PULSCAR ANALYSIS — AIR CONDITIONING COMPRESSOR FAILURES IN NHTSA PUBLIC DATA PulsCar analysis of NHTSA public data, September 2026. License: CC BY 4.0. Attribution: "PulsCar analysis of NHTSA public data, September 2026." Published alongside https://pulscar.io/blog/ac-compressor-replacement-cost ------------------------------------------------------------------------------ 1. SOURCE FILES ------------------------------------------------------------------------------ All inputs are NHTSA's own published flat files, downloaded from static.nhtsa.gov and loaded into a local SQLite database. No third-party data was used. COMPLAINTS — 777,040 owner complaints received 1 January 2015 to 8 September 2026. Field definitions: https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt RECALLS — one row per make/model/year per campaign; campaigns counted DISTINCT. Field definitions: https://static.nhtsa.gov/odi/ffdd/rcl/RCL.txt TSBs — manufacturer communications (service bulletins, dealer messages, owner notification letters), one row per make/model/year; bulletins counted DISTINCT by NHTSA document id. Field definitions: https://static.nhtsa.gov/odi/ffdd/tsbs/FLAT_TSBS.txt Complaints are self-reported. They are not failure rates, and they have no denominator: nobody publishes how many of each model are on the road. Nothing in this analysis should be read as "make X is more or less reliable than make Y". ------------------------------------------------------------------------------ 2. BUILDING THE COMPLAINT SET ------------------------------------------------------------------------------ Step 1 — full-text search of the narrative field: SELECT c.* FROM complaints c JOIN complaints_fts f ON f.rowid = c.rowid WHERE complaints_fts MATCH 'compressor'; -> 1,550 complaints. Step 2 — for every occurrence of the word "compressor" in the narrative, look at a window of 140 characters on each side and keep the complaint only if: (a) the window contains an air-conditioning term: a/c | ac | a.c. | air cond* | hvac | freon | refrigerant | r-134a | r-1234yf (b) the window does NOT contain an air-suspension term: air suspension | air spring | leveling | load level | tire inflat | ride height (air-suspension compressors are a different part and a different article) (c) the word is not immediately followed by a different part: compressor line | hose | pipe | tube | bracket | relay | fuse | switch | mount | fan | wire | wiring | harness | belt tensioner (d) a failure or repair cue appears within ~70 characters AFTER the word (failed, failure, seized, locked, went out, gave out, stopped working, not working, quit, died, bad, blew, blown, broke, grinding, rattling, noise, leak, replace, needs replacing, shattered, exploded, disintegrated, came apart, burned, caught fire, smoking, defective, malfunctioned) OR within ~60 characters BEFORE it (failed, seized, bad, new, replaced, faulty, defective, broken, dead, noisy, grinding, leaking, locked, blown). -> 645 complaints. This is the set used everywhere on the page. Exclusions and known limits: - Mileage statistics drop odometer values <= 0 or >= 400,000. 300 of the 645 complaints carry a usable odometer; the other 345 are blank or zero. - Air-suspension and air-brake compressors are excluded by design. - The set is a keyword set, not an NHTSA component code. NHTSA's own component tag for air conditioning appears on only a few hundred complaints in the whole file, and most A/C compressor narratives are filed under other components, so the tag cannot be used to build this set. ------------------------------------------------------------------------------ 3. SAMPLE READING AND FALSE POSITIVES (done before publishing any count) ------------------------------------------------------------------------------ First query (air-conditioning word anywhere in a 140-character window, any failure cue anywhere in that window): 955 complaints. We drew 30 at random (seed 42) and read each one in full. 7 of 30 were false positives: a leaking A/C hose with a healthy compressor, a compressor replaced during an unrelated repair, a popping noise heard when the compressor cycled, a metal A/C line that chafed a radiator hose, a compressor the owner said explicitly was working. Observed false-positive rate 23%. We tightened the query as described in section 2 (failure cue must sit tight around the word; part-name exclusions added) and drew a second random sample of 30 (seed 1234) from the resulting 647 complaints, read in full. 4 of 30 were false positives: - "A/C Dual Compressor fan went out" (a cooling fan, not the compressor) - a rodent-chewed A/C compressor high-voltage wire - a timing-cover leak located "about the A/C compressor" - "the air compressor had a hole in the tube" (an A/C line) Observed false-positive rate 13%. We then added "fan", "wire", "wiring" and "harness" to the part-name exclusion list, which removed two of those four and left the final set at 645. We did not re-draw a third sample, so 13% is the last rate we actually measured and is the rate quoted on the page. Read the counts as carrying an error