How long does a car defect take to reach the federal record?
Key facts
- 64.2% of the complaints on file about a failure arrived within 30 days of it; 45.0% within a week, 77.3% within 90 days, 91.2% within a year and 96.1% within two years. Computed on 420,008 complaints about failures in 2017-2023.
- Median reporting lag is 11 days as filed, 20 days excluding the same-day mass. The interquartile range is 1 to 74 days and the 90th percentile is 325 days.
- 16.1% of records carry a failure date identical to the filing date. Those records are less likely to involve a fire, crash or injury than later-reported ones, so the spike is a data-entry convention rather than urgency.
- 8.8% of complaints arrived more than a year after the failure they describe; 22.7% more than 90 days after it.
- The apparent 313-fold speed-up in reporting since 2010 (2,192-day median for 2010 failures against 7 days for 2026 failures) is an artefact of two opposite truncations. On the clean band the series is flat at 10 to 13 days.
- No component ranking and no state ranking survived cross-validation (tie-corrected Spearman 0.516 and 0.289). Both are reported here as failures rather than published as findings.
- Nothing here is a failure rate. The file has no registration denominator, so it can describe when reports arrive and never how likely a failure is.
How complete is the complaint record at each point after a failure?
The record fills quickly at first and then takes a year to close. The table gives the share of the complaints on file about a failure that had arrived by each horizon, on the clean 2017-2023 cohort, both as filed and with the same-day filing mass removed.
| Days since the failure | Share on file, as filed | Share on file, excluding the same-day mass |
|---|---|---|
| 0 (same day) | 16.1% | — |
| 7 | 45.0% | 34.4% |
| 14 | 54.0% | 45.2% |
| 30 | 64.2% | 57.3% |
| 60 | 72.6% | 67.3% |
| 90 | 77.3% | 73.0% |
| 180 | 84.6% | 81.7% |
| 365 | 91.2% | 89.5% |
| 730 | 96.1% | 95.4% |
The curve is capped at two years because the most recent cohort in the band, failures in 2023, has only 1,012 days of observation. Computed on the 2017-2022 cohorts alone, which can support a three-year horizon, the day-1,095 point is 97.9%.
This is the figure with a practical consequence. Any tool that screens a model for reported problems — NHTSA's own complaint lookup included — is reading a record that is roughly two thirds complete one month after a failure and still missing about one complaint in eleven a year later.
Does the same completion curve hold across different years?
Yes, and this is the part of the analysis that reproduces most cleanly. Each of the seven failure-year cohorts was computed separately rather than pooled, and they land within a few points of each other at every horizon.
| Failure year | n | By day 7 | By day 30 | By day 90 | By day 365 | By day 730 |
|---|---|---|---|---|---|---|
| 2017 | 61,744 | 44.1% | 61.8% | 74.2% | 89.5% | 95.3% |
| 2018 | 63,138 | 44.5% | 63.9% | 76.5% | 90.4% | 96.0% |
| 2019 | 72,203 | 46.3% | 66.3% | 79.8% | 92.3% | 96.7% |
| 2020 | 54,807 | 46.7% | 66.4% | 79.2% | 92.0% | 95.9% |
| 2021 | 48,666 | 46.1% | 65.5% | 78.3% | 90.8% | 95.6% |
| 2022 | 55,217 | 44.1% | 63.2% | 76.7% | 91.2% | 96.1% |
| 2023 | 64,233 | 43.2% | 62.4% | 76.5% | 91.7% | 97.0% |
The 30-day point spans 61.8% to 66.4% and the one-year point 89.5% to 92.3%. The pandemic years are not visibly different from the others.
Why does the data appear to show reporting getting 313 times faster?
Because the two ends of the series are measuring different populations, and this is the single easiest mistake to make with these two fields. Median lag by failure year looks like a collapse:
| Failure year | Median lag | What is actually observable |
|---|---|---|
| 2010 | 2,192 days | Only complaints filed 5+ years late — the file starts at 2015 receipts |
| 2012 | 1,283.5 days | Only complaints filed 3+ years late |
| 2014 | 354 days | Partly truncated: short lags missing before 2015 |
| 2017 | 12 days | Clean |
| 2019 | 10 days | Clean |
| 2021 | 10 days | Clean |
| 2023 | 13 days | Clean |
| 2025 | 11 days | Long lags cannot have accrued yet |
| 2026 | 7 days | Severely truncated: maximum observable lag is under a year |
A complaint about a failure in 2010 is in this file only if it was filed in 2015 or later, which means only the slowest reports about that year are visible at all — that is left truncation, and it inflates the early medians enormously. At the other end, a failure in 2026 has had at most a few months in which to be reported, so its long lags do not exist yet — right truncation, which deflates the recent medians. Between them sits the 2017-2023 band, where a complaint could have been filed the same day or five years later and either would appear. Across that band the median is flat between 10 and 13 days, with no trend in any direction.
