimated potentially fraudulent votes exceeding that margin, but the paper itself hedges the finding: he writes that Pennsylvania was a “key battleground into which extensive and intensive campaigning and mobilization efforts were directed,” adding that “maybe most or almost all of the incremental manufactured votes are false positives prompted by electors’ strategic behaviors.” The most conservative version of his model puts estimated illegitimate votes well below Trump’s margin.
On peer review, the record is clear. The eforensics model has not been published in a peer-reviewed journal; an early version submitted to the American Political Science Review in 2020 was rejected. Written reviews by Stanford political scientist Justin Grimmer and University of Pennsylvania political scientist Marc Meredith, filed publicly, dispute the model’s ability to distinguish fraud from ordinary variation. Caltech’s Michael Alvarez, with Grimmer and Meredith, ran the model against simulated data built to contain no fraud; it reported fraud at levels comparable to the Pennsylvania paper. Mebane has since revised the analysis, changing which counties he flags and deleting language characterizing the narrower estimate as “a not negligible proportion” of the margin.
Mebane himself has distanced his work from the group’s use of it, telling reporters he saw “many, many things” in the organization’s explanations that he did not think were “sound interpretations.” Pennsylvania’s 2024 results were subject to risk-limiting audits and county spot-check audits, both of which affirmed the count. In federal court, the group’s suit against the Secretary of the Commonwealth and three counties was dismissed; when it was refiled in August, the complaint made no mention of Mebane or his analysis. That case remains pending, with filings on the docket as recently as September 14. This reporting was first developed by Votebeat; TIJ reviewed the underlying certified returns, the court docket and the working papers.
Verdict: Unsupported. The model has not been validated in peer-reviewed publications, the author’s own paper allows that most flagged votes may be false positives, and the organization’s own refiled complaint dropped the analysis.
How TIJ fact-checks
TIJ does its own primary-source research. For each claim, we locate the original statement — a floor transcript, a hearing, a rally, an official release — and then go to the underlying record: federal statistical series, agency publications, certified election returns, court dockets, and academic working papers. Where a number can be recomputed, we recompute it. The gasoline and diesel figures in this edition were pulled from EIA’s weekly price files and calculated independently; the employment figures were queried directly from the BLS Current Employment Statistics database; the inflation adjustment uses the published CPI-U series.
We check claims from across the political spectrum and apply the same standard to each. A claim is rated on what the records demonstrably show, not on whether the underlying argument is persuasive. Where a figure is accurate but the framing omits material context, we say so. Where a claim rests on a forecast or an estimate, we identify the source and its stated limitations. Where we cannot verify a claim against a public record, we label it unverified rather than rating it.
All subjects of this fact-check are public figures or organizations acting in a public capacity. Requests for comment were not individually issued for this edition; all statements examined were made publicly and are quoted from the public record. Corrections and documented rebuttals may be submitted to TIJ’s editors and will be appended to this article.