PulsCar analysis of NHTSA public data, September 2026 Topic: what happens when a car visits more than one repair shop, and what owners say they were charged just to have a problem diagnosed License: CC BY 4.0 (attribution: "PulsCar analysis of NHTSA public data, September 2026") SOURCES - NHTSA Office of Defects Investigation flat file COMPLAINTS (FLAT_CMPL), field dictionary https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt 777,040 owner complaints received 1 January 2015 - 8 September 2026. Downloaded and indexed 17 September 2026; every query below was run 18 September 2026 against a local read-only SQLite copy. - Baselines used for comparison, computed on all 777,040 complaints: towed 5.11%, crash 4.37%, fire 1.82%. A NOTE ON LETTER CASE, BECAUSE IT CHANGES EVERY COUNT NHTSA complaint narratives are stored ENTIRELY IN CAPITALS through 2020 and in ordinary mixed case from 2021 onward. We measured it: of complaints received in each year, the share whose first 200 characters are all uppercase is 100.0% for 2015-2020, 42.5% for 2021, then 1.6% (2022), 1.1% (2023), 0.8% (2024), 0.8% (2025) and 0.6% (2026). Consequence: any case-sensitive search silently deletes the modern half of the file. Running an uppercase-only pattern ("SECOND OPINION", "ANOTHER DEALER", "ANOTHER SHOP", "ANOTHER MECHANIC", "TWO DIFFERENT") against our final set returns 918 of its 5,147 complaints - 17.8%. Every regular expression below was compiled with Python's re.IGNORECASE. Where SQL was used it was SQLite LIKE or FTS5, both of which are case-insensitive for ASCII. No uppercase GLOB was used anywhere. QUERY 1 - complaints that name a repair venue (the denominator) Narratives matching \b(dealer(ship)?s?|shops?|mechanics?|garages?|service centers?|service departments?|technicians?)\b : 366,729 of 777,040 (47.2%). This is the set in which a shop is discussed at all, and it is the comparison set for the percentages below. QUERY 2 - the car went to more than one repair venue for the same complaint Step 1, phrase match: \b(second|2nd|third|3rd|fourth|another|different)\s+ (local |independent |authorized |certified )?(dealer(ship)?s?|shops?| mechanics?|garages?|service centers?)\b OR \b(two|three|four|five|2|3|4|5|several|multiple|numerous)\s+(different\s+)? (same venue list)\b OR "second opinion" -> 10,950 complaints. Step 2, first sample reading (30 random matches, read in context): 6 of 30 were not a second visit at all. They were owners who CALLED several dealers about part availability, backorders, price matching or waiting lists. Measured false-positive rate for the intended meaning: about 20%. Step 3, tightening: a match counts only if, within 160 characters of the phrase, the narrative also contains a taking-the-car verb (took, taken, take, brought, bring, went to, drove it to, towed to, visited, had it at/looked/checked/ inspected/diagnosed, inspected, diagnosed, examined) AND does NOT contain a parts-supply word (part, parts, availab*, backorder, inventory, in stock, price match, waiting list, supply) in the same window. Result: 5,147 complaints. Step 4, second sample reading (30 fresh random matches from the tightened set): 28 of 30 describe a vehicle that was physically taken to a second repair venue for the same complaint. The 2 misses are owners who telephoned or "spoke with" several shops. Measured false-positive rate after tightening: about 7%. The article calls this set "complaints in which the car went to more than one shop", never "how often cars go to more than one shop": complaints are self-reported, and only a fraction of owners write to the federal government at all. 5,147 is 1.40% of the venue-naming set and 0.66% of the whole file. Findings published in second-shop-outcomes.csv, second-shop-complaints-by-make.csv and second-shop-top-models.csv: - towed 10.30% against the 5.11% all-complaint baseline; crash 1.87% against 4.37% and fire 1.09% against 1.82%, both BELOW baseline. - "nobody could find it / could not duplicate it" language: 23.98% of the set against 12.85% of the venue-naming set - the same regular expression run on both, so the ratio is the finding, not the absolute level. - "the repair did not hold / the problem came back": 21.33% against 8.72%. - a recall is mentioned in 