of that order. ------------------------------------------------------------------------------ 4. READING 110 COMPLAINTS IN FULL ------------------------------------------------------------------------------ 110 complaints were drawn at random from the 645 (Python random.seed(2026), random.sample) and each narrative was read end to end. Each was assigned exactly one primary outcome — what the story turned out to be about, not what words it contained. Results: 36 the compressor itself, plainly (failed / seized / replaced) 15 another part was the actual leak or cause — condenser, evaporator, expansion valve, discharge hose, a water pump that dumped coolant onto the compressor, a shorted fuse panel that fed it constant power 13 a sub-part of the compressor, not the whole unit — clutch, clutch coil, field coil, shaft seal, torque limiter, bearing 8 the compressor came apart internally and contaminated the system (metal shards, "imploded", "black death", debris through the lines) 8 the failure took the belt with it — stall, lost power steering, lost brake assist, emergency stop 7 an electric compressor on a hybrid or EV, where it also cools the traction battery (Tesla, BMW i3/i8, Mercedes EQS, Toyota bZ4X, Kia Sportage Hybrid) 6 the story is the warranty extension or parts availability, not a diagnosis 11 off target — the compressor is incidental to a complaint about something else 6 no diagnosis stated, or nobody found one --- 110 Categorisation is a judgement call made by one reader; a second reader would not assign every borderline case identically. The categories are mutually exclusive by construction (one primary outcome per complaint). ------------------------------------------------------------------------------ 5. OWNER-REPORTED DOLLAR FIGURES ------------------------------------------------------------------------------ Within the 645 complaints, every "$" amount between $150 and $8,000 was extracted and kept only when, in a window of 100 characters on each side, the text - contained a cost verb (cost, charge, quote, paid, pay, bill, estimate, repair, replace, fix, price), AND - contained an air-conditioning term (a/c, air conditioning, compressor, freon, refrigerant, condenser, evaporator), AND - contained none of: transmission, engine replacement, catalytic, head gasket, injector, spark plug, down payment, car payment, per month, monthly, purchase price, worth, value of, loan, deductible, insurance, lease, towing fee, rental, diagnosis/diagnostic, brake, tire, battery, alternator, starter, water pump, suspension, head unit, infotainment, radio. Identical amounts repeated inside one complaint were counted once. -> 113 figures from 92 complaints. Distribution: median $1,687; 25th percentile $1,000; 75th percentile $2,311; range $150 to $6,000. under $500 8 (7.1%) $500-$999 17 (15.0%) $1,000-$1,499 24 (21.2%) $1,500-$2,199 31 (27.4%) $2,200-$2,999 14 (12.4%) $3,000 and up 19 (16.8%) Sample reading: 30 of the extracted figures were drawn at random and read in context. 27 were a genuine air-conditioning repair figure. One was a head unit (the word "ac" appeared elsewhere in the sentence) and two bundled the A/C job with radiator or cooling-fan work in a single total. Observed impurity rate about 10%. The head-unit case was excluded by a rule added afterwards. These figures are quotes and payments as owners typed them. Some are parts only, some are parts and labour, some are dealer quotes that were declined. They span eleven years with no inflation adjustment, and people who write to a federal safety agency are not a random sample of people who buy A/C repairs — a bill that felt fair rarely gets reported. Use the distribution as a reality check on a quote, never as a national average. ------------------------------------------------------------------------------ 6. RECALLS ------------------------------------------------------------------------------ Campaigns were selected where the component field names air conditioning, or the defect text names a compressor, refrigerant or "a/c", then filtered the same way as the complaints (air-conditioning term required, air-suspension and air-brake compressors dropped). Campaigns counted DISTINCT; "potentially affected" is the figure the manufacturer filed. 121 campaigns, 2,454,189 vehicles and units in total 54 of those name a compressor: 662,191 units 85 of the 121 have a consequence that mentions FIRE: 646,007 units 28 campaigns name both a compressor and a fire risk: 350,137 units Published as ac-recalls.csv. The remedy_type column is derived from the remedy text: "software" where it mentions software, reprogramming or calibration, "inspect/replace" where it mentions inspection or replacement, "other" otherwise. ------------------------------------------------------------------------------ 7. MANUFACTURER COMMUNICATIONS (TSBs) ------------------------------------------------------------------------------ SELECT t.