PulsCar analysis of NHTSA public data, September 2026
Article: https://www.pulscar.io/blog/car-battery-keeps-dying
License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) — cite "PulsCar analysis of NHTSA public data, September 2026".

SOURCE FILES
- NHTSA Office of Defects Investigation flat files, downloaded September 2026 and loaded into SQLite:
  COMPLAINTS (owner complaints received 1 January 2015 – 8 September 2026, de-duplicated by ODI number: 777,040 rows),
  RECALLS (one row per make/model/year per campaign), TSBS / manufacturer communications (one row per make/model/year per document),
  INVESTIGATIONS.
- Documents opened and read (all on static.nhtsa.gov, links in the article's "Sources and documents"): Subaru 07-207-22R, 15-317-24R,
  15-318-24R, 15-308-23R; Mazda SA-049/20, 01-009/26; Kia TSB ELE 323 (SA555); Toyota T-TT-0625-20; GM 16-NA-227, 24-NA-229;
  Ford TSB 18-2013 and Customer Satisfaction Program 19N09; VW 27-15-03; NHTSA recall records 19V-348, 13V-044, 15V-145, 19V-813.

COMPLAINT SET ("battery drain complaints", n = 5,163)
Step 1. Full-text pre-filter: complaints whose description contains the word "battery" (34,089 rows).
Step 2. Symptom regex (case-insensitive), any one of:
  battery|batteries + (keeps|kept|continues|continued|constantly|repeatedly|always|again) + (dying|going dead|draining|dies|died|drains|drained|dead|discharging)
  "parasitic draw|drain|load"
  (new|replacement|second|third|2nd|3rd) battery + (died|dies|dead|drained|drains|went dead|is dead|was dead|also died)
  battery + (died|dies|drained|drains|went dead|was dead|is dead|was completely dead|was/is/had/has drained|drained overnight|goes dead|went completely dead|was totally dead)
  (drains|drained|draining|kills|killed|killing) + (the|my|our|his|her) + [12 volt] battery
  "dead battery|batteries", "battery drain", "battery drains", "battery drainage", "battery discharge"
Step 3. Exclusions:
  - high-voltage / EV / hybrid-pack / key-fob contexts: hybrid battery, high voltage, HV battery, traction battery, lithium, li-ion, charging station,
    supercharg*, charge port, state of charge, range loss/drop/anxiety, kWh, electric vehicle, EV, PHEV, BEV, plug-in, battery pack, drive battery,
    propulsion, fob/remote/key battery — UNLESS the text also says "12V"/"12 volt";
  - makes TESLA, RIVIAN, LUCID, POLESTAR, FISKER, VINFAST, SMART;
  - motorcycle wording (motorcycle, Harley, R1200, Ninja, Goldwing, scooter, bike);
  - hypothetical-only matches: if every symptom match is preceded within 45 characters by afraid/worried/concern/fear/could/would/will/may/might/can
    (e.g. "I am afraid this will drain my battery"), the complaint is dropped.
Result: 5,896 matches before exclusions, 5,163 after.

SAMPLE READING (false-positive check)
- First pass (looser regex, 5,613 rows): 40 random complaints read; about 10 of 40 were off-topic (hypothetical drains, motorcycle ABS,
  an incidental "dead battery" in a complaint about something else). Query tightened as in Step 3.
- Second pass (final set): 30 random complaints read; 2 of 30 were arguably off-topic (a lock failure blamed on "battery drainage",
  a battery that died during a highway drive with no drain described). Estimated false-positive rate about 7%.
- Third pass: 40 random complaints that contain fix wording (fixed, resolved, replaced the, turned out, found that, diagnosed, traced) read to
  categorize the stated outcome. Counts from those 40: battery replaced with no root cause named 10; a module or telematics unit named 5;
  a switch, latch, relay or light that stayed on 7; alternator or battery current sensor 2; steering-column or power-steering module 2;
  shop found nothing 5; the rest were about other faults or gave no outcome.
  In total 110 complaints were read by a person; all other counts below are keyword counts over the 5,163 texts.

