PulsCar study: owner complaints, recalls and bulletins about a car that will not accelerate
PulsCar analysis of NHTSA public data, September 2026. License: CC BY 4.0, credit "PulsCar analysis of NHTSA public data" with a link to https://www.pulscar.io/blog/car-wont-accelerate

SOURCE FILES (downloaded from https://static.nhtsa.gov/odi/ffdd/ in September 2026)
- Complaints: COMPLAINTS_RECEIVED_2015-2019.zip, 2020-2024.zip, 2025-2026.zip. One row per ODINO (NHTSA report number), vehicle complaints only. Received (LDATE) 2015-01-01 to 2026-09-08. Unique complaints: 777,040.
- Recalls: FLAT_RCL.zip, one row per make/model/year per campaign; campaigns counted as DISTINCT campaign numbers.
- Manufacturer communications (bulletins): TSBS_RECEIVED_2015-2019, 2020-2024, 2025-2026; one row per make/model/year per document; documents counted as DISTINCT NHTSA document ids.
- Investigations: FLAT_INV.zip.
Counts run on 17 September 2026.

1. THE ACCELERATION-COMPLAINT SET (25,949 complaints)
Step 1, candidate pull (full-text index over the owner description, CDESCR): any of
  accelerate, accelerating, acceleration, "reduced engine power", "reduced power", "limp mode", "limp home", "loss of power", "lost power", "loses power", "losing power", "lose power", "no power"  -> 77,118 rows.
Step 2, exclusions (Python re, case-insensitive; a complaint matching any of these was dropped):
  power (steering|brake|window|lock|seat|door|outlet|liftgate|tailgate|mirror|outage|cord|button|source|supply|to the (radio|dash|lights|car's electronics|infotainment|screen|display|battery)) | (electrical|electric|battery|12v|12-volt) power | loss of power steering | lost power steering | sudden unintended acceleration | unintended acceleration | accelerat(ed|es|ing) on its own | (self|un)-?(intended|commanded) accelerat | accelerated (by itself|without) | surge
Step 3, the symptom itself had to be described (CORE regex, case-insensitive):
  \b(will|would|does|did|could|can|do)\s?(n[o']?t|not)\s+accelerate | won't accelerate | wont accelerate | no acceleration | loss of acceleration | lost acceleration | (fail(s|ed|ing)?|unable|refus(es|ed|ing)?|inability|failure)\s+to accelerate | (not|stopped|quit|barely|hardly)\s+accelerat(e|ing) | reduced (engine )?power | limp[- ]?(home )?mode | sluggish | hesitat(es|ed|ion|ing) | (lack|loss) of (engine )?power (when|while|on|under|during|at|going|trying|accelerat) | los(t|es|ing|e) (all |engine )?power (when|while|on|under|during|at|going|trying|accelerat)
  Generic "loss of power" / "lost power" with no driving context was NOT enough on its own (that broader set, 39,800 rows, is not used in the article).
Hand audit: 30 random complaints from the final set read in full: 28 described a car that would not accelerate, lost power while driving, hesitated, or entered limp/reduced-power mode; 2 were marginal (an electrical laundry list that mentioned hesitation once; a CVT failure described as a stall). An earlier audit of 40 from the broader 39,800-row set found 36 on topic, 2 stalls and 2 that only mentioned acceleration hypothetically, which is why the CORE regex was tightened.

2. STATISTICS
- Mileage: FAILURE MILEAGE as reported; only values >0 and <400,000 used (14,185 of 25,949 complaints state one). Median 70,000; quartiles 32,000 / 105,000. All-complaints baseline (392,935 with mileage): median 70,000; quartiles 33,000 / 110,000.
- Vehicle age = year received minus model year, model years 1990-2027. Median 5 years; quartiles 3 / 8.
- Outcome flags as filed by the owner: towed 5.2% (baseline 5.1%), crash 1.0% (baseline 4.4%), fire 0.8% (baseline 1.8%), any injury 0.52% (baseline 2.82%), 3 reports list a death. The baseline is every vehicle complaint in the file, including air-bag and crash reports, so it is not a "safe" comparison group.
- Share by make and model = acceleration complaints / all complaints for that make or model in the same file and period (complaints_by_make.csv, complaints_by_model.csv; makes with >=100 and models with >=100 acceleration complaints). Complaints are self-reported, unverified and not normalised to vehicles on the road; the tables do not measure failure rates.
- Sub-phrases inside the set (share of 25,949): "reduced engine power" 2,025 (7.8%; 64% of them are Chevrolet, GMC, Cadillac or Buick); "limp mode" 4,843 (18.7%); hesitation 4,441 (17.1%); "wrench" light 744 (2.9%; 687 of them Ford); "CVT" 645 (2.5%; 488 Nissan or Infiniti); highway/freeway/interstate/merging 9,290 (35.8%); intermittent or cleared by a restart 2,729 (10.5%).
- No warning light: regex  no (check engine|warning|dash|dashboard|engine)( warning)?( light| lights| lamp| message| messages| indicator) | without (any )?(prior )?warning | no warning (light|lamp|message|indicator)s? | no lights? (on the dash|came on|illuminated) | no (dtc|code)s?  -> 3,529 (13.6%).
- Shop could not find / could not duplicate: the SHOP_COULD_NOT_FIND regex from our car-noise study (could not duplicate|replicate|reproduce|find the problem..., no problem found, nothing found, operating as designed, normal) -> 1,663 (6.4%).