* FROM tsbs t JOIN tsbs_fts f ON f.rowid = t.rowid WHERE tsbs_fts MATCH 'compressor'; -> 1,225 distinct documents; after requiring an air-conditioning term in the summary and dropping air-suspension documents: 770 distinct bulletins, dated 27 June 2015 to 12 September 2026. Of those 770, counted by reading the summary text for each phrase family: leak 140 (18.2%) software / reprogram 144 (18.7%) noise 112 (14.5%) no cooling / warm air 89 (11.6%) oil 94 (12.2%) replace the compressor 69 (9.0%) clutch 67 (8.7%) condenser 48 (6.2%) warranty extension / service campaign 29 (3.8%) shaft seal 21 (2.7%) contamination / debris 10 (1.3%) By make, the largest counts are Honda 87, Chevrolet 60, Ford 44, Audi 41, Ram 37, Mercedes-Benz 31, Jeep 30, Acura 28, Volkswagen 25. Per-make bulletin counts are published as a column in ac-complaints-by-make.csv. These are counts of documents, not of failures, and a manufacturer that publishes more bulletins is not thereby a worse manufacturer. ------------------------------------------------------------------------------ 8. WHAT THE SET SHOWS (headline numbers used on the page) ------------------------------------------------------------------------------ 645 complaints, 1 Jan 2015 - 8 Sep 2026 Towed: 16 (2.48%) against 5.11% across all 777,040 complaints Fire: 11 (1.71%) against 1.82% baseline Crash: 9 (1.40%); 8 injuries reported; 0 deaths Median failure odometer 64,000 miles (n=300; quartiles 40,000 and 100,000) Mentions a belt: 82 (12.71%) Mentions a seized or locked unit: 60 (9.30%) Mentions defrost / defog / visibility:49 (7.60%) Mentions metal, debris or contamination:48 (7.44%) Mentions a recall: 166 (25.74%) Mentions a leak: 163 (25.27%) Mentions a receiver-drier or accumulator: 2 (0.31%) Mentions a system flush: 15 (2.33%) Honda (all models): 235 (36.4%) — Civic 125, CR-V 77 Electric compressor on a hybrid or EV: 20 (3.10%) ------------------------------------------------------------------------------ 9. LIMITATIONS, STATED PLAINLY ------------------------------------------------------------------------------ - Complaint counts are reporting counts. A warranty-extension letter teaches thousands of owners that NHTSA exists and lifts a model's count without changing a single compressor. Honda's counts in this set are affected by that and are not evidence that Honda compressors fail more often than anyone else's. - Only 300 of 645 complaints carry a usable odometer, so the median failure mileage describes those 300, not the set. - Complaints skew towards early failures. A compressor that lasted twelve years does not generate a safety complaint, so the median here is a median of complained-about failures, not of compressor life. - Model-level counts below about ten complaints are small samples and are published as counts, without percentages. - Nothing here measures how often a shop replaced a compressor that did not need replacing. The closest honest measure on the page is the 110-complaint read and the Honda dealer message quoted in the article, both of which are labelled as what they are. ------------------------------------------------------------------------------ 10. FILES ------------------------------------------------------------------------------ ac-complaints-by-make.csv 35 makes; complaints, share, median failure miles, towed, belt/clutch-or-seal/contamination mentions, and A/C compressor bulletin count per make ac-complaints-by-model.csv 54 models with 3 or more complaints; counts, median failure miles, model-year span, condenser / evaporator / clutch-or-seal / warranty mentions ac-owner-reported-costs.csv 113 owner-reported dollar figures with vehicle, odometer and a short context snippet ac-recalls.csv 121 campaigns with population, fire flag and remedy type method.txt this file No personal data is published: no VINs, no names, no cities, no dealer addresses. Context snippets are short extracts of public complaint narratives that NHTSA already publishes in full. ------------------------------------------------------------------------------ CASE SENSITIVITY — SECTION ADDED 19 SEPTEMBER 2026 ------------------------------------------------------------------------------ NHTSA stores complaint narratives in ALL CAPS for records filed through roughly 2020 and in mixed case for records filed from 2021 on. A case-SENSITIVE search of the narrative field loses about 92% of the complaints filed after 2020, so this is the single largest way an analysis of this database can go wrong while still looking finished. Every query behind this file is case-insensitive, in two layers: * the complaint set is built with an FTS5 MATCH full-text search, which folds case before matching, so it cannot miss a mixed-case record; and * where the narrative is then tested with patterns, the narrative is upper-cased and whitespace-normalized first, so the patterns compare upper against upper. No query in this work uses GLOB or any case-sensitive comparison. This was not taken on trust. On 19 September 2026 every published folder was re-checked against the same database. Re-run here: the step-1 search MATCH 'compressor' returns 1,550 complaints today, exactly the figure published above. The same word searched case-sensitively in capitals would return far fewer.