The practical rule is that this field supports a distribution, not a time series.
What does the spike of same-day reports actually mean?
It looks like a data-entry convention, not a wave of urgent reports. 16.1% of records in the cohort carry a reporting lag of exactly zero, which would be a striking finding if those were the serious cases. They are the opposite.
| Same-day records (n=67,565) | Reported after at least a day (n=352,443) | |
|---|---|---|
| Involved a fire | 1.05% | 2.14% |
| Involved a crash | 1.93% | 4.85% |
| Involved an injury | 1.21% | 3.14% |
| References a recall | 11.5% | 11.4% |
If the same-day mass were urgency, fires and crashes would be over-represented in it. They are under-represented by roughly half on every severity flag. Recall correspondence is present in equal measure in both groups, so that is not the explanation either. The most plausible reading is that the failure date is entered as the filing date when the owner does not recall the exact day — which means the first bar of the distribution is partly an artefact of how the form is completed.
Because that mass sits at exactly the point where it has most leverage, it moves the headline a long way: the cohort median goes from 11 days to 20 when it is excluded, and the share arriving after 90 days from 22.7% to 27.0%. Every figure in this analysis is therefore given on both bases.
Do some components get reported faster than others?
We computed this ranking, it failed its own cross-validation, and we are publishing the failure rather than the ranking. Ordering the component groups by median reporting lag on failures in 2017-2019, and then independently on failures in 2020-2023 — same model-year 2015-2018 cohort, recall correspondence excluded from both — produces two orderings that correlate at only 0.516 across the 18 groups present in both.
| Component group | Rank on 2017-2019 failures | Rank on 2020-2023 failures |
|---|---|---|
| Electronic stability control | 1st of 18 (8 days) | 1st of 18 (3.5 days) |
| Service brakes | 1st of 18 (8 days) | 11th of 18 (10 days) |
| Steering | 3rd of 18 (8.5 days) | 3rd of 18 (6 days) |
| Visibility / wiper | 7th of 18 (11 days) | 2nd of 18 (5 days) |
| Engine | 8th of 18 (12 days) | 17th of 18 (14 days) |
| Wheels | 8th of 18 (12 days) | 3rd of 18 (6 days) |
| Seats | 15th of 18 (18 days) | 5th of 18 (7 days) |
| Suspension | 18th of 18 (25 days) | 8th of 18 (9 days) |
Ranks are taken within the 18 groups present on both bases so the two columns are comparable, and tied medians share a rank.
A ranking in which service brakes can be either the fastest-reported group or the eleventh of eighteen, and suspension either the slowest of all or the eighth, is not measuring a property of the component. The one group that did appear to separate from the pack — engine and engine cooling, at a 41-day median against 13 for the plain engine bucket — dissolved on inspection: 70% of its records are Ford (Escape 214, Edge 197, Fusion 136 of 1,041), so the number dates one manufacturer's coolant problem rather than describing engine cooling in general. That is the same trap that has caught model-level medians in this data before, and it is the reason a 1,000-record component figure needs its make composition checked before it is quoted.
The per-state version failed in the same way and for the same reason: across the 48 states with at least 400 complaints in each half of the period, the spread is only 9 to 15.5 days, the 48 states take just 9 distinct median values between them, and the split-half correlation is 0.289.
Does recall mail distort this the way it distorts complaint counts?
It runs in the opposite direction from the intuition and it barely touches the headline. When an owner contacts NHTSA about a recall — usually because the repair part is unavailable — that contact is filed as a complaint under the recall's own component label, and on component counts this contamination is large enough to reverse leaderboards.
| Basis | n | Median lag | Arrived after 90 days | Arrived after a year |
|---|---|---|---|---|
| As filed | 420,008 | 11 days | 22.7% | 8.8% |
| Recall-referencing only | 48,076 | 29 days | 35.2% | 12.4% |
| Excluding recall correspondence | 371,932 | 10 days | 21.1% | 8.4% |
Recall correspondence is slower, not faster — a median of 29 days against 10 for everything else — which makes sense, because a notification letter arrives and then the owner waits on a part before complaining. Our loose screen puts that correspondence at 15.51% of the corpus, against 19.7% measured corpus-wide by an earlier and independently written screen; that 4.2-point gap is the honest error bar on this kind of text classification, so the share should be read as a range rather than a point. Either way, removing it moves the cohort median by a single day, so unlike the count studies, the lag result does not depend on where that line is drawn.
