PULSCAR ANALYSIS — VIBRATION AT IDLE OR AT A STANDSTILL (NOT A MISFIRE) IN NHTSA'S PUBLIC FILES PulsCar analysis of NHTSA public data, September 2026. License: CC BY 4.0. Attribution: "PulsCar analysis of NHTSA public data, September 2026." Published alongside https://www.pulscar.io/blog/car-vibration-at-idle ------------------------------------------------------------------------------ Prepared September 2026. Source data: NHTSA public flat files (complaints received 2015 to 21 September 2026, 780,071 records; manufacturer communications) and documents downloaded from static.nhtsa.gov. 1. Complaints (2,918, all read in full, two readers) Searches: a first search (vibration word within one sentence of an idle or standstill phrase) found 1,806; a second, wider search for spellings and phrases the first missed found 689; a third, independent completeness audit found 100; a fourth search, wave 2, added 323 (below). Details in codebook.txt, "Pool history". The found_by column says which search found each complaint. Coding: the codebook (codebook.txt) was written, dated and hashed before any complaint was coded (2026-09-26T17:01:38Z). Two independent classification passes coded every complaint separately, each from the codebook and its own batch only, without seeing the other's codes: one category (N not a vibration at a standstill; M mounts; D cylinder deactivation; A A/C or accessory load; T transmission, converter or clutch; S stop-start or hybrid engine start; I engine running rough; X another named cause; U no cause stated) and four flags (in_gear, ac_related, light_or_code, fixed). The two passes ran independently of each other. Agreement, first three searches (2,595): category 1,883, Cohen's kappa 0.63; in_gear 2,130, kappa 0.45; ac_related 2,572, kappa 0.74; light_or_code 2,340, kappa 0.76; fixed 2,550, kappa 0.52. Every disagreement was read in full and resolved under the written adjudication rules R1-R18 and R20 (codebook.txt; R21 added in wave 2); a consistency pass then re-read agreed codes flagged by script as contradicting the rules and corrected 41. Wave 2 (323 complaints): the completeness audit's family "vib_any_standstill" (any vibration word, including misspellings and verbs such as rumble, rough, chug or jolt, within one sentence of a new standstill phrase: drive-thru, warming up, cold start, came to a stop, parked with the engine running, foot on / holding the brake, waiting in line) matched 417 complaints, 352 of them outside the pool the audit started from; 29 of those had already been coded in the third search, so wave 2 coded the other 323. Before coding, a short addendum (codebook_wave2, dated and hashed 2026-09-28T10:06:15Z, reproduced in codebook.txt) restated the pool rule and clarified one flag, in_gear, on which the readers of the first three searches had agreed only moderately; nothing else in the codebook changed. Wave-2 agreement: category 271 of 323, kappa 0.77; in_gear 315, kappa 0.85; ac_related 322, kappa 0.67; light_or_code 298, kappa 0.80; fixed 315, kappa 0.70. Most wave-2 complaints turned out not to describe a shake at a standstill (260 of 323 are N), so its agreement is not directly comparable with the first searches; it shows that the in_gear clarification was applied consistently, not that it would lift agreement by the same amount on the earlier set. The 72 wave-2 disagreements were resolved under the existing rules (rule R9 was applied to variable cam timing parts and head gaskets; R21, a shake while braking to a stop is N, was written down afterwards from the codebook's own "braking judder" example), and a consistency pass re-read every agreed wave-2 code: 81 agreed relevant codes whose narrative has no word for a felt vibration, shake, shudder, shimmy, jerk, rock or jolt were changed to N under R20, and 16 more were changed under R6, R7, R8, R11, R12, R15, R16, R18 and R20. Whole set (2,918): category agreement 2,154, kappa 0.65; in_gear 2,445, kappa 0.47. The complaints file carries both readers' original codes (coder1_*, coder2_*), the final code, the resolution and the rule applied. Final: not a vibration at a standstill (N) 752; of the 2,166 that describe one: no cause stated (U) 1,102, engine running rough (I) 600, another named cause (X) 224 (156 of them mention the steering), transmission, converter or clutch (T) 158, mounts (M) 49, stop-start or hybrid (S) 18, cylinder deactivation (D) 9, A/C or accessory by a diagnosis (A) 6. In gear at a stop 383; A/C brings it on or makes it worse 37; check engine light or a code 563; the shake stopped after a repair 39. Wave 2 added 63 relevant complaints: 28 U, 27 I (most a shake at a cold start with a cam-timing, head-gasket, coolant or injector finding), 4 X, 4 T; 11 of them in gear at a stop. Makes among the 2,166: Ford 478, Honda 337, Chevrolet 306, Hyundai 120, Kia 118, Nissan 110, Toyota 87, Jeep 71, GMC 70, Chrysler 56, and others. Median model year 2015; median reported mileage 66,729 (1,387 complaints give one). By year received: 2015 274, 2016 175, 2017 160, 2018 202, 2019 269, 2020 190, 2021 133, 2022 120, 2023 125, 2024 156, 2025 202, 2026 (to 21 September) 160. Figures count complaints, not vehicles. 2. Owner-reported amounts (212 figures from 163 complaints) Every dollar or "k" figure in the relevant complaints was extracted by script and decided under one rule: a figure is included when the owner gives it for a diagnosis, repair, part, quote or bill aimed at the vibration or at the fault the complaint blames for it, whether it worked or not; vehicle prices, loans, towing, rentals, warranty terms, mileages and repairs for unrelated faults are left out. Status is paid only when the owner says so; otherwise quoted, billed, offered or not stated. 3. Manufacturer documents (494 listed; 213 read in full, 80 compared sentence by sentence) First and widened searches of the NHTSA index summaries for a vibration, shake, shudder, boom or roughness at idle, at a stop or in gear, and for mounts, cylinder deactivation and accessory load, gave 291 records; the completeness audit found 199 more. Of these, the 32 on this article's topic were downloaded and read (28 PDFs; 4 have no PDF on NHTSA), and 167 about engine rough idle or misfire are listed by summary only, left to the misfire and rough-idle guides. Near-identical revisions (5-word shingle Jaccard of 0.6 or more) were read as sentence differences against the version read in full; byte-identical copies are marked as duplicates; page images were read where the text layer was empty or scrambled. Of the 213 read in full: 147 direct, 33 related, 1 lists the topic among others, the rest not relevant. Every verbatim quote in the article was checked by script against the document or complaint text. Coverage status in September 2026 is our computation from the terms printed in each document, not a statement by the manufacturer. No recall filings were read for this study. Limits: complaints over-represent unresolved and frightening cases; owners report the diagnosis as they understood it; the searches cannot see narratives that describe the shake in other words, or bulletins whose index summary omits the symptom. The completeness audit's remaining low-yield families were not read: other verbs next to an idle phrase (2,730 complaints outside the pool, about 7.5% on topic in a sample of 40, roughly 202 on-topic), cause words near any vibration word (1,541, about 2.5%, roughly 38) and cross-sentence proximity within 400 characters (6,326, about 15%, roughly 943) — about 10,600 complaints and an estimated 1,180 on-topic ones, from samples of 40, not verified. Nothing here is a failure rate.