PulsCar analysis of NHTSA public data, September 2026 Finding a coolant leak: 10,523 owner complaints, 267 of them describing a leak nobody could find, and 1,041 manufacturer communications Licensed CC BY 4.0. Queries run 18 September 2026. -------------------------------------------------------------------------- SOURCE FILES -------------------------------------------------------------------------- NHTSA ODI complaint flat files COMPLAINTS_RECEIVED_2015-2019, _2020-2024, _2025-2026 -- https://static.nhtsa.gov/odi/ffdd/cmpl/ (dictionary CMPL.txt) 777,040 distinct vehicle complaints (ODINO, product type V) received 1 January 2015 to 8 September 2026. NHTSA manufacturer communications flat file FLAT_TSBS.zip -- 5,267,097 rows, one row per make/model/model year per communication; counted here as distinct NHTSA communication IDs. Bulletin PDFs opened individually at https://static.nhtsa.gov/odi/tsbs/... (listed under PRIMARY DOCUMENTS below). -------------------------------------------------------------------------- RELATIONSHIP TO OUR OTHER COOLANT ANALYSIS -- READ THIS FIRST -------------------------------------------------------------------------- Our companion page, coolant leak repair cost, publishes a complaint set built to answer a cost question. It used to publish 3,512 against the 10,523 here, and this file explained the gap as a difference of method. THAT EXPLANATION WAS WRONG, and is corrected here on 18 September 2026: the cost page's query had been run without case folding, so it kept almost no complaint received after 2020 (NHTSA narratives are upper case only through 2020). Re-run case-insensitively it returns 10,451 -- within 0.7% of the set on this page, because the two queries really are close. The remaining differences (sentence splitting, and whether a narrative is dropped when another fluid is named in the same sentence) are small. NEITHER TOTAL IS A FAILURE RATE. Every share quoted on this page is a share of THIS set, and every share quoted on the cost page is a share of THAT set. Where the two pages quote the same fact (the recall campaigns, the owner-reported repair figures), this page cites the cost page's numbers rather than recomputing them. -------------------------------------------------------------------------- WHICH COMPLAINTS COUNT (the coolant-loss set: 10,523) -------------------------------------------------------------------------- Step 1. Pre-filter in SQL: the narrative contains COOLANT, ANTIFREEZE, ANTI-FREEZE, RADIATOR, WATER PUMP, HEATER CORE, EXPANSION TANK, OVERFLOW TANK or HEAD GASKET. 24,675 complaints. Step 2. Split the narrative on [.!?;] plus newline. A complaint counts when ONE of those fragments contains a cooling-system word and a leak word whose match positions are within 70 characters of each other. cooling-system words: COOLANT | ANTIFREEZE | ANTI-FREEZE | RADIATOR | WATER PUMP | HEATER CORE | THERMOSTAT HOUSING | COOLANT RESERVOIR | EXPANSION TANK | OVERFLOW TANK | HEAD GASKET leak words: LEAK* | SEEP* | DRIP* | BURST* | RUPTUR* | CRACK* | SPRAY* | SPEW* | POOLING/POOLED UNDER | PUDDLE | LOSING COOLANT | LOSS OF COOLANT | COOLANT LOSS | LOW COOLANT | EMPTY Step 3. Exclusions, applied to the fragment: NO LEAK, NOT LEAK, DID NOT LEAK, DOES NOT LEAK; a leak of a different fluid named in the same fragment (OIL LEAK, BRAKE FLUID LEAK, POWER STEERING FLUID LEAK, TRANSMISSION FLUID LEAK, FUEL LEAK, GAS LEAK); cabin water intrusion (WATER LEAK INTO, SUNROOF, WINDSHIELD LEAK). Step 4. Motorcycle, heavy-truck, bus and RV manufacturers excluded by make. Result: 10,523 complaints. NOTE on the radiator's internal transmission cooler: narratives describing coolant and transmission fluid mixing inside a radiator are KEPT, because those sentences name a radiator and a coolant leak. They are a real cooling-system failure, not a transmission complaint. SAMPLE READING (done before any count below was published) Thirty matches were drawn at random (seed 41) and read in full, including re-reading the coolant sentences of the five that were ambiguous in the first 430 characters. All thirty described a coolant leak or a coolant loss. FALSE-POSITIVE RATE OBSERVED: 0 IN 30. That is a small sample; treat it as "few obvious false positives", not as