PULSCAR ANALYSIS OF NHTSA PUBLIC DATA — TRANSMISSION FAILURE AND REPAIR COST Published September 2026. License: CC BY 4.0 (credit "PulsCar analysis of NHTSA public data, September 2026"). ------------------------------------------------------------------------------- CORRECTION, 18 September 2026 ------------------------------------------------------------------------------- Step (c) below — the failure-cue regex — was run against the narrative WITHOUT case folding, as the words "(regex, upper case)" recorded. NHTSA complaint narratives are upper case only through 2020; from 2021 on they are mixed case. The uppercase-only match therefore kept 846 complaints received in 2021 and 72, 49, 60, 92 and 78 in 2022-2026, against 2,200-3,700 a year before that: the published set was effectively a 2015-2021 file described as a 2015-2026 one. Steps (a) and (b) are SQL LIKE, which folds ASCII case, so they were never affected. Re-running step (c) case-insensitively — same word list, same order — returns 36,138 complaints instead of 17,842, and every complaint-derived figure on this page, in the article and in the three complaint CSVs has been recomputed from it. The recall, bulletin and investigation sections are unchanged: they query NHTSA's recall and bulletin text through SQL LIKE or an upper-cased copy of the text, and re-running them changes nothing. This correction is exact, not a reconstruction: re-running the documented query case-sensitively reproduces every previously published figure to the digit (17,842 complaints, 546 amounts, median $4,000), which is what identified the defect. What changed materially: complaint set 17,842 -> 36,138 towed 5.0% -> 6.9% (baseline 5.1%) -- the page used to say a transmission failure is towed at the baseline rate; it is towed somewhat more often crash 1.6% -> 1.3% fire 0.6% -> 0.4% median failure mileage 70,000 -> 75,000 owner-reported amounts 546 -> 1,240 (483 -> 1,098 complaints) median amount $4,000 -> $5,500; quartiles $3,000/$5,100 -> $3,800/$8,000 by make Ford 4,715, Nissan 3,181, Jeep 1,655 -> Ford 12,204, Nissan 4,473, Chevrolet 3,235 The 110 complaints read in full (section 4) were drawn BEFORE this correction, from the 12,508 cue-carrying complaints the old query returned rather than from today's 27,200. They have not been re-drawn or re-read; the article says so where it uses them. ------------------------------------------------------------------------------- 1. SOURCE FILES ------------------------------------------------------------------------------- All inputs are NHTSA's public ODI flat files, downloaded 17 September 2026 from https://static.nhtsa.gov/odi/ffdd/ (field definitions: cmpl/CMPL.txt, rcl/RCL.txt, ftsb/FLAT_TSBS.txt, inv/INV.txt) and loaded into a local SQLite database. complaints 777,040 rows, received 1 January 2015 - 8 September 2026 recalls one row per make/model/year per campaign (campaigns counted DISTINCT) tsbs manufacturer communications, one row per make/model/year investigations ODI investigations Complaints are self-reported by owners. They are NOT failure rates: there is no denominator (nobody publishes how many of each model are on the road), no verification of the owner's diagnosis, and no follow-up when a problem is later fixed. Nothing below should be read as "brand X is unreliable". ------------------------------------------------------------------------------- 2. THE COMPLAINT SET (36,138 complaints) ------------------------------------------------------------------------------- Query, in order: a) components LIKE '%POWER TRAIN%' -> 104,152 b) AND the narrative mentions the transmission: descr LIKE '%TRANSMISSION%' OR '% CVT%' OR 'CVT%' -> 47,023 c) AND the narrative contains a failure cue (regex, CASE-INSENSITIVE since 18 September 2026 -- see CORRECTION): SLIP | SHUDDER | JERK | HARD SHIFT | HARSH SHIFT | WILL NOT SHIFT | WON'T