Cases That Closed Empty: What Happened to the 2,262 Federal Defect Investigations That Found Nothing

This site is built on three federal tables. The recall table is the makers admitting something. The complaint table is the owners alleging something. And the investigation table is the referee: the Office of Defects Investigation opening a docket, gathering the evidence, and deciding whether the allegation becomes an admission. The Airbag Biography showed what that table looks like when a docket lands, six Takata case numbers exploding into a hundred thousand vehicle rows. This article is about the other outcome, which turns out to be the usual one. Collapse the table's 145,336 rows into distinct case numbers and the register has opened 3,525 investigations in its history. Of the 3,427 that have a close date, 1,165 ended with a recall campaign on file. The other 2,262 closed with nothing. Two out of every three times the referee blew the whistle, the final ruling was: no foul.

Two out of three

Start with the shape of the whole file. The register opened 596 investigations in the 1980s, peaked at 1,248 in the 1990s, ran 1,002 in the 2000s, and then the docket thinned dramatically: 450 in the 2010s, 229 so far in the 2020s. The agency investigates far less than it used to. And what it investigates, it increasingly converts. Of the dockets opened in the 1980s, 82.9% closed empty. The 1990s ran 71.6%. The 2000s dropped to 56.7%, the 2010s to 52.9%, and the 2020s sit at 55.4%. Read those two trends together and a story emerges: the early register threw open a case at the first scattering of complaints and usually found nothing, while the modern register opens half as many cases and hits on nearly half of them. The screening got better, and the thing doing the screening is the other table on this site. The complaint register, a quarter million entries a decade, became the instrument that decides which allegations are worth a docket, which is exactly the reading the percentile ladder was built for.

The ladder

The case numbers carry their type in the prefix, and the types are a ladder. A PE, a preliminary evaluation, is the standard first rung: the agency has seen enough complaints to ask the question formally. The register holds 2,079 closed PEs and 69.8% of them ended empty, which is the ladder working as designed, because the cases that survive a PE climb to an EA, an engineering analysis, the full-instrumentation second rung. Closed EAs number 771 and only 48.9% ended empty. Escalation cuts the odds of nothing roughly in half, because by the time engineers are instrumenting cars, somebody has already decided the smoke looks like fire.

Then there are the rungs the public never sees clearly. An RQ is a recall query, the agency auditing a recall that already exists, asking whether it went far enough, and accordingly it is the type most likely to end with campaigns on file: only 35.5% of 172 closed empty. An AQ is an audit query, a review of how a process behaved rather than whether a part is defective, and all 23 that closed, closed empty, which is the type doing exactly what it says. And then there is the DP, the defect petition, the one rung a private citizen can place on the ladder themselves. Federal law lets anyone petition the agency to open an investigation. The register holds 272 closed petitions. It granted recalls out of 7. The other 265, 97.4%, were denied or closed with nothing, which makes the petition docket the emptiest drawer in the building and the clearest sentence in the file about where the power to start a recall actually lives.

Finding nothing takes just as long

Here is the number I expected this cut to disprove, and it refused. You would think the empty cases are the quick ones, a docket opened on thin smoke and shut in a month. They are not. The median investigation that closed empty ran 138 days. The median investigation that produced a recall ran 162. At the slow end the gap stays just as narrow: the ninetieth percentile is 500 days for empty, 550 for producing. The register spends almost exactly as long establishing that nothing is wrong as it spends establishing that something is. Which is, if you think about it, how it has to work. You can confirm a defect the day the third inflator ruptures on a test stand. Proving an absence means running out every complaint, every fleet query, every maker response, and absence has no test stand moment. The file's most expensive product is the word no.

The long goodbyes

The slow tail of that distribution is its own exhibit. Ten dockets ran more than six years and still closed with nothing. The champion is an audit query into electrical issues that ran from 2009 to 2018, 8.8 years. Second place belongs to a defect petition, DP13001, a steering column case that took 7.1 years to deny, which must stand as the longest no in the register. A power train PE opened in December 2016 closed empty 7.1 years later. The sunroof file is here too: EA14002, the engineering analysis into spontaneous sunroof breakage, instrumented and studied for 6.7 years before closing without a campaign. And one entry closed this year: PE19014, a seats docket opened in 2019, which spent six and a half years becoming nothing.

The ghost in the speed control

Sort the 2,262 empty dockets by component and the top of the list is a monument. Vehicle speed control, 282 distinct investigations closed with nothing, the largest pile of empty dockets any component family ever generated. This is the register's long war with sudden unintended acceleration, the complaint that arrives in waves, terrifies everyone, and then, docket after docket, declines to leave physical evidence. Hydraulic brakes are second at 239 and steering third at 186, the other families where the symptom is dramatic, the complaint rate is high, and the hardware usually checks out. Air bags, for contrast, closed only 98 dockets empty out of the family's whole investigative history, because when that family is investigated, there tends to be shrapnel. The make table reads mostly as market share with a Detroit accent: Ford leads at 443 empty dockets, Chevrolet 320, and the first import does not appear until Toyota at 72, partly because the emptiest decades were the decades the domestic makers owned the road.

The shelf that never closes

Ninety-eight dockets have no close date at all, and they split into two very different shelves. Sixty-one were opened in the 2020s: that is simply the live docket, the cases still being worked. But twenty-eight were opened in the 1990s, and those are fossils. Read their summaries and some literally describe recalls that happened decades ago; the investigation did its work, the campaign ran, and nobody ever wrote the ending into this field. The register, which exists to record endings, has a small shelf of its own unfinished paperwork, and this site reads those entries as record-keeping gaps rather than thirty-year investigations. Even the referee's own file has a junk drawer, which is worth remembering whenever any single field in any register is treated as gospel.

So the doctrine, for reading any vehicle page on this site. A recall on the page is a verdict. A complaint is an allegation. And an investigation is a question, nothing more, because the register's own history says the answer came back no two times out of three, and the no took just as long to write as the yes. The parts bin tells you what breaks and the largest-recalls ledger tells you what happens when the answer is yes. This is the ledger of the other answer, which is the bigger pile: 2,262 times, the full machinery of a federal defect investigation ran its course, and the file it produced ends with the regulatory equivalent of a shrug, dated and signed.

Notes on the cut. The investigation table stores one row per make and model, so everything above is deduplicated to distinct case numbers, 3,525 of them. A docket counts as producing a recall if any of its rows carries a campaign number; where an outcome was never written into that field, the docket reads as empty, so the producing share is a floor, not a ceiling. Durations use dockets where both the open and close dates parse. The still-open shelf means no close date on file, which for the 1990s entries is missing paperwork rather than live cases. Docket types are read from the case number prefix, and the rarer prefixes, a few dozen dockets across several older series, are counted in the totals but not broken out.