PulsCar analysis of NHTSA public data, September 2026 "Car starts then dies": what 525 owner complaints, 54 recall campaigns and 23 manufacturer communications say License: CC BY 4.0. Credit "PulsCar analysis of NHTSA public data, September 2026" with a link to https://pulscar.io/blog/car-starts-then-dies All queries were run on 18 September 2026. -------------------------------------------------------------------------- 1. SOURCE FILES -------------------------------------------------------------------------- NHTSA public flat files from https://static.nhtsa.gov/odi/, in local SQLite. Complaints FLAT_CMPL. 777,040 vehicle complaints (one row per ODINO) received 1 January 2015 to 8 September 2026. Dictionary: https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt Recalls FLAT_RCL, campaigns counted DISTINCT on the campaign number, populations taken at each campaign's largest reported POTAFF. Communications FLAT_TSBS, counted DISTINCT on the NHTSA ID. https://static.nhtsa.gov/odi/ffdd/tsbs/TSBS.txt CASE SENSITIVITY. NHTSA complaint narratives are ALL CAPITALS through 2020 and mixed case from 2021 onward. An uppercase Python regex without re.IGNORECASE, or an uppercase SQLite GLOB, silently drops most post-2020 complaints. Every regex below was compiled with re.IGNORECASE; SQLite LIKE and FTS5 MATCH are case-insensitive and these queries rely on that. -------------------------------------------------------------------------- 2. WHY THIS SET IS SMALL ON PURPOSE -------------------------------------------------------------------------- "Starts then dies" has no fixed vocabulary. Owners write "starts and stalls", "fires and quits", "runs three seconds then shuts off", "won't stay running". A loose query for a start word near a death word returns tens of thousands of complaints, and reading them shows most are stalls while driving. We chose precision over size and ended with 525. Step 1 — FTS5 candidates (37,780): NEAR(start* di*, 12) plus the same NEAR construction for "shuts off", "shut off", stall*, "turns off", "cuts off", "shut down", quit*, "few seconds", NEAR(fire* di*, 10), NEAR(crank* di*, 12), NEAR(start* idle*, 10), and the phrases "stay running" / "staying running". Step 2 — three case-insensitive patterns, any one of which qualifies: A. a start verb (start/starts/started/fire/fires/fired), optionally followed by up/it/the car/the engine, then a connector (and / then / but / only to / comma), then within 30 characters a death verb (dies, shuts off, shuts down, cuts off, cuts out, stalls, quits, stops running, turns itself off, goes dead, dying). E. the same relationship written tightly: "start(s|ed) then|and [immediately] dies|shuts off|stalls|quits|cuts off". D. "won't / will not / doesn't / wouldn't / couldn't (stay|keep|continue| remain) running", or a start verb followed by "but|and|then" and one of those phrases. The start verb carries a negative lookahead so that inchoative uses do not match: "started knocking", "started to leak" and "started leaking" are rejected, "started and stalled" is kept. This one rule removed several hundred false positives. Step 3 — two exclusions applied to the matched window: - a non-engine subject within 55 characters before the match (sound, noise, pecking, beep, chime, alarm, horn, radio, wiper, blower, fan, light, display, screen, air conditioning, heater, clicking, ticking, buzzing, music, speaker), which removes reports where the thing that "started and quit" was a noise; - the reverse complaint "would not stop running" / "would not shut off". Step 4 — drop motorcycles, powersports and heavy trucks by make. FINAL SET: 525. -------------------------------------------------------------------------- 3. THE SPLIT: 272 AT A START ATTEMPT, 253 AFTER A ROADSIDE STALL -------------------------------------------------------------------------- A complaint is marked "roadside" when the narrative anywhere contains a driving context (while driving, driving at/down/on/home/to, at NN mph, highway, freeway, interstate, in traffic, on the road, at speed, in motion, going NN, merging, intersection, red light, turnpike, exit ramp, pulled over/onto, mid-drive, down the road). Everything else is "at a start attempt": 272 complaints. Those 272 are what the article's percentages are computed on. SAMPLE READING. Thirty at-start-attempt complaints were drawn at random (random.seed(31337)) and read in full. Twenty-five described a vehicle that started and then died shortly after, or that would not stay running. Five did not: two were no-start reports with nothing running, two were stalls on the road whose narrative happened to avoid every driving word, and one was an engine shutdown caused by a door-module fire. FALSE-POSITIVE RATE 5 of 30, about 17%. WHY THE SPLIT MATTERS. Almost half of what a search engine will return for this symptom is a stall while driving, told