PULSCAR ANALYSIS - P2002 (DIESEL PARTICULATE FILTER EFFICIENCY BELOW THRESHOLD) AND DIESEL PARTICULATE FILTER TROUBLE IN NHTSA'S PUBLIC FILES PulsCar analysis of NHTSA public data, October 2026. License: CC BY 4.0 Attribution: "PulsCar analysis of NHTSA public data, October 2026." Published alongside https://www.pulscar.io/blog/p2002-code ------------------------------------------------------------------------------ Prepared October 2026. Source data: NHTSA public flat files (complaints received 1 January 2015 to 4 October 2026 - the latest received date in the files; manufacturer communications; recalls since 2010) and documents downloaded one at a time from static.nhtsa.gov. 1. Complaints (618) First search (566): every complaint whose narrative names P2002 or P2003 by any spelling we searched (a letter O for zero, a space, hyphen or period after the P, the suffix 00); or a companion filter code (P244A to P244F, P245x, P246x, P242F, P2BAD, P2BAE, P20EE); or a particulate filter in owners' words (particulate, DPF, soot trap, diesel or exhaust filter, particle filter, FAP); or the word regeneration, or soot, in a narrative that also shows a diesel engine (diesel, Duramax, Cummins, Power Stroke, TDI, BlueTec, EcoDiesel, CRD, a 6.4, 6.6 or 6.7 engine, DEF, AdBlue). Eighteen rows of electric-car makers that matched only the regeneration or soot words were left out, because there "regen" is regenerative braking. A wider search by symptom (a diesel vehicle with exhaust or emissions wording and a power or warning word) was tried first and rejected: it returned 450 narratives, almost all about the urea (DEF/SCR) system, a fuel pump or an exhaust leak, none naming the filter. Second search (52): a completeness check of the whole complaint file, run after the first batch was sorted, with other spellings and phrasings: regeneration without the word diesel, Ford's "exhaust cleaning" and "drive to clean exhaust" messages, misspellings of regeneration, a particulate-matter filter sensor, and a few heavy-truck narratives. All 52 were added. 62 Ford 6.7L narratives about an exhaust-gas-temperature sensor shutdown that never name the filter, regeneration or soot were looked at and not added. found_by in the complaints file says which search found each complaint. Sorting: the sorting rules (sorting-rules.txt) were written, dated and locked before any complaint was sorted. Every complaint was sorted twice, independently, by the same written rules, on its whole text. The two sortings matched on the main category in 576 of 618 (93.2%), on the loss-of-power flag in 603 (97.6%), on the short-trip flag in 613 (99.2%), on the recall flag in 606 (98.1%) and on the returned flag in 598 (96.8%). Where they differed on any field (88 complaints) the full text was read and the difference decided under the written rules; where the rules were silent a rule was added (R1 to R12, in sorting-rules.txt). The resolution column says "agreed" or "settled by a written rule", with the rule number. A sweep of the agreed sortings for contradictions with those rules (complaints naming P2002 or a companion code, complaints marked not relevant that contain filter wording, and the short-trip flag against the text) found none that needed a change. Categories: P filter named or replaced, C cleaned or regenerated only, F software or campaign only, S sensor or wiring, O another cause, K checked and cleared, U no diagnosis or repair stated, N not a filter event on the owner's vehicle. Flags: derate (loss of power, limp mode, stall or no-start), shorttrip (driving pattern tied to the trouble), recall (a recall, bulletin, update or program mentioned), returned (the trouble came back or the repair did not hold). Result: 405 relevant, 213 not relevant. Counts are counts of complaints, not cars. A complaint is evidence that an owner wrote to NHTSA; nothing here is a failure rate. 2. Documents (533 files, 488 distinct documents) Search: NHTSA's index of manufacturer communications, entries whose summary names P2002 or P2003, a companion filter code, or filter wording (particulate, DPF, soot, regeneration, exhaust filter): 1,866 entries. 