PULSCAR ANALYSIS — WHAT A CASE-SENSITIVE SEARCH LOSES IN NHTSA'S COMPLAINT FILE PulsCar analysis of NHTSA public data, 20 September 2026. License: CC BY 4.0. Attribution: "PulsCar analysis of NHTSA public data, September 2026." ------------------------------------------------------------------------------ 1. THE FINDING, IN ONE PARAGRAPH ------------------------------------------------------------------------------ NHTSA's public complaint file stores owner narratives in ALL CAPS for records filed up to 2020 and in ordinary mixed case for records filed from 2021 on. The change is not documented in the field dictionary and nothing in the file marks it. Anyone who searches that text case-SENSITIVELY — a grep, a regular expression, a GLOB in SQLite, a case-sensitive spreadsheet filter — gets the old records and silently misses almost all of the new ones. The query returns rows, so it looks like it worked. ------------------------------------------------------------------------------ 2. WHEN THE CHANGE HAPPENED ------------------------------------------------------------------------------ Share of narratives containing at least one lowercase letter, by the year the complaint was received (lowercase-by-year.csv): 2015 0.0% 2019 0.0% 2023 98.5% 2016 0.1% 2020 0.1% 2024 98.9% 2017 0.1% 2021 55.9% 2025 99.0% 2018 0.1% 2022 97.9% 2026 99.2% 2021 is the switch year: it is mixed within itself. Before it, mixed case is a rounding error; after it, ALL CAPS is. Of 777,040 complaints received 1 January 2015 to 8 September 2026, 360,813 (46.4%) arrived from 2021 on, and 335,239 of those — 92.9% — contain lowercase text. ------------------------------------------------------------------------------ 3. WHAT IT COSTS, BY TERM ------------------------------------------------------------------------------ what-a-case-sensitive-search-loses.csv gives, for twenty common search terms, the count a case-insensitive search returns and the count an upper-case search returns, across all years and for 2021 onward. Across the whole file a case-sensitive search loses between 24% and 78% of the matches depending on the term. Restricted to complaints filed from 2021 on it loses 88% to 96%. Examples, all years / since 2021: engine failure 78.0% lost / 96.3% lost coolant 72.6% / 96.3% catalytic converter 58.1% / 94.1% check engine light 55.3% / 93.8% fire 52.6% / 93.0% brake 46.0% / 91.0% airbag 24.1% / 88.5% The spread between terms is not noise: a term that owners tend to write in the older, all-caps era loses less, because more of its matches are old. ------------------------------------------------------------------------------ 4. METHOD ------------------------------------------------------------------------------ Source: NHTSA's published complaint flat file (COMPLAINTS_RECEIVED), downloaded from https://static.nhtsa.gov/odi/ffdd/cmpl/ in September 2026 and loaded into SQLite without modification. 777,040 complaints, received 1 January 2015 to 8 September 2026. Field definitions: https://static.nhtsa.gov/odi/ffdd/cmpl/CMPL.txt "Case-insensitive count" is SQLite LIKE '%term%', which folds ASCII case. "Upper-case-only count" is SQLite GLOB '*TERM*', which does not. Lowercase detection is GLOB '*[a-z]*'. One row per complaint (ODINO); no de-duplication was needed. Both numbers in every row were produced by the same query against the same table in the same run, so the comparison is not affected by the file changing between downloads. ------------------------------------------------------------------------------ 5. LIMITS, STATED PLAINLY ------------------------------------------------------------------------------ - This measures the NARRATIVE field only. Structured fields (make, model, component, dates, flags) are unaffected. - The twenty terms are ones PulsCar happened to need. They are not a random sample of what anyone searches, and the per-term percentages should be read as examples of the size of the effect, not as a distribution. - "Lost" means "not returned by an upper-case-only search". It does not mean the complaints are missing from the file — they are there, and a case-insensitive search finds them. - Complaints are self-reported. Nothing here is a failure rate, and this file makes no claim about any vehicle. - We did not ask NHTSA why the formatting changed, and this note does not speculate. It describes the file as published. ------------------------------------------------------------------------------ 6. WHY PULSCAR PUBLISHED THIS ------------------------------------------------------------------------------ We hit it ourselves. Several of our own published analyses were re-checked in September 2026 for this exact error; four had to be corrected, and one headline finding was retracted because the corrected figure fell below its own baseline. Every one of our 124 published research folders now states how it handles case, and every one was re-verified against the database on 19 September 2026. Contact: admin@pulscar.io