KEYWORD COUNTS OVER THE 5,163 (case-insensitive regex, one count per complaint)
- "jump" 1,281 (24.8%) · "recall" 1,062 (20.6%) · towed/tow 734 (14.2%) · stranded 564 (10.9%) · stall / died while driving 539 (10.4%)
- door ajar / latch / dome or interior light stays on 485 (9.4%) · radio / head unit / infotainment / Uconnect / Sync 448 (8.7%)
- alternator 421 (8.2%) · parasitic 385 (7.5%) · shop could not find / duplicate / reproduce / "no problem found" 352 (6.8%)
- battery replaced two or more times 310 (6.0%) · software / reflash / reprogram / firmware 211 (4.1%)
- telematics module (DCM, Starlink, OnStar, HFL, HandsFreeLink, Bluetooth) 205 (4.0%) · TIPM / Totally Integrated Power Module 96 (1.9%)
- start-stop / auxiliary battery / idle stop 136 (2.6%) · relay 101 (2.0%) · "mA"/milliamp figure given 13 · multimeter / ammeter / clamp 7
- Ford/Lincoln Edge, MKX, Flex, MKT, Explorer 2011-2014: 317 complaints, 284 of them mention a door-ajar, latch, dome or interior light.
- Subaru "class vehicles" (2015-2020 Legacy, Outback, Forester, WRX/STI; 2019-2020 Ascent): 1,107 of the 1,243 Subaru complaints.
- Honda CR-V 2017-2019: 107.

SEVERITY FLAGS (NHTSA fields CRASH, FIRE, TOWED) vs the all-complaints baseline (777,040)
- towed 1.55% vs 5.11% · crash 0.68% vs 4.37% · fire 0.87% vs 1.82%.

MILEAGE: NHTSA "miles" field where 0 < miles < 400,000 (2,335 of 5,163 give a usable figure): median 50,000; quartiles 21,000 and 87,000.
Subaru median 18,000 (n=405), Honda 28,300 (n=170), Ford 55,000 (n=442).

BULLETINS ("battery_drain_bulletins.csv", n = 993 distinct NHTSA document IDs)
Step 1. tsbs_fts full-text: "battery" AND (drain OR drains OR draining OR drained OR parasitic OR discharge OR discharged OR discharges OR dead OR dies OR died OR dying)
        → 50,785 make/model/year rows.
Step 2. Kept only summaries that name the condition: parasitic; battery drain; drains/draining the battery; dead battery; battery goes/went/is/was dead;
        battery discharge(d); key-off load/draw; IOD; ignition-off draw; dark current; excessive draw; battery may drain/discharge; battery dies/dying/died.
Step 3. Excluded high-voltage/EV/key-fob/TPMS contexts unless "12V" appears; excluded motorcycles, trailers, motorhomes, buses, fire trucks,
        ambulances, and storage/transport/pre-delivery bulletins unless the word "customer" appears.
Step 4. Counted DISTINCT nhtsa_id → 993 documents; per make, a document is counted once for each make it names.
Remedy class is a KEYWORD class from the summary, not a reading of each document: software (reprogram/reflash/software/calibration/firmware),
hardware (replace/install/repair/harness/connector/switch/module without software words), diagnostic guidance (procedure/how to/tips/measurement
without "replace"), unclassified (300 documents whose summary does not say). Component keyword columns are also keyword flags and can overlap.

RECALLS
Query: defect text contains "battery" AND one of drain, discharg*, dead, not start, no-start, no start, deplet*, parasitic, loss of electrical, loss of power;
high-voltage / EV / trailer / breakaway contexts removed unless "12V" appears → 27 campaigns (passenger cars, trucks, buses, RVs, equipment).
Only one passenger-vehicle campaign describes the battery draining while parked (15V-145, 2011-2013 Chevrolet Volt, 50,236 vehicles, software).
The passenger-vehicle campaigns with the largest populations are loose or degrading battery connections that cut power while driving:
13V-044 BMW (516,791), 19V-348 Chrysler Pacifica (198,731), 19V-020 Ram 1500 (159,740), 19V-472 BMW (16,641), 20V-694 and 21V-457 Land Rover Discovery.

LIMITATIONS
- Complaints are self-reported, unverified, not a failure rate, and skew toward vehicles with publicity (class actions, recalls).
  Per-1,000 figures normalize by the make's total complaints, not by vehicles on the road.
- The regex cannot tell a drain from a worn-out battery; the set is "owners whose 12V battery went dead repeatedly or was drained".
- Keyword categories over the 5,163 texts were not individually read; the 110 read complaints are the check on them.
- Bulletin summaries are NHTSA's short abstracts; where a summary does not state the remedy the document is "unclassified".