3. PARTS NAMED (parts_named_in_complaints.csv)
Column 2-3: complaints in the set whose text names the part or system (regex per row, e.g. transmission = \btransmission\b|\bcvt\b|torque converter|clutch pack|valve body|\btransaxle\b; throttle = throttle body|throttle position|throttle actuator|electronic throttle|accelerator pedal (position |assembly|sensor)|pedal (position )?sensor|APP sensor|TPS; catalytic = catalytic|cat converter, deliberately NOT "converter" alone, which matched "torque converter" in an earlier draft).
Column 4-5: the same count inside the "diagnosed" subset, 11,563 complaints whose text contains a diagnosis or repair word (diagnos|determined that|needs to be replaced|replaced the|was replaced|dealer/mechanic/shop said/stated/found/told|confirmed|repaired).
Hand audit of 20 random diagnosed complaints: in 13 the tags fired; in about 9 of those 13 the tagged part was the one the shop identified, in the rest the part was named in passing (for example "transmission limp mode" where the shop replaced an ABS module). Two engine failures were missed because the text said "lower bearing" or "power transfer unit". Treat the table as "what owners talk about", not as verified repairs.

4. RECALLS
- Recalls whose consequence text contains "motive power", "loss of power", "loss of drive", "reduced power" or "loss of propulsion": 638 distinct campaigns, 15,679,223 vehicles (sum of the largest potentially-affected figure per campaign). Reported by year, "loss of motive power"/"loss of propulsion"/"loss of drive power" phrasing only: 2015: 11, 2016: 10, 2017: 10, 2018: 4, 2019: 9, 2020: 14, 2021: 29, 2022: 42, 2023: 66, 2024: 73, 2025: 83, 2026 to 8 Sep: 35. Part of the rise is phrasing: NHTSA summaries adopted "loss of motive power" as standard wording in the 2020s.
- Recalls whose component is ACCELERATOR PEDAL or THROTTLE and whose defect/remedy text is not about sticking pedals, floor mats or unintended acceleration: 53 campaigns, 1,056,777 vehicles.
- Recalls whose defect text names a CVT or continuously variable transmission: 14 campaigns, 1,608,758 vehicles (1,303,530 of them Subaru 19V-149, a brake-light-switch recall that only mentions the CVT in passing).

5. BULLETINS (reduced_engine_power_bulletins.csv)
Every manufacturer communication whose NHTSA summary contains the phrase "reduced engine power": 171 documents (139 distinct bulletin numbers once revisions are merged). By make: Chevrolet 47, GMC 31, Cadillac 16, Lexus 13, Ford 11, Buick 7, Porsche 6, MINI 6, others 34. Keyword class (column class_by_keyword): software = summary names a reprogram/calibration/software update and no part; software and hardware = both; hardware or inspection = names a part, harness, sensor, pump, hose, pipe, valve or kit; diagnostic or information only = engineering information requests, tech-line articles or diagnostic steps with no fix named. Result: 108 hardware or inspection, 21 diagnostic/information only, 18 software only, 14 software and hardware, 10 unclassified. All 171 summaries were read; the CSV keeps a 160-character snippet of each.
The phrase count for "limp mode" or "limp home" in summaries is 349 documents (Jeep 140, Dodge 77, Ram 73, Mercedes-Benz 39, Chrysler 22, Audi 17). Bulletins naming CVT together with hesitation, lack of power, judder, fail-safe or loss of power: Nissan 75, Subaru 10, Infiniti 8, Mitsubishi 3, Honda 2.

6. LIMITS
Complaints are unverified owner statements; one incident can be filed more than once; the same text can name several parts; a mention of a recall does not mean the vehicle was in it. The set is defined by words, so an owner who wrote "the car died" without an acceleration phrase is not in it, and a hypothetical ("the dealer said it will lose acceleration") can be. Nothing here is a failure rate.