990 are for heavy trucks, buses and engine makers (Mack, Volvo trucks, International, Freightliner, Western Star, Hino, Cummins engine bulletins and the like) and were set aside; 876 are for light-duty vehicles. Of these we downloaded the 145 that name P2002 in their summary, all those naming a companion filter code or filter wording once per group of entries with an identical summary, and 15 more found by the completeness check: 533 files (the file list is in p2002-documents.csv; 248 index listings with an identical summary and 115 whose summary mentions only the urea system or generic aftertreatment were not downloaded and are listed with the reason). Reading: 379 families of revisions and look-alikes. 178 representative documents (all that print P2002 or six or more filter words in their own text) were read in full; 154 further files (revisions, copies and look-alikes) were compared with their representative line by line; the remaining 201 families were checked for the codes and the filter wording in their own text, and 15 of them proved to be about the filter and were read in full. Files whose text would not read (a few dozen) were read from images of their pages; that read added eight papers that print P2002. Every verbatim quotation in the article was checked against the text of the named document; quotations from pages read only as images were not used. 164 files print P2002 in their own text (156 in the readable text, 8 from images); none prints P2003. 45 of the 533 files are exact copies, leaving 488 distinct documents. Model years, engines, build dates and coverage terms are as printed in the documents, never taken from NHTSA's index, which expands years. Coverage status "as of September 2026" is the authors' computation from printed terms. 3. Owner-reported amounts (38 figures in 33 complaints) One written inclusion rule (fixed before the figures were decided): INCLUDE a dollar figure when the owner states it as a price, charge, bill or quote for diagnosing, cleaning, regenerating, repairing or replacing the diesel particulate filter (or the exhaust/aftertreatment unit that contains it), its pressure or temperature sensors, or the repair of the filter trouble (including a stated total for "the DPF repair"), or a diagnostic fee asked for that filter event, on THIS vehicle. EXCLUDE: mileage written as 'NNk'; vehicle value, payments, trade-in, loans, buy-back or refund/compensation requests; repairs of other systems (turbo, injectors, EGR cooler, DEF tank/pump, transmission, engine internals) unless the narrative attributes the figure to the filter; totals that mix filter work and other work unless the narrative attributes the figure to the filter; restatements of a figure already counted in the same narrative (count once); hypothetical or other owners' costs; tow charges; rental cars; DEF fluid purchases. STATUS (from the owner's own words): paid (paid / cost me / spent), billed (bill, invoice, charged, cost), quoted (told, quoted, estimate, wants, would cost, needs), offered, not stated. GROUP: filter, clean (cleaning or forced/service regeneration), sensor, software, diag (diagnostic fee), mixed (filter plus other work in one figure), unspecified (a repair of the filter trouble whose parts are not named). 99 candidate figures in 72 complaints were looked at and 38 kept (26 quoted, 7 billed, 3 paid, 2 not stated); each exclusion has a short reason in p2002-amounts-left-out.csv. A figure that names filter work together with other work in one sum is shown only in the group "mixed". Check of the result: a third, independent sorting of 70 randomly drawn complaints (seeded draw, whole text, same written rules) matched the final category in 61 (87%), whether the complaint is a filter event or not in 64 (91%), the loss-of-power flag in 67, the short-trip flag in 70, the recall flag in 65 and the returned flag in 64. The differences are borderline readings, mostly between 'not a filter event' and 'checked, nothing found' or 'no diagnosis stated'. Shares are therefore approximate. 4. Limits The complaint file holds the cases where an owner wrote to NHTSA, in the owner's words, with no verification; owners whose trouble ended are mostly not in it. The shares reported (for example how many filter replacements were followed by a return of the problem) describe the complaints, not all trucks. Most complaints do not name a code, so the study set follows the filter and its regeneration, not the code string; only 8 print P2002. The document set holds what NHTSA's index lists; a maker paper that is not in the index is not in it, and papers for the same code differ by maker and market (P2002 is printed as a cracked or failed filter, as a monitoring fault and, in two Stellantis markets, as a blocked filter).