zero. -------------------------------------------------------------------------- RESULT 1 -- DISCOVERY SIGNALS (coolant-leak-discovery-signals.csv) -------------------------------------------------------------------------- Each row is a regex over the narrative. A narrative can match several rows, so the shares do not sum to 100. THESE MEASURE WHAT OWNERS WROTE DOWN, not what happened: a complaint is filed after something frightening, so alarming signals are over-represented relative to how leaks are found in daily life. check-engine light or a misfire code 3,248 30.87% temperature gauge climbed or the engine overheated 2,643 25.12% low-coolant or coolant-temperature message 1,148 10.91% a smell 639 6.07% white smoke from the tailpipe 573 5.45% a low or empty reservoir, found by looking 532 5.06% a puddle, pool or drip on the ground 361 3.43% a shop found it during other work 184 1.75% steam or smoke from under the hood 132 1.25% names none of the above 4,107 39.03% -------------------------------------------------------------------------- RESULT 2 -- INSIDE OR OUTSIDE (coolant-leak-internal-vs-external.csv) -------------------------------------------------------------------------- Markers for coolant going INSIDE the engine (cylinder, combustion chamber, block, oil, transmission, white exhaust smoke) versus markers for coolant going OUT (puddle, drip, spray, soaked bay, steam under the hood). any internal marker 3,587 34.09% any external marker 737 7.00% internal markers only 3,448 32.77% external markers only 598 5.68% both 139 1.32% neither 6,338 60.23% fire flag: external-only 6.19%, internal-only 0.44%, whole set 1.53%, all 777,040 complaints 1.82%. towed: external-only 4.68%, internal-only 5.77%, whole set 5.02%, baseline 5.11%. median miles: external-only 85,850; internal-only 90,000; whole set 88,000 (4,587 complaints report a usable mileage; 0 < miles < 400,000). WHY THIS IS NOT A BASE RATE: NHTSA receives complaints about failures that frighten, strand or cost thousands. Coolant going into an engine does all three; a weeping hose clamp does none. The internal share here is a fact about the complaint file, and the article says so where it is used. -------------------------------------------------------------------------- RESULT 3 -- THE LEAK NOBODY COULD FIND (coolant-leak-unfindable-leaks.csv) -------------------------------------------------------------------------- Within the 10,523, narratives containing one of: COULD NOT FIND / COULDN'T FIND / CANNOT FIND / CAN'T FIND / UNABLE TO FIND / UNABLE TO LOCATE / COULD NOT LOCATE / NO VISIBLE LEAK / NO SIGN(S) OF A LEAK / NO LEAK(S) FOUND / DID NOT FIND A LEAK / FOUND NO LEAK / NO OBVIOUS LEAK / NO PUDDLE / NO EXTERNAL LEAK(S) / NEVER FOUND THE LEAK. FIRST PASS, no proximity rule: 329 complaints. Thirty were read at random (seed 7): FIVE were about a different failure in the same narrative -- an airbag light, a steering column, a charging fault, a death wobble, a sputter. FALSE-POSITIVE RATE 5 IN 30, ROUGHLY 17%. TIGHTENED: the phrase must sit within 150 characters of COOLANT, ANTIFREEZE, RADIATOR, WATER PUMP, HEATER CORE, HEAD GASKET, EXPANSION TANK or OVERFLOW. 267 complaints. Thirty read at random (seed 3): all thirty were about a coolant leak nobody could locate. 0 IN 30. What the narrative says it turned out to be (shares of 267; a narrative can name more than one, and many end unresolved): coolant inside the engine (cylinder, block, head gasket, misfire, white smoke, borescope, spark plugs) 133 49.8% an engine, short block or long block replaced 46 17.2% a water pump named 20 7.5% a radiator, hose or coolant pipe named 15 5.6% a reservoir, expansion tank or cap named 8 3.0% a thermostat or its housing named 8 3.0% a heater core or heater line named 3 1.1% names a pressure test, dye test, block test or borescope 38 14.2% CONCENTRATION CHECK, because one documented defect family dominates: Ford/Lincoln/Mercury rows 168 (62.9%), of which 97 (57.7%) internal. All other makes 99 (37.1%), of which 36 (36.4%) internal. Outcome flags: fire 2 (0.75%), towed 3 (1.12%), crash 0 (0.00%). Miles: 99 complaints report a usable figure; median 78,000, quartiles 50,000 and 119,000. -------------------------------------------------------------------------- RESULT 4 -- MANUFACTURER COMMUNICATIONS (coolant-leak-bulletins-by-component.csv) -------------------------------------------------------------------------- FTS5 query over bulletin summaries: (coolant OR antifreeze OR radiator OR "water pump" OR "heater core") AND (leak OR leaks OR leaking OR seep OR seepage OR "coolant loss" OR "losing coolant") Counted as distinct NHTSA communication IDs, after removing heavy-truck, bus, RV and motorcycle brands: 1,041 bulletins. Component named, share of 1,041 (a summary can name more than one): radiator 126 (12.1%); head gasket, cylinder head or block 105 (10.1%); hose, pipe or coolant line 104 (10.0%); water pump 102 (9.8%); reservoir, expansion tank, degas bottle or cap 71 (6.8%); thermostat or housing 68 (6.5%); oil cooler, heat exchanger, EGR cooler, intercooler or transmission cooler 64 (6.1%); turbocharger and its coolant lines 55 (5.3%); heater core 12 (1.2%); intake manifold 3 (0.3%); freeze or core plug 1 (0.1%). Summary names a leak-finding method (dye, pressure test, borescope, smoke test): 41 (3.9%). LIMITATION THAT MATTERS: the flat file carries an ABSTRACT of each bulletin, not its full text. A bulletin whose abstract does not mention a pressure test may still contain one; that is why the 3.9% is described in the article as what the abstracts say, and why the bulletins quoted in the article were downloaded and read as PDFs instead. -------------------------------------------------------------------------- PRIMARY DOCUMENTS OPENED AND READ, 18 SEPTEMBER 2026 -------------------------------------------------------------------------- Ford SSM 47849, 7 March 2019, 2014-2019 Fusion / 2017-2019 Escape 1.5L EcoBoost -- https://static.nhtsa.gov/odi/tsbs/2019/MC-10156569-0001.pdf Ford SSM 48345, 5 December 2019, 2019-2020 Edge / Escape / Nautilus / Corsair 2.0L EcoBoost, EGR cooler -- https://static.nhtsa.gov/odi/tsbs/2019/MC-10168812-0001.pdf GM PIP5465, January 2017, 2016-2017 Chevrolet Malibu 1.8L, exhaust heat exchanger -- https://static.nhtsa.gov/odi/tsbs/2017/MC-10113359-9999.pdf BMW SI B11 11 17, September 2018, turbocharged V8 leak diagnosis -- https://static.nhtsa.gov/odi/tsbs/2018/MC-10149961-9999.pdf Mercedes-Benz LI09.41-N-079165 version 1, 21 April 2025, OM654 charge air cooler -- https://static.nhtsa.gov/odi/tsbs/2025/MC-11017563-0001.pdf Read in the flat file (no separate PDF): GM 14417 / 14417A customer satisfaction program, 3 March 2017, Chevrolet Cruze 1.4L; GM PIP4959D, 15 August 2017; Land Rover JLRTB02127NAS5 (28 March 2025), NAS6 (5 November 2025) and NAS7 (13 May 2026, rescinded); Land Rover TUCC_AUG2025_30 (3 September 2025). -------------------------------------------------------------------------- LIMITATIONS -- READ BEFORE QUOTING ANY NUMBER -------------------------------------------------------------------------- 1. Complaints are self-reported, unverified and voluntary. They are not failure rates and cannot be converted into one: NHTSA publishes no denominator. 2. Counts favour high-volume models and models whose problem has been publicized. Nothing here ranks reliability, and nothing here says one brand leaks more than another. 3. Keyword shares measure vocabulary. "The owner did not write 'puddle'" is not the same as "there was no puddle". 4. Fire, towed, crash and mileage are the owner's own entries. 5. The unfindable-leak set is 267 narratives, and 168 of them come from one manufacturer with a documented defect family. The article reports the split with and without that group. -------------------------------------------------------------------------- FILES -------------------------------------------------------------------------- coolant-leak-discovery-signals.csv signal, complaints, share, median miles, miles reported, fire %, towed %. coolant-leak-internal-vs-external.csv group, marker, complaints, share, fire %, towed %, crash %, median miles, miles reported, plus summary rows and the all-complaint baseline. coolant-leak-unfindable-leaks.csv what the narrative says it turned out to be, the Ford concentration check, makes, models, outcome flags. coolant-leak-bulletins-by-component.csv component named, diagnosis rows, makes, total. No personal data is published: no VIN, no city, no dealer name, no owner free text beyond short quoted snippets in the article itself.