PULSCAR ANALYSIS — TRANSMISSION FLUID CHANGE, FLUSH, TOP-UP AND LIFETIME-FLUID COMPLAINTS AND MANUFACTURER DOCUMENTS IN NHTSA'S PUBLIC FILES PulsCar analysis of NHTSA public data, October 2026. License: CC BY 4.0 Attribution: "PulsCar analysis of NHTSA public data, October 2026." Published alongside https://www.pulscar.io/blog/transmission-fluid-change-cost ------------------------------------------------------------------------------ Prepared October 2026. Source data: NHTSA public flat files (complaints received 1 January 2015 to 4 October 2026, 782,908 records; manufacturer communications index) and documents downloaded from static.nhtsa.gov, one at a time with pauses. 1. Complaints (3,280, all read in full, two readers) First search: case-insensitive regular expressions over every narrative that names a transmission (transmission and common misspellings, trans, tranny, transaxle, CVT, DCT, DSG, gearbox): a fluid word near an action word (change, flush, drain, refill, top off, add, fill, replace, service, exchange, wrong, overfill, low, level, dipstick, lifetime, sealed); a flush, exchange or drain-and-fill of the transmission; "transmission service(d)" or "serviced the transmission" outside a dashboard-message context; lifetime, sealed, maintenance-free or no-dipstick wording; a flush near a transmission word; and a fluid service phrase not preceded by brake, coolant, steering, washer, radiator, engine, differential, case, clutch, hydraulic, wiper or windshield. Exact patterns are in the appendix of codebook.txt. An early version that required a fluid word found 2,878; it dropped "transmission serviced" narratives and was replaced before coding. Final first search: 3,002 complaints. Completeness audit: before writing, an independent completeness search of the same file for what the first search missed — a transmission "serviced" or "maintained" without the word fluid, flushes phrased away from "transmission fluid", level and amount wording ("too much", "a quart low", "overfilled the transmission"), incidental level checks, "no way to check", lifetime wording outside the original windows, and misspellings — and listed 278 more complaints, each read before listing. Patterns with low precision (L4, Q5, S1, R15, S3) were sampled, and broad scouting patterns (M3, P1, P2, P4, P5, Q1, Q3) were not read in full; their unread remainders are not in the pool. The found_by column says which search found each complaint, and matched_by which pattern. Coding: the codebook (codebook.txt) was written, dated and hashed before any complaint was read, and an addendum for the 278 audit additions was hashed before they were read. Two independent classification passes coded all 3,280 separately, each from the codebook only, without seeing the other's codes, each seeing only the codebook and its batch of complaints (14 batches of up to 215 for the first search, 2 of 139 for the audit). The Mac running the work slept during batches 07 and 08; both passes on those batches were resumed (same codebook, same independence) from the last complaint written, and every complaint in those batches has two codes. One category (N not about a transmission fluid service or fluid question on the owner's vehicle; W wrong fluid or a service done wrong; F flush or machine exchange; C change, drain-and-fill, "transmission service", fluid and filter; T top-up or level check; L lifetime or sealed fluid, no dipstick, no interval; R a fluid service recommended, quoted, prescribed or refused) and four flags (fix_attempt, helped, worse_after, unit_replaced). Agreement, first search (3,002): category 2,939 (97.9%), Cohen's kappa 0.973; relevance 0.957; fix_attempt 0.937; helped 0.910; worse_after 0.927; unit_replaced 0.962. Agreement, audit additions (278): category 260 (93.5%), kappa 0.920; relevance 0.837; fix_attempt 0.892; helped 0.919; worse_after 0.818; unit_replaced 0.964. Every disagreement (253 complaints) was read in full and settled under written rules R1-R24 (codebook.txt), and a consistency pass corrected 37 agreed codes that contradicted those rules; the resolution and rule_applied columns record both. Final: not relevant (N) 561; of the 2,719 relevant: C 1,004, F 731, R 361, T 356, L 188, W 79. Fix attempts 1,788; helped 289; worse after the service 241; transmission replaced or replacement recommended 628. Routine (not a fix