PulsCar analysis of NHTSA public data, September 2026 Topic: does a noise complaint come with a diagnostic trouble code? 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 Rows loaded: 777,040 owner complaints received 1 January 2015 - 8 September 2026. - NHTSA ODI flat file MANUFACTURER COMMUNICATIONS (FLAT_TSBS), field dictionary https://static.nhtsa.gov/odi/ffdd/tsbs/TSBS.txt Rows loaded: 187,827 distinct communications (NHTSA_ID) over the same period, one database row per make/model/year covered by each communication. - Downloaded and indexed 17 September 2026; queries run 18 September 2026. QUERY 1 - "noise complaints" Full-text search (SQLite FTS5) over the complaint narrative for: noise, noises, noisy, grinding, grind, squeal(ing), squeak(ing), knock(ing), rattle/rattling, clunk(ing), whine/whining, hum(ming), tick(ing), screech(ing), growl(ing), chirp(ing), groan(ing) -> 57,202 complaints. Tightening 1 (negation): a complaint counts only if at least one noise word is NOT preceded within 40 characters by no / not / never / without / n't / any. This removes narratives such as "no unusual noises" and "NO WARNING LIGHTS ON DASH, NO NOISES". -> 55,554 complaints (204 of the later no-light matches were removed by this step). QUERY 2 - "the dashboard said nothing" Within query 1, narratives matching any of: no (trouble|fault|error|diagnostic|stored) code(s), no DTC, "did not set a code", no check engine light, no engine light, no warning light(s), no dashboard light, "no lights came on", "nothing showed up on the scan", "scan showed/found nothing", "codes were not set/stored/present", "without a check engine light/warning light/code". -> 3,697 complaints. Tightening 2 (proximity): the affirmative noise word and the no-code phrase must appear within 600 characters of each other, so that the two statements belong to the same described failure. -> 3,512 complaints. This is the published set. SAMPLE READING (required before publishing any keyword count) 40 random matches were read at the untightened stage. Off-target cases found: narratives that mention a noise only to deny it ("No unusual noises", "NO NOISES"), and narratives where the no-light sentence belonged to a non-audible failure. After both tightenings, 30 fresh random matches were read in full context: 29 described an audible symptom together with the owner's statement that no warning light or code appeared. One (a Kia K5 wind noise from the rear doors) is an audible symptom but not a mechanical wear failure. Observed false-positive rate for the published set: about 3% in a sample of 30, which is a small sample. EXCLUSIONS - Mileage statistics drop rows with miles <= 0 or >= 400,000 and rows with no mileage; 2,014 of the 3,512 complaints carry a usable mileage. - Model-year statistics drop years outside 1996-2026. - Median mileage is published per model only where at least 5 usable values exist. - No personal data is published: no VIN, no city, no dealer name, no free text. WHAT WAS DIAGNOSED (file: no-light-noise-what-was-diagnosed.csv) Within the 3,512, the text after "diagnosed with/as", "determined that", "found that", "replaced the", "replacing the", "needed a new", "needed to be replaced" was captured (up to 60 characters) and matched against part keywords. 969 of the 3,512 complaints contain such a phrase; that 969 is the denominator. A complaint can fall in more than one category, so shares do not sum to 100. This is a keyword categorization of owner-written text, not a repair-order audit. BULLETINS (file: noise-bulletins-by-make.csv) Same noise word list over the communication summary, counted by distinct NHTSA_ID: 13,868 of 187,827 communications (7.4%) describe a noise. Of those, 1,455 (10.5%) mention DTC / "diagnostic trouble code" / "trouble code" / a P0xxx code / "check engine light" / MIL / "malfunction indicator" anywhere in the summary. 64 summaries state explicitly that no code is set. The summary is NHTSA's published abstract of the communication, not its full text; a bulletin could mention a code in its body and not in the summary, so 10.5% is a floor for "mentions a code", and the true share is likely somewhat higher. BASELINE FIGURES QUOTED IN THE ARTICLE All 777,040 complaints: 5.11% report the vehicle was towed, 4.37% report a crash. 55,554 noise complaints: 4.72% towed, 1.81% crash, 1.83% fire. 3,512 noise + no-light complaints: 10.79% towed, 1.37% crash, 2.39% fire. Mileage in the 3,512: median 80,000, quartiles 48,913 and 110,000 (n = 2,014). LIMITATIONS - Complaints are unverified owner reports. They are not failure rates, and they are not normalized by how many vehicles of each model are on the road, so a large count for a high-volume model such as the F-150 partly reflects sales volume. - Nothing here ranks makes by reliability, and no such ranking can be derived from these counts. - Owners who file with NHTSA are self-selected, and owners describing a frightening noise may be likelier to file than owners with a quiet fault. - "Towed" is the owner's own answer to NHTSA's towed field; the higher towed share in the no-light group is an association in self-reported data, not a causal finding. ------------------------------------------------------------------------------ CASE SENSITIVITY — SECTION ADDED 19 SEPTEMBER 2026 ------------------------------------------------------------------------------ NHTSA stores complaint narratives in ALL CAPS for records filed through roughly 2020 and in mixed case for records filed from 2021 on. A case-SENSITIVE search of the narrative field loses about 92% of the complaints filed after 2020, so this is the single largest way an analysis of this database can go wrong while still looking finished. Every query behind this file is case-insensitive. SQLite's LIKE and its FTS5 full-text index both ignore case by default, and no query in this work uses GLOB or any case-sensitive comparison. This was not taken on trust. On 19 September 2026 every published folder was re-checked against the same database, comparing the figures here with what a case-sensitive query would have returned. The result: the complaint set published here (57,202) is LARGER than the absolute ceiling a case-sensitive search of the same terms could have returned (23,696 narratives). A case-sensitive query cannot produce a bigger set than itself, so the bug is ruled out for this folder by arithmetic, not by assurance.