SHIFT | DOES NOT SHIFT | NOT SHIFT | FAIL | WENT OUT | GAVE OUT | REBUILD | REPLACE | LOSS OF POWER | LURCH | HESITAT | LIMP | STUCK IN GEAR | NO GEARS | NEUTRAL -> 36,138 Step (a) is what keeps the set honest: the component tag is assigned by NHTSA, not by our keywords, so a complaint about a squeaky seat that happens to use the word "transmission" is excluded before the text search runs. Headline figures computed on those 36,138: towed 2,483 6.9% (baseline: 39,739 of 777,040 = 5.1%) crash reported 467 1.3% (baseline: 33,940 = 4.4%) fire 154 0.4% (baseline: 14,121 = 1.8%) shop could not find it / "operating as designed" / no codes 1,184 3.3% (regex: COULD NOT DUPLICATE, CANNOT DUPLICATE, NO PROBLEM FOUND, NOTHING WRONG, OPERATING AS DESIGNED, NO FAULT FOUND, COULD NOT REPLICATE, UNABLE TO DUPLICATE, NORMAL OPERATION, NO CODES) software/reprogramming named 1,599 4.4% (REPROGRAM, SOFTWARE UPDATE, REFLASH, FLASH THE, TCM UPDATE, UPDATED THE SOFTWARE, CALIBRATION) Failure mileage: 21,724 of the 36,138 report a usable figure (we drop 0, blanks and anything >= 400,000). Median 75,000. 25th percentile 36,000, 75th percentile 110,000. 5,480 (25.2%) are at or under 36,000 miles; 8,486 (39.1%) at or under 60,000; 15,110 (69.6%) at or under 100,000. Note that the MILES field is blank far more often in the post-2020 half of the file, which is why the mileage n grows less than the set does. CVT subset: 2,123 complaints whose narrative contains the standalone word CVT. Median failure mileage 84,000 (n=1,234) against 75,000 for the rest (n=20,490). By make and by model: transmission-complaints-by-make.csv (makes with >= 10 complaints) and transmission-complaints-by-model.csv (models with >= 15), each with the median failure mileage and the three most common model years. ------------------------------------------------------------------------------- 3. WHAT OWNERS SAY THEY WERE QUOTED (1,240 dollar figures, 1,098 complaints) ------------------------------------------------------------------------------- Owners often write the number the shop gave them. We extracted every dollar amount in the 36,138 narratives with the regex \$\s?([0-9][0-9,]{2,7})(\.\d{1,2})? and kept an amount only when ALL of these held for the 120 characters on either side of it: - the window names the unit: TRANSMISSION | TRANSAXLE | TRANS | CVT - the window contains a work verb: REPLAC | REBUILD | NEW TRANSMISSION | NEW CVT | NEW TRANS - the window does NOT name a different big-ticket part: RADIATOR, TRANSFER CASE, BATTERY, HARNESS, ENGINE REPLAC, NEW ENGINE, TIRES, BRAKE, SUSPENSION, AIR BAG, BODY REPAIR, CATALYTIC, DIFFERENTIAL - the window is not about money other than repair: WORTH, OWE, LOAN, TRADE, PURCHASE, BOUGHT, VALUE, DEDUCTIBLE, DOWN PAYMENT, PER MONTH, MONTHLY, LEASE, RENTAL, REFUND, LAWSUIT, SETTLEMENT, MSRP, DIAGNOSTIC, DIAGNOSIS FEE - the amount is between $1,000 and $15,000 Result: 1,240 amounts across 1,098 complaints. Distribution: $1,000-1,999 82 6.6% $2,000-2,999 93 7.5% $3,000-3,999 146 11.8% $4,000-4,999 196 15.8% $5,000-5,999 152 12.3% $6,000-7,999 227 18.3% $8,000 and up 344 27.7% median $5,500 | 25th percentile $3,800 | 75th percentile $8,000 median odometer of the cars behind those numbers: 91,840 miles (only 486 of the 1,240 figures come from a complaint that reports an odometer) SAMPLE READING (required before publishing any keyword count). A first pass on the loose query — any dollar amount in a narrative containing a repair word — returned 1,252 amounts; reading 8 of them showed obvious contamination (a radiator, a transfer case, a battery, an engine harness), which is why the "different big-ticket part" exclusion list above exists. On the tightened set we drew 30 at random (seed 5) and read each in context. 