attempt) flushes 162, of which 50 worse after; routine changes 371, of which 63 worse after. Makes among the 2,719: Ford 539, Chevrolet 382, Honda 363, Nissan 360, GMC 183, Toyota 148, Jeep 107, Dodge 83, Hyundai 82, Acura 56, Subaru 48, Kia 47, and others. Models: Ford F-150 162, Chevrolet Silverado 1500 114, Honda Pilot 110, Ford Escape 100, GMC Sierra 1500 88, Honda Odyssey 66, Nissan Pathfinder 64, Chevrolet Colorado 63, Nissan Altima 62, Nissan Rogue 61. Median model year 2016; median reported mileage 85,000 (1,252 complaints give a mileage from 1,000 to 400,000). By year received: 2015 176, 2016 162, 2017 143, 2018 175, 2019 275, 2020 188, 2021 170, 2022 185, 2023 216, 2024 270, 2025 387, 2026 372 (2026 to 4 October). Mention counts (by script, not coded; case-insensitive in the relevant narratives): dealer 1,648, shudder 398, CVT 206, metal shavings, flakes or particles 76, burnt 57; shudder counts the misspelling "shutter". Figures count complaints, not vehicles. 2. Owner-reported amounts (522 figures in 410 complaints) Every dollar figure in the relevant complaints, plus written amounts without a $ sign found by a second script, was decided under one rule: kept if the owner ties it to transmission work on their own vehicle (diagnosis, a fluid service, a level check, a part, module or software, a repair, a replacement or rebuild) and the narrative says or implies paid, billed, quoted or offered; excluded (646 numbers, listed with reasons in our working files): mileages and other numbers that are not prices, vehicle prices and values, loans, rentals and towing, warranty mileages and intervals, repeated figures, and repairs unrelated to the transmission. 310 figures are quotes, 123 described as paid, 29 billed, 5 offered, 55 not stated. Each figure was checked against the narrative and sorted into one group (group column) by the job the narrative ties it to. Fluid service (change, flush, exchange or top-up): 104 figures, median $400; paid or billed 57, median $358; quoted 36, median $500. Shop-estimate prices: the estimator site we tried (RepairPal) returned HTTP 403 on 7 October 2026; it was not retried by another route, and no shop-estimate ranges are used. 3. Manufacturer documents (546 distinct, read) First search: every manufacturer communication whose NHTSA index summary names transmission fluid near a service word, a flush near a transmission word, a maintenance schedule with a transmission item, or lifetime or sealed fluid (538 records). Completeness audit: the independent audit added 94 records (fluid application guides, capacity charts, maintenance schedule notes, fluid-level videos and others whose summaries named the fluid without a service verb). Of the 632 records, 19 have no PDF on NHTSA; the other 626 files, after removing 80 byte-identical copies (md5), are 546 distinct documents: 462 from the first search, 84 from the audit. Near-identical revisions (5-word shingle Jaccard 0.6 or more) were read sentence by sentence against a copy read in full (168 documents); 378 were read in full; page images were read where the text layer was empty or scrambled. Of the 546: 276 direct (fluid type, level procedure, interval, flush instruction or fluid service as the repair), 111 with a fluid step inside another repair, 107 related, 52 not relevant. Recalls: NHTSA's recall index was searched for campaigns whose defect or remedy involves transmission fluid level, overfill, fluid expelled, or a drain-and-fill; 5 campaigns (21V539, 22V220, 22V225, 22V835, 20V043), 50 files, were read. None of the five has a Part 573 report among its published files; one file filed under 20V043 (RCRN-20V043-1000) belongs to a different campaign. Every verbatim quote in the article was checked by script against the document or complaint text; quotes from image-only pages were checked against a transcription of the page image. Coverage status in September 2026 is our computation from the terms printed in each document, not a statement by the manufacturer. Limits: complaints over-represent bad outcomes and expensive repairs; owners report the service and its result as they understood them; a fluid service named in a complaint is not proof of what the shop did; searches of NHTSA's index cannot see bulletins whose summary is blank or whose fluid content appears only inside a PDF. Nothing here is a failure rate or a price survey.