PulsCar analysis of NHTSA public data, September 2026 Shaking or vibration under acceleration: 4,463 owner complaints, 589 manufacturer bulletins, 22 recall campaigns Licensed CC BY 4.0. Queries run 18 September 2026. -------------------------------------------------------------------------- SOURCE FILES -------------------------------------------------------------------------- NHTSA ODI complaint flat files COMPLAINTS_RECEIVED_2015-2019, _2020-2024, _2025-2026 -- https://static.nhtsa.gov/odi/ffdd/cmpl/ (dictionary CMPL.txt) 777,040 distinct vehicle complaints (ODINO, product type V) received 1 January 2015 to 8 September 2026. NHTSA recall flat file FLAT_RCL.zip -- campaign, make, model, model year, component, manufacturer, potentially affected, report date, defect, consequence, remedy. NHTSA manufacturer communication files TSBS_RECEIVED_2015-2019, _2020-2024, _2025-2026 -- https://static.nhtsa.gov/odi/ffdd/tsbs/ (dictionary TSBS.txt) Loaded into SQLite and queried with Python 3 and the sqlite3 module. -------------------------------------------------------------------------- CASE SENSITIVITY -- READ THIS BEFORE REUSING THE QUERIES -------------------------------------------------------------------------- NHTSA complaint narratives are ALL CAPITALS for complaints received through 2020 and mixed case from 2021 onward. A case-sensitive regex, or an upper-case SQLite GLOB, silently drops most of the post-2020 file while still returning thousands of rows, so it looks like it worked. Every query below is case-insensitive and was written that way on purpose: - the SQL stage uses LIKE, which is case-insensitive for ASCII in SQLite; - every Python regular expression carries re.IGNORECASE; - no GLOB, no upper()/lower() comparisons against a cased literal. The word lists below are printed in lower case for readability. They match regardless of case. -------------------------------------------------------------------------- WHICH COMPLAINTS COUNT (the acceleration-shake set: 4,463) -------------------------------------------------------------------------- STAGE 1 -- SQL pull, case-insensitive LIKE on the narrative: descr LIKE '%shak%' OR '%vibrat%' OR '%shudder%' OR '%judder%' OR '%tremb%' 32,543 candidate complaints. STAGE 2 -- Python, per sentence. The narrative is split on sentence punctuation and newlines, and a complaint counts when ONE SENTENCE contains both a vibration word and an acceleration word. vibration words (regex, IGNORECASE): shak* | vibrat* | shudder* | judder* | trembl* acceleration words (regex, IGNORECASE): accelerat* | on the gas | gas pedal | press* the gas | step* on the gas | give it gas | under load | throttle | speeding up | speed up | picking up speed | going uphill | up a hill | on takeoff | take off | from a stop EXCLUSIONS, applied to the same sentence: unintended / sudden / unintentional acceleration, "accelerated on its own" -- these are runaway-throttle complaints, a different defect; "shudder to think" and the common misspelling "shutter to think" -- a figure of speech, not a symptom; "does not / doesn't / did not / didn't speed up | vibrate | shake" -- denials. Makes excluded (the same list our other analyses use): motorcycle, heavy-truck, bus, trailer and RV manufacturers -- Harley-Davidson, Indian, Triumph, Ducati, Kawasaki, Yamaha, Suzuki, Polaris, Can-Am, KTM, Vespa, Piaggio, Aprilia, Moto Guzzi, Arctic Cat, Royal Enfield, Husqvarna, Zero, Mack, Hino, Kenworth, Peterbilt, Paccar, International, IC Bus, Freightliner, Western Star, Blue Bird, Thomas Built, Navistar, Prevost, Workhorse, Spartan, Autocar, Forest River, Volvo Trucks, Volvo Bus, Isuzu Trucks, UD Trucks, Sterling, Gillig, U-Haul, Elpers, Regency. Removing them took 4,477 complaints to 4,463. SAMPLE READING (required before publishing any keyword count) Two samples of thirty randomly drawn matches were read in full. - The first query used a 90-character proximity window instead of a sentence boundary and measured FOUR mis-sorts in thirty (13%): a metaphorical "I shudder to think", a sentence saying the vibration does NOT change with speed, and two idle-only narratives. The window was replaced with the sentence rule and the three exclusions above. - The final query, sampled again with a different seed, measured THREE mis-sorts in thirty, ROUGHLY 10%: one narrative about a vehicle