from the roadside. The two have different causes and different urgency, and mixing them is how a page about a driveway problem ends up quoting numbers from a highway one. -------------------------------------------------------------------------- 4. WHAT WAS COUNTED -------------------------------------------------------------------------- start-then-die-by-context.csv The three groups and the whole-file baseline: TOWED, CRASH and FIRE flags, injuries, mileage (MILES, 0 < miles < 400,000) and the share at or under 36,000 miles. Headline: at a start attempt the crash flag is 0.74% against a 4.37% baseline (this failure happens where the car is parked), while the fire flag is 2.94% against 1.82%. Median mileage 80,000 across 163 readings; 19.6% at or under 36,000 miles. start-then-die-what-owners-name.csv Case-insensitive regex families across the 272 and across all 525. Mentions, not confirmed repairs; a complaint can name several. THE MOST IMPORTANT CAVEAT ON THIS PAGE. The two largest families are the fuel pump / fuel relay / power module group and the battery / alternator / charging group, at 17.28% each. The idle-control family that dominates service-counter advice — throttle body 1.10%, mass air flow 0.37%, vacuum or intake 0.74% — is almost invisible. That is NOT evidence that idle control is a rare cause. It is evidence that a $15 can of throttle-body cleaner does not produce a federal safety complaint, while a car that has to be towed does. Every number in this table is filtered by "was it bad enough to report to the government", and the article says so. start-then-die-by-make-and-model.csv Counts with the visible denominator: for a make, the rate per 10,000 of that make's own complaints in the same file; for a model, per 10,000 of that model's own complaints. NOT failure rates. 34.19% of the at-start- attempt narratives already name a recall, and recall publicity is itself a reason to file. start-then-die-recall-campaigns.csv Campaigns whose DEFECT or CONSEQUENCE text contains a stall word AND a no-start/restart word (SQLite LIKE, case-insensitive): 54 campaigns, 2,259,387 potentially affected units, each campaign counted once at its largest population. Manufacturer communications (counted, not published as a CSV) tsbs_fts MATCH on the phrases "starts and stalls", "start and stall", "starts then stalls", "start then stall", "stalls after start", "dies after start", "will not stay running", "does not stay running", "engine stalls immediately", "start and die", "starts and dies" -> 23 distinct NHTSA IDs. This is a small number and is reported as one; the phrase set is narrow by design, and a bulletin that describes the same condition in other words will not be in it. -------------------------------------------------------------------------- 5. DOCUMENTS READ -------------------------------------------------------------------------- Recall 19V813000 (Chrysler, 528,594 2011-2013 Dodge Durango and Jeep Grand Cherokee): "The fuel pump relay inside the Totally Integrated Power Module (TIPM-7) may fail, causing the vehicle to stall without warning." The text also records that these vehicles were "previously recalled under NHTSA Recall 14V530 or 15V115." Recall 18V735000 (Ford, 1,282,596 vehicles): the largest campaign in our set by population. Manufacturer communication NHTSA ID 10216450, Chrysler bulletin 18-015-13, read in NHTSA's manufacturer-communications flat file (the PDF was not retrievable at the static.nhtsa.gov paths we tried, and we did not work around that): "Flash: No Start or Start and Stall When Using Remote Start ... Engine starts then immediately stalls when using Remote Start system (Instrument Cluster may display the message 'Remote Start Disabled, Start Vehicle to Reset') ... Engine may be hard to re-start at ambient temps below 10F(-12C)." The remedy is reprogramming the PCM and BCM. Manufacturer communication NHTSA ID 10220495 (Jeep Cherokee 2020-2021, KL): "Owner reports in colder temperatures the vehicle may exhibit a No Start or and/or a Start and Stall fail to Start condition. Vehicle will start normally when PDC or vehicle is warmed up. This symptom may also be accompanied by DTC P1279 - STARTER CONTROL CIRCUIT 2 OPEN." The remedy is one Denso Run/Start relay. -------------------------------------------------------------------------- 6. LIMITATIONS -------------------------------------------------------------------------- - Complaints are self-reported, unverified, and are not failure rates. - 272 is a small file. Percentages on it move by 0.37 points for every single complaint, and no claim in the article rests on a difference smaller than a few points. - Regex families count words, not diagnoses. - The severity filter described above biases the whole table toward expensive, stranding causes. Read it as "what gets reported", never as "what usually goes wrong".