29 of 30 were a real transmission repair or replacement figure; 1 was wrong (a Subaru complaint where $4,000 had been spent on axles and CV joints before the transmission was mentioned). Observed false-positive rate 3.3% (1/30), 95% binomial interval roughly 0.1%-17%. LIMITATIONS of these 1,240 figures, all material: - They mix dealer quotes, independent-shop quotes and bills actually paid. Only 375 windows name a dealer; only 46 clearly name an independent or chain transmission shop, which is too few to compare the two honestly, so we do not. - A few complaints report a part price and a labor price separately, so one job can appear as two amounts. - Some are ranges ("$3,600-$5,000"); we take the first number in the window. - People who are angry about a big number write to NHTSA. People quoted $600 for a solenoid usually do not. This set is therefore biased upward and is NOT a national average price. Use it as "what the expensive end of this market looks like when owners report it", not as an estimate. - Amounts are not inflation-adjusted; the complaints span 2015-2026. Published as transmission-repair-quotes.csv (complaint id, make, model, model year, failure mileage, amount). No free text, no VIN, no city, no dealer name. ------------------------------------------------------------------------------- 4. 110 COMPLAINTS READ IN FULL — WHAT THE SHOP FOUND ------------------------------------------------------------------------------- Counting words tells you what people complained about, not what the car needed. From the 12,508 complaints in the set whose narrative contains an outcome cue (DEALER, SHOP, MECHANIC, TECHNICIAN, REPLACED, REBUILT, DIAGNOS, TOLD ME, FOUND) we drew 110 at random (Python random.seed(2026)) and read every one, assigning each a single primary outcome. THAT POOL IS THE PRE-CORRECTION ONE: the same cue list over the corrected set returns 27,200 complaints, and the 110 were not re-drawn, so this section describes the 2015-2021 part of the file (see CORRECTION): whole transmission named as the answer (recommended or performed) 23 a named part or module instead of the unit 20 (valve body, clutch pack, TCM, lead frame, torque converter, CVT belt, speed sensor, oil cooler/adapter, filter, bushing) no cause found / "that is normal" / wrong part replaced 18 no repair outcome stated in the complaint 19 still unresolved after repeated visits, no single cause named 12 radiator-to-transmission coolant cross-contamination 6 software reprogramming named as the repair 5 off target (not a transmission repair at all) 5 fluid service named as the repair 2 Derived figures used in the article: - 27 of 110 (a part, software or fluid) versus 23 of 110 (a whole unit). - 30 of 110 (18 + 12) end with the owner still not knowing what is wrong. - Off-target rate of the complaint set as read: 5/110 = 4.5%. Read "unresolved" as a floor on frustration, not a repair-failure rate: a complaint is filed at a moment in time, often while the car is still at the shop, and nobody files a follow-up saying it was fixed. ------------------------------------------------------------------------------- 5. RECALLS (411 campaigns) ------------------------------------------------------------------------------- recalls WHERE component LIKE '%TRANSMISSION%', campaigns counted DISTINCT: 411 campaigns, report dates 14 May 2010 - 15 July 2026 26,426,053 vehicles potentially affected (sum of the maximum potaff per campaign) 316 campaigns reported since 1 January 2015, 19,794,463 vehicles 134 campaigns (13,046,539 vehicles) have a remedy containing REPROGRAM, SOFTWARE or FLASH — i.e. the fix is a software update Published as transmission-recalls.csv with a remedy_type column derived from the remedy text (software / inspect-replace / other). The largest campaigns in the file, each read individually: 22V413000 Ford, 2,925,968 units, transmission gear position indication 11V395000 Honda 2005-2010 Accord/CR-V/Element, 1,512,107, secondary shaft bearing; remedy: