As of 2026-08 no federal rule governs the wayside heat detectors on trains carrying hazardous materials, none of the 34 FRA proposed rules since the NTSB report of 2024-06-25 concerns bearing detection, and FRA inspection reports ended the audit series in 2023 below their 2018 level

What is happening?

FRA Safety Advisory 2023-01, published 2023-03-03, records that on 3 February 2023 a freight train rolled past three trackside heat detectors in Ohio. The failing bearing read 38 degrees F above ambient at the first, 103 about ten miles later, and 253 about twenty miles after that — past the 200 degree critical threshold the operating railroad had set for itself. 38 railcars derailed in an Ohio village, 11 hazardous materials tank cars ignited, and about 2,000 residents were evacuated. That threshold was one railroad's own number and not a public standard: the NTSB survey of the six Class I railroads, taken later that year, found six different answers on how far apart detectors sit and on how hot is too hot.

Three and a half years later there is still no federal regulation that says how far apart those detectors may be, how hot is too hot, or how they must be inspected and maintained. FRA said so itself in March 2023, and a full-text search of the current Code of Federal Regulations returns zero sections containing the phrases hot bearing detector, wayside detector, or bearing defect detector. The NTSB told FRA to fix this in June 2024 in recommendations R-24-2 through R-24-5. Since that date FRA has published 34 proposed rules in the Federal Register and not one of them concerns bearing defect detection.

What FRA did publish instead was a deregulatory round — 21 proposed rules on a single day, 2025-07-01, and 11 final rules on a single day, 2026-04-28. One of the eleven, effective 2026-05-28, formally confirms that the FRA Office of the Chief Counsel may decline or dismiss a violation. Meanwhile the inspection workforce that would catch such defects by hand filed 71,326 inspection reports in 2018 and 63,427 in 2023, an 11 percent decline, with nearly 400 federal and more than 150 state inspectors covering the entire national network.

Whose problem is this?

The party that sets the temperature at which a train stops is the railroad, and the party that lives where the train stops too late is the town. Nothing in federal law connects the two.

Where does this problem end?

The boundary matters because this is not a case of a rule being broken. There is no rule to break, so the ordinary machinery of citation and penalty has nothing to attach to.

What is the state now, and what should it be?

Now

Needs a new measurementThe target state has a shape but no number. NTSB recommendation R-24-3 asks FRA to establish minimum requirements including criteria for alert and alarm thresholds and maximum distances between wayside detectors, and R-24-4 asks for installation, inspection and maintenance requirements. But R-24-2 asks FRA to research the effectiveness of current systems first, so no source we opened states what the correct spacing or threshold actually is. The target is a rule that does not yet exist, not a value that FRA is failing to hit.

How big is it?

The population this dossier counts is the people evacuated in FRA-reportable rail accidents in which at least one hazardous materials car released its contents. Over 2021 through 2025 that was 64 accidents and 3,577 people, a five-year mean of about 715 per year. In 2025 alone it was 380 people across 15 such accidents.

That measure is narrow on purpose, and it is narrow in a way that matters. Evacuation is the only harm in this chain that the federal record actually counts, and it counts only the people who were told to leave. It does not count people who sheltered in place, and it does not count the far larger number who live beside the routes that carried 6,738 hazardous materials cars through reportable accidents in 2025, of which 24 released.

The distribution is dominated by single events. In 2023 the figure was 2,355; in 2021 it was 36. A range built on this measure is a range of outcomes, not a range of uncertainty.

Under what conditions does it arise?

1. The detector is private infrastructure doing a public job. Wayside heat detectors are owned, sited, calibrated and threshold-set by the railroads themselves, so the number that decides whether a train stops is a private business rule rather than a public standard. The NTSB survey found the six Class I railroads had picked six different answers. 2. The gap is invisible between accidents. Detectors spaced 25 miles apart instead of 12.7 miles produce no citation, no report and no statistic until a bearing fails in the interval. There is no compliance record because there is nothing to comply with. 3. The enforcement arm has not grown while its discretion is expanding. Inspection report volume in 2023 sat 11 percent below its 2018 level after peaking in 2020, and the 2026 rule confirms in regulation that enforcement attorneys may decline or dismiss violations without publishing each dismissal. 4. Legislative attention is periodic, not cumulative. The Railway Safety Act has been introduced four times in the 119th Congress and each version restarts at the first stage rather than building on the last.

What has been tried?

Every one of the rules in the last row removed or relaxed an existing requirement rather than adding one. None of them touched wayside detection.

What was found?

Why is it still unsolved?

Institutional gap — the safety measure everyone points to after the 2023 Ohio derailment was never written into federal law, so there is no rule to break and nothing for enforcement to attach to; the second and lesser axis is that the hand-inspection capacity covering the same network is shrinking and turning more discretionary at the same time.

The federal rulebook has nothing to say about the equipment that is supposed to catch an overheating bearing before it burns through an axle. FRA wrote that sentence itself in March 2023, in the safety advisory it issued a month after the derailment, and the sentence is still accurate in August 2026. An advisory is a request. It creates no inspection item, no citation, no penalty and no data series. The consequence is that the single variable that determined the outcome in that derailment — how far apart the detectors sat and at what temperature they were set to shout — remains a matter of six private judgments by six companies.

The NTSB closed the loop in June 2024 by naming exactly what FRA should do, in four recommendations that build on one another: research first, then minimum requirements for thresholds and maximum spacing, then installation and maintenance requirements, then rules for how railroads must respond to an alert. None of those four has produced a Federal Register document. Instead the agency spent 2025 and 2026 on a deregulatory program carried out under two executive orders, in which the only enforcement-adjacent rule to reach final form was one confirming the discretion to let violations go.

