PulsCar analysis of NHTSA public data, September 2026 Shaking when braking: what 2,799 owner complaints and 533 manufacturer bulletins say 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 received 1 Jan 2015 to 8 Sep 2026. NHTSA manufacturer communication flat files TSBS_RECEIVED_2015-2019, _2020-2024, _2025-2026 -- https://static.nhtsa.gov/odi/ffdd/tsbs/ NHTSA recall flat file FLAT_RCL.zip, investigation flat file FLAT_INV.zip. -------------------------------------------------------------------------- CASE SENSITIVITY -- READ BEFORE REUSING ANY QUERY -------------------------------------------------------------------------- NHTSA complaint narratives are ALL CAPS through the end of 2020 and mixed case from 2021 onward. Measured on the file: of 58,465 complaints received in 2020, 58,386 are identical to their own uppercase form; of 62,775 received in 2023, only 939 are. An uppercase-only regular expression, or an uppercase SQLite GLOB, silently drops most complaints filed after 2020 while still returning thousands of rows -- a failure that looks like a result. That matters more here than on most topics, because the single largest cluster in this file is a 2023-2025 model-year vehicle. An uppercase-only query would have missed it entirely. Every query behind this page is case-insensitive: Python regexes compiled with re.IGNORECASE, SQL text matching with LIKE over lower(), which is case-insensitive in SQLite for ASCII. -------------------------------------------------------------------------- WHICH COMPLAINTS COUNT (the brake-shudder set: 2,799) -------------------------------------------------------------------------- Stage 1 -- SQL prefilter: narratives containing "brak" AND one of shak, vibrat, shudder, judder, puls, wobbl, tremb, chatter. 6,297 complaints. Stage 2 -- whole-narrative exclusions: (a) braking STOPS the shake. "Death wobble" narratives, and any phrasing where applying the brakes ends or cures a shake or wobble, are the inverse symptom and are dropped. (b) "shaken up" (a person, not a car) and "phantom braking" narratives. Stage 3 -- sentence test: split at . ; ! ? and keep the complaint if any sentence puts a SHAKE word within 70 characters of a BRAKE word, in either order, with per-sentence exclusions for engine-idle shake (idle, rpm, stall, in gear, foot on the brake, motor mount). SHAKE: \bshake|shakes|shaking|shaky\b, vibrat*, shudder*, judder*, puls*, wobbl*, trembl*, chatter*, jitter*, shimmy* BRAKE: \bbrake|brakes|braked|braking\b, slow* down, stopping, come to a stop, deceler*, brake pedal Word boundaries are load-bearing and were added only after reading samples. Without \b, "puls*" matched the word PROPULSION, pulling in loss-of-power narratives that have nothing to do with braking. The same class of error, in the other direction, is the uppercase trap above. Result: 2,799 complaints. -------------------------------------------------------------------------- SAMPLE READING (the false-positive rate we found) -------------------------------------------------------------------------- Three rounds of 30 random complaints, read in full, in three versions of the query. Round 1 (3,318 rows) surfaced death-wobble narratives where braking stops the shake, and engine-idle shake with a foot on the brake. Round 2 (3,060 rows) surfaced the PROPULSION bug and "shaken up". Round 3, on the final 2,799, found four misfits in thirty: a third-brake-light product, a cruise-control complaint, and two transmission complaints. Measured false-positive rate: about 4 in 30, roughly 13%. Every count in this file should be read with that margin. -------------------------------------------------------------------------- STRUCTURED FIELDS AND THE BASELINE -------------------------------------------------------------------------- All 777,040 complaints: towed 5.11%, crash 4.37%, fire 1.82%, median mileage 70,000 (392,935 usable mileages). The 2,799 brake-shudder complaints: towed 2.89%, crash 4.39%, fire 0.39%, 70 injuries, 0 deaths, median mileage 49,000 (1,642 usable mileages, first quartile 18,375, third quartile 98,000). The crash figure is the one to notice: 4.39% against a 4.37% baseline. Unlike most comfort-and-noise