Does vehicle age confound this, as it does mileage at failure?
No, and that is worth stating because our own prior work predicted it would. An analysis of the odometer-reading field in this same database found that median mileage at failure falls steeply with model year for no reason other than newer cars having driven less, and recorded the general warning that the same trap applies to any date-at-failure or age-at-failure cut. Measured directly on the reporting lag, it largely does not:
| Model year | n | Median lag | Arrived after a year |
|---|---|---|---|
| 2008 | 14,879 | 9 days | 7.5% |
| 2010 | 16,279 | 8 days | 6.7% |
| 2012 | 27,404 | 11 days | 8.0% |
| 2014 | 34,114 | 11 days | 8.8% |
| 2016 | 36,053 | 12 days | 9.5% |
| 2018 | 31,245 | 15 days | 11.2% |
| 2020 | 15,341 | 12 days | 10.1% |
| 2022 | 11,272 | 11 days | 9.2% |
The spread across fourteen model years is 8 to 15 days, against a two-fold spread in the mileage field across four model years. The reason is structural: an odometer reading accumulates with the age of the car, so an older vehicle can register a failure at a mileage a new one cannot reach, whereas a lag is the difference between two dates and has no such floor. The model-year 2024 figure is 30 days but rests on 671 records, which is too thin to quote. This is a useful narrowing of the earlier warning — the trap belongs to quantities that accumulate, not to every field with a date in it.
What a VIN check can and cannot tell you here
It cannot resolve complaint data to your car. NHTSA records complaints against a make, model and model year, and the VIN field in the public file is partial, so complaint data is a model-level signal. No vehicle history report can tell you that a particular complaint was filed about the specific car you are looking at, and any product that implies otherwise is overclaiming.
What a Zilocar VIN check does return, tied to that one vehicle: accident and damage records including airbag deployment, odometer readings over time and rollback indicators, salvage-auction and theft records, ownership history, sales-listing history, specifications, NHTSA and IIHS safety ratings, a valuation, and whether a recall is present on the car.
What it does not return, and where to go instead: whether a recall has actually been performed on that car — a dealer or the manufacturer can confirm the remedy status from the VIN; open NHTSA defect investigations, which are published by NHTSA itself and sit upstream of any recall; and the legal title brand, which is the business of NMVTIS and the state DMV rather than ours. We show the salvage auction record and the title and ownership history; the brand classification is theirs.
And the honest limit this analysis puts on everyone, including us: we have measured the lag on NHTSA's complaint record and found it roughly two thirds complete at 30 days. We have not measured our own ingest latency on the same basis, so nothing here should be read as a claim that Zilocar's records arrive faster. That would need its own study.
Method
Source. NHTSA ODI consumer-complaint period files COMPLAINTS_RECEIVED_2015-2019.zip, COMPLAINTS_RECEIVED_2020-2024.zip and COMPLAINTS_RECEIVED_2025-2026.zip from https://static.nhtsa.gov/odi/ffdd/cmpl/, public domain, downloaded 2026-10-11 07:14 UTC. The zip members carry timestamps of 2026-09-25 09:46:20, 2026-10-10 09:26:26 and 2026-10-10 09:27:30 and all three archives passed a CRC check. The combined FLAT_CMPL.zip download was deliberately not used because it has silently served stale extracts under fresh timestamps on three separate occasions.
Freshness assertion, run before any computation. The newest filing date across the three files is 2026-10-08, three days before retrieval. A stale-but-parseable file is the characteristic failure mode of this source, so the newest filing date is asserted rather than the retrieval date assumed.
Universe. The three files hold 1,111,402 component rows. Restricting to PROD_TYPE = 'V' drops 13,216 equipment, tyre and child-seat rows, leaving 1,098,186. Deduplicating on ODINO — one complaint can carry several component labels, while both dates are properties of the complaint — gives 784,010 distinct vehicle complaints with receipt dates from 2015-01-01 to 2026-10-08. No ODINO appears in more than one period file (0 of 784,010), so the files do not overlap.
Fields and the lag. · FAILDATE (position 8 of the layout) is the date the complainant gives for the failure; LDATE (position 17) is the date NHTSA received the complaint. The reporting lag is LDATE − FAILDATE in days.