that accelerated in reverse and shook separately, one that listed a cold-start shake and an acceleration kickback in the same sentence, and one about "accelerated frame corrosion", where "accelerated" is an adjective. Read every count below as carrying roughly one mis-sort in ten. The dominant residual error is a vibration word and an acceleration word landing in the same sentence without describing the same event. -------------------------------------------------------------------------- FIELDS AND RESULTS -------------------------------------------------------------------------- Outcome flags (share answering Y, of 4,463): towed 2.29% (102), crash 1.14% (51), fire 0.29% (13), 25 injuries, 0 deaths. All-complaint baseline (777,040): towed 5.11%, crash 4.37%, fire 1.82%. Mileage: NHTSA MILES field, rows with 0 < miles < 400,000 (2,566 rows, 57.5% of the set). Median 60,000; quartiles 22,000 and 93,000; 34.4% at or under 36,000 miles; 50.7% at or under 60,000. All-complaint baseline median (392,935 usable rows): 70,000. Speed bands owners name in the text: regex \b(\d{1,3})\s*(?:-|to|and)\s* (\d{1,3})\s*mph, IGNORECASE, keeping pairs with 0 < low < high <= 100. 344 complaints (7.7%) name at least one band; 389 bands in total. Median band 30-45 mph. Grouped by band midpoint: under 20 mph 43 (11.1%); 20-34 mph 114 (29.3%); 35-49 mph 124 (31.9%); 50-64 mph 72 (18.5%); 65 mph and above 36 (9.3%). This is a count of the minority who volunteer a band, not of all 4,463. Phrase subsets, share of the full 4,463 narratives: "no fault found" family 213 (4.77%) -- all-complaint baseline 2.64%. The family is: no fault / problem / issue / code found or detected; could not / couldn't find, duplicate, replicate, reproduce; unable to find, duplicate, replicate, reproduce, diagnose; nothing found or nothing wrong; no codes; cannot duplicate; normal operating characteristic; operating as designed; within specification. mentions a recall, warranty extension, service campaign, class action, settlement or TSB 1,207 (27.0%) -- all-complaint baseline 32.5%. warranty-friction words (out of warranty, warranty expired, denied, refused, wear item, not covered) 346 (7.75%) -- baseline 5.13%. The three baselines above were computed by scanning all 777,040 narratives with the identical case-insensitive regexes. WHAT PART FAMILY THE NARRATIVE NAMES (4,463 complaints, categories overlap) Each category is a case-insensitive regex run over the whole narrative. A complaint can match several categories, so the shares do not sum to 100, and a complaint that names no part at all matches none. Transmission or torque converter 1,908 42.75% Misfire, plugs, coils or injectors 358 8.02% Tires, wheels, balance or alignment 275 6.16% Software, reprogram or relearn 209 4.68% Engine internals or engine replaced 207 4.64% Fuel system or fuel pump 194 4.35% Catalytic converter or exhaust 103 2.31% CV axle, half shaft or drive axle 95 2.13% Suspension or steering parts 61 1.37% Driveshaft, U-joint or carrier bearing 55 1.23% Engine, motor or transmission mount 49 1.10% Wheel bearing or hub 20 0.45% The tire row is deliberately wide: it matches the bare word "tire" anywhere in the narrative. The narrower question -- how many narratives name a balance, an out-of-balance wheel, a road-force check or a bent rim -- returns 21 complaints in the whole set. THIS IS NOT A CAUSE RANKING. It counts which acceleration shakes end in a federal safety complaint. Owners file when a repair is expensive, repeated or refused; almost nobody files after a $25 balance fixed it. Read the table as a map of which shakes become disputes. MANUFACTURER BULLETINS (589 distinct documents) Rows whose Summary contains shudder, vibrat*, shake or judder AND an acceleration marker (accelerat*, light throttle, under load, tip-in, take off, launch, torque converter, lock-up, TCC), all case-insensitive. The flat file carries one row per make/model/year; collapsing to distinct NHTSA document IDs gives 589 documents, after removing the heavy-truck and engine-maker rows (Cummins, Detroit Diesel, Allison, Freightliner and the rest of the exclusion list above). Of the 589, by full summary text: names the torque converter, lock-up clutch or TCC 220 37.4% names a driveshaft, half shaft, axle shaft, U-joint 53 9.0% mentions software, reprogramming, a calibration 