24.15%; a part was replaced in 41.56%; the words "second opinion" appear in 8.12%. - mileage is usable (>0 and <400,000) in 3,196 of 5,147: median 70,000, lower quartile 27,000, upper quartile 106,000; 954 (29.85%) are at or under 36,000. - makes and models are counts of complaints, NOT failure rates. No registration or sales denominator exists in this file, so a make with more cars on the road will produce more complaints. The CSVs say so in their own right and the article never ranks brands by reliability from them. QUERY 3 - what owners say they were charged to have a problem diagnosed Step 1, loose match: a dollar figure within 60 characters of diagnostic / diagnosis / diagnose / diag, either order, value between $20 and $1,000 -> 2,278 complaints, median $176.48. Step 2, sample reading (30 random matches): 5 of 30 were not diagnostic charges. They were repair quotes that happened to sit near the word "diagnose" ("$789.88 to diag & replace steering gear"), rental-plus-diagnostic totals, and one narrative where the pattern grabbed the repair price offered as an alternative to the fee. Measured false-positive rate: about 17%. Step 3, tightening: the dollar figure must be adjacent to the diagnosis word with no more than a short connector between them, and the matched span itself must not contain replace*, repair*, rental, rent, tow*, part(s), install*, quote(d), estimate(d), labor, engine or transmission. One figure per complaint, the first qualifying one. Result: 1,089 complaints. Step 4, second sample reading (30 fresh random matches from the tightened set): 30 of 30 are charges to diagnose, quoted or paid. Measured false-positive rate after tightening: 0 of 30. A sample of 30 cannot distinguish 0% from about 10%. Findings published in owner-reported-diagnostic-charges-by-year.csv: - all years: median $164, quartiles $125 and $200, range $30-$1,000, n=1,089. - by year of the complaint, the median rises every year or holds: $114 (2015), $120, $125, $135, $145, $150 (2020), $150 (2021), $174, $179, $193, $200 (2025), $210 (2026). Per-year counts run from 57 to 142, so each year's median rests on a small sample and the year-to-year steps are not significant on their own; the direction across twelve years is the finding. - 526 of the 1,089 narratives also contain the word "recall" and 318 contain "warranty": owners write about a diagnostic charge most often when they believed somebody else should have been paying. - where the fee sits within 250 characters of dealer language and no independent language (667 complaints), the median is $160. The mirror case - independent language and no dealer language - has only 21 complaints, which is too small to publish as a comparison, and the article says so instead of quoting it. LIMITATIONS - NHTSA complaints are unverified owner reports about vehicle defects. They are not a sample of repair visits, not a sample of shops, and not a failure rate. People write to the federal government when something went badly, so every percentage here describes complained-about repairs. - The file records the owner's account only. There is no shop-side record, no invoice and no confirmation that a diagnosis was right or wrong. - "Nobody could find it" is the owner's summary of what a shop told them. It is not evidence that the shop was careless, and the article does not say it is. - Dollar figures are as typed by owners: some are quotes rather than paid bills, some are rounded, and a few are plainly approximate. - Model-year and mileage fields are frequently blank; denominators are stated wherever a subset is used. - No personal data is published: no VIN, no city, no state, no dealer name, no free text beyond the short quoted snippets in the article itself. OTHER DOCUMENTS OPENED FOR THIS ARTICLE (not part of the CSVs) FTC "Auto Repair Basics"; Michigan Compiled Laws 257.1332; New York Vehicle and Traffic Law 398-c and 398-d; California Bureau of Automotive Repair "Write It Right"; the Texas Department of Licensing and Regulation program list; the ASE Blue Seal of Excellence requirements; the AAA Approved Auto Repair program page; and the FTC Consumer Sentinel Network Data Book 2024. All opened 18 September 2026 and cited with links in the article's sources section.