The Failure Mode

A state agency publishes a list of the dams it regulates. One column is labeled hazard potential, and a number of rows read High. The reasonable inference is that those are the dams in bad shape. The federal guideline that defines the term rules that inference out in one sentence, and the federal inventory keeps the shape of the dam in a different field. Both fields sit in the same record and answer different questions: one by looking downstream of the dam, the other by looking at the dam. Nothing requires them to agree. A High hazard class beside a Satisfactory condition is not an error, and neither is a Low class beside an Unsatisfactory one. What follows is the chain from a barrier across a creek to a number in a headline, with the governing text at each link. Six links, and where each one stops carrying something Links 3 and 4 fill two different fields in the same record. Link 6 reads them as one. 1 A barrier meets the statutory definition of a dam Height or storage...

A contractor's excavator catches a buried pipeline. Product escapes, crews respond, the street is opened, damaged pipe is cut out and replaced, and some weeks later the operator's accountants settle on a number for what the day cost: roughly $100,000. Whether that release ever becomes part of the federal record of American pipeline failures — whether it is counted at all — depends almost entirely on what the pipe was carrying. Two sections of federal regulation decide it, and the same agency administers both. Under 49 CFR 191.3 , a gas release becomes a reportable incident when estimated property damage reaches a figure that, for the year beginning July 1, 2026, stands at $153,600 . Under 49 CFR 195.50 , a hazardous liquid release becomes a reportable accident when estimated property damage exceeds $50,000 . A $100,000 release on a liquid line is an accident in the federal record. The same $100,000 on a gas line produces no report, no row, and no entry in any trend line. ...

A bridge classified as Poor sounds like a bridge about to close, and the number of them in a state reads like a scoreboard. The federal system does not keep score that way. The National Bridge Inventory carries a condition class for each public highway bridge, but the measure written into federal regulation does not add the bridges up. It adds up their decks. In the 2026 inventory, 41,148 of 624,939 bridges were classified Poor — 6.58 percent of the structures, and 4.77 percent of the deck area. That 1.81 percentage point gap is not a rounding artifact. It is the design of the measure. On the network the regulation actually governs, the same comparison runs the other way: the Poor class holds 2.81 percent of the count and 3.96 percent of the deck area. What follows is the chain that turns an inspector's coding sheet into a percentage, and the arithmetic that decides which way the gap opens. The Chain Seven steps separate a person standing under a girder from a percentage in ...

Suppose a household reads 48 ppb of lead at its kitchen tap, and in the same six-month window its utility reports no exceedance of the lead action level. Both can hold at once, and neither has to be a clerical error. For most systems the compliance decision is not made on any single tap, nor on an average: it is made on one ranked position in a short ordered list of results; results ranked above that position do not enter the comparison. Where a state has approved fewer than five samples in writing, the highest result is the answer. The Chain The pool is drawn first. Under 40 CFR 141.86(a) a system evaluates the materials in its distribution system and identifies a pool of targeted sampling sites. The pool is ranked by plumbing, not by geography. The sample is defined by standing time and by tap. A first-draw litre after at least six motionless hours. The laboratory returns one number per site , in milligrams per litre. Results are sorted ascending and numbered 1 throug...