177 30.1% names a misfire, spark plug or ignition coil 22 3.7% names an engine, motor or transmission mount 13 2.2% names a tire 12 2.0% names a balance, road-force check or bent wheel 4 0.7% Counts by make (top): Honda 78, Acura 74, Chevrolet 56, Toyota 46, GMC 35, Ford 32, Cadillac 31, Ram 25, BMW 20, Mazda 20, Nissan 19, Land Rover 18, Dodge 13, Jeep 12. "Mentions software" counts the word, not the outcome. RECALL CAMPAIGNS (22) Campaigns whose defect or consequence text contains vibrat*, shudder or shak* AND an acceleration or driveline marker (accelerat*, under load, throttle, torque converter, half shaft, axle, driveshaft, propeller, engine mount, motor mount), case-insensitive. 22 distinct campaigns. Summing each campaign's potentially-affected count once gives 139,505 campaign-vehicles. Most of the 22 are trailer, bus and heavy-truck campaigns about axle spindle nuts and U-joint bolts; the two that cover ordinary passenger vehicles are 20V-788 (Ford Explorer and Lincoln Aviator, 1,405 vehicles, right-hand motor mount fasteners) and 14V-341 (Cadillac CTS, 16,932 vehicles, rear driveshaft vibration triggering the roll-over sensor). For contrast, the same style of search run for stalling on the same file returns 556 campaigns. Acceleration vibration is almost never a recall subject; it is a bulletin and warranty-extension subject. PRIMARY DOCUMENTS OPENED FOR THE ARTICLE (PDFs fetched from static.nhtsa.gov/odi/tsbs/, all returned HTTP 200 on 18 September 2026) Ford Customer Satisfaction Program 25N09, 17 September 2025 (11023857) GM PIP5337H, December 2016 (10111732) Toyota T-SB-0312-17, 13 November 2017 (10146516) Stellantis 03-007-16, 30 November 2016 (10121893) Mazda 03-001/18, 5 June 2018 (10143961) Mazda 01-007/25, November 2025 (11026092) GM 24-NA-274, January 2025 (11013094) Ford CSP 14M02 Supplement 7, 4 May 2023 Ford TSB 23-2251, 1 September 2023 (10241524) -------------------------------------------------------------------------- LIMITATIONS -- READ BEFORE QUOTING ANY NUMBER -------------------------------------------------------------------------- 1. Complaints are self-reported, unverified and voluntary. They are not failure rates and cannot be converted into one: NHTSA publishes no denominator, so nobody knows how many of each model are on the road. 2. Counts favour high-volume models and models whose problem has been publicized. Nothing here ranks reliability, and no honest reading of complaint counts can. 3. The keyword method carries roughly one mis-sort in ten (see SAMPLE READING). Counts under about 30 should be read as "a handful of owners". 4. Towed, crash, fire, mileage and the mph bands are the owner's own entries. Mileage is blank or unusable in 42.5% of this set. 5. The crash share here (1.14%) is lower than the all-complaint baseline (4.37%) because the baseline is dominated by airbag and seat-belt complaints filed BECAUSE of a crash. It is not evidence that an acceleration shake is harmless. 6. Recall "potentially affected" is the manufacturer's estimate of vehicles in the population, not of vehicles that have the defect. 7. Bulletin counts are counts of documents, not of vehicles or repairs. A manufacturer that writes many short bulletins will out-count one that writes few long ones. -------------------------------------------------------------------------- FILES -------------------------------------------------------------------------- shake-complaints-by-make.csv make, complaints, median miles, miles reported, percent towed, percent crash, percent naming a transmission or torque converter. shake-top-models.csv make, model, complaints, median miles, miles reported, share at or under 36,000 miles, top model year, percent naming a transmission. 60 most-reported models. shake-what-owners-name.csv part family, complaints naming it, share of the 4,463. shake-speed-bands-owners-name.csv band group, mentions, share of the 389 bands owners wrote down. acceleration-vibration-bulletins.csv 589 rows: NHTSA document ID, bulletin number, communication date, make, model years covered, component, whether the summary names the torque converter, whether it mentions software, and the first 200 characters of the summary. No personal data is published: no VIN, no city, no dealer name, no owner free text beyond short quoted snippets in the article itself.