The legislative route has been open the whole time and has never closed a single step. Four Railway Safety Act bills have been introduced across the 119th Congress, including one carried by a senator from the state where the derailment happened, and every one of them sits at the introduced stage. A bill that is reintroduced each session without advancing is not a slow fix. It is a stable equilibrium in which the appearance of action substitutes for the thing itself.

What observation would mean it is solved?

Candidates — (a) a Federal Register final rule that sets maximum wayside detector spacing, minimum alarm thresholds, and maintenance requirements (b) a sustained rise in FRA inspection reports and in inspector headcount (c) several consecutive years with no hazardous materials release accident above some agreed size.

(c) alone is worthless. A year with no large release can happen by luck alone, so a single quiet year proves nothing about whether the gap was closed. The 2021 figure of 36 people evacuated sat two years before the 2023 figure of 2,355, with no rule change in between.

(a) alone is not enough either, because a rule with no inspection behind it repeats the current pattern one level up: the CFR would gain text while the field capacity kept shrinking. And (b) is gameable — inspection report counts are a volume measure, and the DOT OIG already found coding errors and missing detail in the underlying records, so more reports can mean less inspection.

The honest reading is that (a) and (b) must both move, and (c) must be watched over a decade rather than a year.

What is it connected to?

Fills with researchthis almost certainly connects to PHMSA tank car standards, to the two-person crew rule, to the precision scheduled railroading operating model and train length, and to local emergency responder funding. Relation type and evidence grade were not established in this round.

What these sources do not say

• What share of the national network is inspected in a year. No source we opened gives track miles inspected, route miles per inspector, or any coverage ratio. The inspector headcount and the report count exist; the denominator does not.
• Whether the NTSB recommendations to FRA are open or closed. NTSB classifies every recommendation as acceptable, unacceptable or closed, but the public query interface rejected every request we constructed, so the current classification of R-24-2 through R-24-5 is unknown to us.
• How many detectors exist, and where. The NTSB survey reports average spacing per railroad, not a count, not a map, and nothing at all for the short line and regional railroads that carry the same tank cars.
• How many violations FRA has actually declined or dismissed. The 2026 rule says dismissals will appear in an annual enforcement report, and FRA explicitly refused to publish them individually. We did not locate a fiscal year 2025 or 2026 enforcement report.
• Whether inspection reports continued to fall after 2023. The OIG series stops at 2023. Whether 2024 and 2025 continued the decline, leveled off, or recovered is not in any document we read.
• What any of this costs. No source quantifies what a national detector standard would cost the railroads, nor what the 2023 evacuation and cleanup cost the public.

See the evidence

What was read directly, and what was not. The NTSB final report PDF, the two Federal Register full texts, and the DOT OIG report PDF were downloaded and read in full; the OIG figures come from the report file at the path /sites/default/files/library-items/FRA Roadway Worker Protection Final Report_5-14-2025.pdf reached from the library item above. The accident and hazardous materials counts were computed by us from the FRA Form 54 dataset rather than copied from a published table, which means they will not match FRA summary publications exactly if those use a different accident-type filter or a different reporting cutoff. Several pages, including federalregister.gov article pages, congress.gov and railroads.dot.gov, refused the fetch tool from this environment and were retrieved instead by direct HTTP request or through the corresponding public API; the FRA safety data query site returned an internal error and was not used.

One count needs explaining, and one pair of numbers only looks like a disagreement. The NTSB investigation page and the executive summary of the final report both give 34 new safety recommendations, while the numbered rail series runs R-24-1 through R-24-33. The thirty-fourth is I-24-1, addressed to PHMSA on placard survivability, and it sits inside the same list. The report also reiterated one previously issued recommendation, R-13-26, and reclassified four others. Separately, the NTSB page describes 5 vinyl chloride tank cars and 3 mechanically breached cars while FRA Safety Advisory 2023-01 describes 11 hazardous materials tank cars that ignited; these count different things and we cite each to its own source. This is a Path A output (research-based definition), so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range 380–715.4 As of 2021-2025

Sensitivity The interval 380 to 715 is not a confidence interval. It is the distance between the most recent complete year and the five-year mean of the same measured quantity, and the spread exists because the annual figure is driven by a handful of large events rather than by a stable rate: the five yearly values run from 36 to 2355. Three limits push the true affected population upward and are not corrected for here. First, FRA counts only people evacuated, so anyone told to shelter in place, anyone who left without an evacuation order, and anyone exposed without leaving is outside the count. Second, FRA collects Form 54 only for accidents above the annual monetary damage threshold, so smaller releases never enter the series. Third, late and amended railroad reporting can revise recent years upward after the extract date. One limit pushes the other way: an evacuation count measures realized harm in a given year, not the standing population at risk, and the far larger number of people living beside routes that carried 6738 hazardous materials cars through reportable accidents in 2025, only 24 of which released, is not in this figure at all. We chose the evacuated count because it is the only element of this chain that the federal record actually measures.

Regional breakdown The dataset does carry a state field, so a per-state evacuation tally is technically computable, but the quantity being reported here is a five-year annual mean dominated by single events. Splitting a mean of that shape by state would attribute the whole national figure to whichever few states happened to host a large derailment in the window, which describes accident history rather than exposure. The state where an evacuation occurred is also not the state where the risk lives, since hazardous materials routes cross many states between origin and destination. No source we opened gives hazardous materials route miles or trackside population by state, so the denominator that would make a state breakdown meaningful does not exist in our material.