symptoms, a brake-shudder complaint is as likely as the average complaint to involve a crash. Mileage distribution, on the 1,642 rows with a usable mileage: at or under 12,000 miles 318 19.4% at or under 20,000 miles 454 27.6% at or under 36,000 miles 666 40.6% at or under 60,000 miles 922 56.2% -------------------------------------------------------------------------- THE ABS SPLIT (brake-shudder-abs-subset-vs-rest.csv) -------------------------------------------------------------------------- 449 of the 2,799 narratives (16.04%) mention ABS or anti-lock braking. That subset behaves differently from the rest of the file on every measure: ABS mentioned everything else complaints 449 2,350 median mileage 88,000 45,000 towed 4.68% 2.55% crash 9.58% 3.40% injuries 16 54 mentions a bump or uneven road 17.15% 9.70% mentions low speed 20.27% 10.26% mentions a rotor 7.80% 23.62% says it could not stop 4.23% 1.74% Read the last two rows together: the ABS subset is three times less likely to name a rotor and more than twice as likely to say the car would not stop. This is a different symptom wearing the same words, and its crash share is nearly triple the rest of the file's. We are not claiming ABS causes crashes. Selection is at work in both directions -- people are likelier to file, and to mention ABS, after a frightening stop -- and mentioning ABS is not the same as ABS malfunctioning. What we are claiming is narrower and checkable: within one complaint file, using one query, the ABS-mentioning subset carries a materially higher crash share than the rest. -------------------------------------------------------------------------- THE PALISADE CLUSTER -------------------------------------------------------------------------- The largest single nameplate in the file is the Hyundai Palisade: 134 complaints, 74 of model year 2024 and 32 of 2023, median reported mileage 1,750. Within those 134: mention a bump, pothole, railroad crossing or uneven surface 97 72.4% mention ABS or anti-lock 64 47.8% mention low or slow speed 52 38.8% say the car could not stop or kept going 15 11.2% mention a rotor 0 0.0% Zero of 134 name a rotor. Hyundai's own Technical Service Bulletin 25-01-070H, "Unintended ABS Activation Software Update (Service Campaign 9C8)," November 2025, covers 2023-2025 Palisade built 13 May 2022 to 16 June 2025 and describes the same symptom: vibration in the brake pedal and/or steering wheel and noises from the ABS when braking over uneven road surfaces or bumps, with the note that "the brake system remains functional under these conditions." The remedy is an ABS software update. -------------------------------------------------------------------------- MANUFACTURER BULLETINS -------------------------------------------------------------------------- Manufacturer communications whose summary contains a brake word and a vibration word within 100 characters, counted DISTINCT by NHTSA record id: 545, or 533 after removing heavy-truck and bus manufacturers by name. rotor named 57 10.69% hub, flange, corrosion, rust or cleaning 57 10.69% software or calibration 35 6.57% resurfacing, machining or an on-car lathe 21 3.94% says the condition is normal or needs no repair 21 3.94% disc thickness variation, DTV or runout 20 3.75% ABS or electronic stability 16 3.00% tire, wheel balance or road force 14 2.63% torque or lug nut 10 1.88% Honda and Acura together account for 152 of the 533 (28.5%), which ranks documentation, not reliability. -------------------------------------------------------------------------- LIMITATIONS -------------------------------------------------------------------------- 1. About 13% of rows are mis-sorts (see the sample reading). 2. Complaints are self-reported and unverified. 3. There is no denominator. Nothing here ranks any vehicle's reliability, and the Palisade cluster is a documented service-campaign population, which is exactly the kind of thing that drives owners to the complaint form. 4. Counts of what owners "name" measure vocabulary, not verified causes. 5. Mileage is owner-entered and missing in 1,157 of 2,799 rows. 6. The complaint file ends 8 September 2026. Contact: support@pulscar.io