update the transmission control module software 26V237000 Ford 2015-2017 F-150, 1,392,935, loss of signal between the transmission range sensor and the PCM -> unexpected downshift into second gear; remedy: PCM software update (Ford number 26S28) 19V075000 Ford 2011-2013 F-150, 1,263,051, unexpected downshift into first gear; remedy: reprogram the powertrain control module (Ford 19S07) 24V797000 GM 2020-2022 full-size trucks/SUVs (diesel), 463,295, transmission control valve failure -> rear wheel lock-up; remedy: new TCM software plus a special coverage program paying for transmissions the software identifies as having the defective valve Caveat: the potaff column is "potentially affected", a population, not a count of failures, and vehicles appear in more than one campaign, so the 26.4 million total is not 26.4 million distinct cars. ------------------------------------------------------------------------------- 6. MANUFACTURER COMMUNICATIONS ------------------------------------------------------------------------------- tsbs WHERE component LIKE '%TRANSMISSION%': 4,907 distinct documents. Restricting to summaries naming the symptoms this article is about — SHUDDER, SLIP, HARSH SHIFT, FLARE or DELAYED ENGAGEMENT — leaves 421 distinct documents. By make: Chevrolet 75, GMC 62, Cadillac 53, Toyota 40, Ford 22, Acura 18, Honda 17, Nissan 13, Buick 13, Lincoln 11, Jeep 11, Dodge 8. Of the 421, 154 (37%) name reprogramming, software, calibration or a flash; 52 name a fluid or flush procedure. Document counts are paperwork counts, not problem counts: many are revisions (NTB17-034, NTB17-034a through NTB17-034d are five rows for one bulletin), so the by-make numbers describe how a manufacturer files, not how often its cars break. Separately, tsbs whose summary contains WARRANTY EXTENSION, EXTENDED WARRANTY, CUSTOMER SATISFACTION or SPECIAL COVERAGE on a transmission component: 174 distinct documents (Chevrolet 35, GMC 18, Jeep 15, Nissan 14, Honda 11, Ford 9). These are free or discounted repairs outside any recall. Documents opened and read in full for this article: - Nissan NTB17-034, 26 April 2017, 108 pages, "CVT JUDDER AND/OR P0746/P0965 STORED", downloaded from static.nhtsa.gov/odi/tsbs/2017/MC-10109251-9999.pdf. Its repair overview is a three-way decision tree: replace only the valve body; replace the sub-assembly and valve body; or replace the CVT. - NTB14-118c, 13 October 2017 (MC-10123345-9999.pdf), superseded by NTB17-034. ------------------------------------------------------------------------------- 7. INVESTIGATIONS ------------------------------------------------------------------------------- investigations WHERE component LIKE '%TRANSMISSION%': 78 distinct actions. The current one named in the article is EA26001 (Ford F-150 2015-2016, "Unintended Transmission Downshift and Rear Wheel Lock-up"), opened 30 January 2026 as an upgrade of PE25002 (opened 21 March 2025) and still open at the time of writing. ------------------------------------------------------------------------------- 8. WHAT WE DID NOT DO ------------------------------------------------------------------------------- - No normalization by sales or vehicles on the road: we do not have that denominator, so no "most/least reliable" claim appears anywhere. - No re-reading of the 110-complaint sample after the 18 September 2026 correction: re-drawing them would mean reading 110 fresh narratives, which we did not do in this pass, so the sample is labelled as pre-correction wherever it is used instead of being quietly restated. - No price re-verification: the repair-price ranges in the article are prior published estimates and are labelled as such. Consumer estimator sites block automated access, and we do not work around that. - No attempt to match a complaint to a specific transmission model (4L60E, DPS6, ZF8) beyond what the owner wrote.