PULSCAR ANALYSIS OF NHTSA PUBLIC DATA - SQUEAKS AND CREAKS WHEN TURNING THE WHEEL PulsCar analysis of NHTSA public data, September 2026. License: CC BY 4.0. Files in this folder may be reused with attribution to PulsCar (pulscar.io). -------------------------------------------------------------------------- 1. SOURCE FILES -------------------------------------------------------------------------- Complaints FLAT_CMPL.txt 777,040 unique complaints (ODINO), received 1 January 2015 - 8 September 2026. Field definitions: https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt Communications FLAT_TSBS.txt manufacturer communications, counted DISTINCT on the NHTSA document id. Recalls FLAT_RCL.txt campaigns counted DISTINCT on the campaign number. Complaints are OWNER REPORTS, not verified repairs, and there is no denominator of vehicles on the road. Nothing here ranks reliability. -------------------------------------------------------------------------- 2. COMPLAINT QUERY (the 495) -------------------------------------------------------------------------- Step 1 - full-text search for noise words (squeak, squeal, creak, groan, whine, pop, clunk, grind, rattle, knock, click) -> 24,738 complaints. Step 2 - proximity filter: a noise word and a turning phrase (turn / turning / steering wheel / cornering / parking lot) within 80 characters of each other inside one clause -> 4,025. Restricting the noise word to the four this article is about (squeak, squeal, creak, groan) -> 531. Step 3 - FIRST SAMPLE READING. 30 random matches read in full; 5 of 30 (17%) were not the article's symptom. The failure mode was the word "turn" meaning something else: "turning off my car", "turn the volume up", "rotors were turned", "turned into a screeching noise", plus one negation ("no squealing noted on full wheel turn"). Step 4 - tightening. Discard a match when the matched span contains turn(ing|ed) on|off|up|down|into|over, turn signal, "rotors turned", "rotors machined", "turning on the car" or when the 90 characters before it contain a negation (no|not|never|without|n't) ... (squeak|squeal|creak|groan). -> 495 complaints, 8 January 2015 to 4 September 2026. This is the published set. Step 5 - SECOND SAMPLE READING. 30 fresh random matches read in full: 2 of 30 were still off-target (a squeal when shutting the car off; a squeal while cranking). Residual false-positive rate in the read sample: 2 of 30, about 7%. Reports where the squeak when turning came from the brakes or from a seat were KEPT, because they are what a driver hears when turning and they are categorized separately in the outcome table. Structured-field counts: towed 1.4%, crash 0.8% - both far below the all-complaint baselines (5.1% and 4.4%), which fits a symptom that is usually an annoyance. Mileage: 257 records carry a usable figure (blanks, zeros and 400,000+ dropped); median 65,000 miles, quartiles 35,000 and 99,776, and 68 of 257 (26%) are at or under 36,000 miles. -------------------------------------------------------------------------- 3. OUTCOME READING (the 110) -------------------------------------------------------------------------- 255 of the 495 contain a diagnosis or repair cue (diagnos*, was told, found that, determined, replaced, repair*, fixed, the cause, turned out, dealer said, technician). 110 were drawn at random from those 255 - which is 43% of the cue-bearing subset, a large fraction, so the sample is close to a census of the complaints that say what a shop found. Each was read in full and given ONE primary outcome: the last or most serious component the owner names. Counts are in squeak-turning-outcomes-110-read.csv. Limitation: "no cause found" includes cars still in the process. And an owner's report of what a shop said is not a verified repair. -------------------------------------------------------------------------- 4. MANUFACTURER COMMUNICATIONS (the 342) -------------------------------------------------------------------------- Full-text search of bulletin summaries for a squeak/creak/squeal/groan word together with steering or turning returns 446 distinct documents; requiring the two to appear within 90 characters of each other leaves 342, dated 17 January 2015 to 9 September 2026. They are published in steering-noise-bulletins.csv with a cause category assigned by matching the summary text: 86 steering column, intermediate shaft, boots and seals 35 suspension: strut bearing, shock, spring seat, bushing 9 steering rack or gear 6 body, interior or trim rubbing 4 steering wheel, airbag module or clock spring 3 hydraulic power steering: pump, fluid, hose, belt 199 not classifiable from the summary That last number is honest, not lazy: many summaries name no component (" CREAKING/SQUEAKING NOISES FROM THE FRONT AXLE WHEN STEERING"), and a few are concatenations that bundle unrelated concerns, so a handful of the 342 are not about steering noise at all. Category counts are therefore a floor for each category, never a share of all steering noises. Bulletin counts are paperwork counts. Mazda's 41 documents are dominated by re-issues of the same front shock absorber and strut bearing advice across model years, and Land Rover's 38 repeat the same steering pinion seal lubrication note. -------------------------------------------------------------------------- 5. RECALLS -------------------------------------------------------------------------- Searching recall defect and consequence text for a noise together with steering returns 7 distinct campaigns covering 253,704 vehicles. -------------------------------------------------------------------------- 6. WHAT IS NOT HERE -------------------------------------------------------------------------- Repair prices. They are in no federal file, and the estimator sites that carry them refuse automated access, so no price on the article page was re-verified in this pass. ------------------------------------------------------------------------------ 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 (24,738) is LARGER than the absolute ceiling a case-sensitive search of the same terms could have returned (873 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.