Cleaning. · 12 records whose failure date falls after the receipt date and 659 whose failure date precedes 1980 are dropped as impossible rather than corrected, leaving 783,339 usable pairs (99.91%). Medians are used throughout because the distribution has a long right tail.
The cohort, and why it exists. The headline figures are computed on the 420,008 complaints whose reported failure falls in 2017-2023. Earlier failure years are left-truncated, because a file that begins with 2015 receipts can only contain complaints about a 2010 failure if they were filed five or more years late. Later failure years are right-truncated, because long lags have not had time to accrue. The 2017-2023 band is affected by neither.
Robustness, six checks, four of which bite, all reported either way. (1) The two truncations above make the naive by-failure-year time series worthless: it reads as a 313-fold speed-up (a 2,192-day median for 2010 failures against 7 days for 2026 failures) and the clean band is flat at 10 to 13 days. (2) The same-day mass, 16.1% of records, is a filing convention rather than urgency — those records are under-represented on every severity flag (fire 1.05% against 2.14%, crash 1.93% against 4.85%, injury 1.21% against 3.14%) with recall correspondence equal in both (11.5% against 11.4%) — and excluding it moves the median from 11 days to 20 and the after-90-days share from 22.7% to 27.0%. Both bases are published everywhere. (3) The component ranking does not reproduce: tie-corrected Spearman 0.516 between the orderings from failures in 2017-2019 and failures in 2020-2023 on the same model-year 2015-2018 cohort excluding recall correspondence, with service brakes moving 1st of 18 to 11th, suspension 18th to 8th and seats 15th to 5th; and the one apparent outlier, engine and engine cooling at a 41-day median, is 70% Ford records (Escape 214, Edge 197, Fusion 136 of 1,041), so it dates one manufacturer's defect wave. No component ranking is published. (4) The per-state ranking does not reproduce either: a 9-to-15.5-day spread (1.72x), only 9 distinct median values across the 48 states, and a tie-corrected split-half Spearman of 0.289 across the 48 states with at least 400 complaints in each half. No state ranking is published. (5) Fleet age does not confound this field, although our own earlier work on the odometer field predicted it would: median lag by model year runs 9, 8, 11, 11, 12, 15, 12, 11 days from model year 2008 to 2022, because a date difference does not accumulate with vehicle age the way an odometer reading does. (6) Recall correspondence is slower, at a 29-day median against 10 days for the rest, and removing it moves the cohort median by one day; our loose screen puts it at 15.51% of the corpus against 19.7% from an earlier independently written screen, a 4.2-point spread that is the honest error bar on the classification.
Rank correlations are tie-corrected, and that is not a detail. Median reporting lags are small integers, so a great many of the units being ranked tie: 6 of the 18 component groups share a median with another group on the first basis and 8 do on the second, and the 48 states take only 9 distinct median values between them. A Spearman calculation that breaks ties by input order rather than averaging their ranks therefore returns a figure that depends on how the rows happened to be sorted. Computed that way the two correlations came out as 0.492 and 0.466; computed correctly, with tied ranks averaged, they are 0.516 and 0.289. The published figures are the tie-corrected ones. The conclusion is unchanged in both cases — neither ranking reproduces — but the per-state correlation is substantially weaker than the naive calculation suggested, so the error was not in a harmless direction.
Cohort reproduction. The completion curve was computed on each of the seven failure-year cohorts separately as well as pooled. The 30-day point spans 61.8% to 66.4% and the one-year point 89.5% to 92.3%. The pooled curve is capped at day 730 because the 2023 cohort has only 1,012 days of observation; the day-1,095 point of 97.9% is computed on the 2017-2022 cohorts alone (n=355,775).
Independent corroboration, exact on both vehicles. · api.nhtsa.gov/complaints/complaintsByVehicle was queried for the 2017 Honda CR-V and the 2016 Toyota Tacoma and compared against our extract. Complaint counts match exactly (1,773 against 1,773 and 323 against 323) and the ODINO sets are identical with none on either side only. The lag statistics also match to the digit: median 7 days against 7 and 29 against 29, 75th percentile 50 against 50 and 103 against 103, same-day share 19.9% against 19.9% and 12.7% against 12.7%, and the 30, 90 and 365-day completion points agree with a gap of 0.0 points on all six. The API serves dates as MM/DD/YYYY against the flat file's YYYYMMDD, so the two paths parse independently.
Known limits. The denominator throughout is complaints on file at 2026-10-08, so the curve measures the share on file and not the share of all complaints a failure will ever generate. The file carries no registration denominator, so no figure here is or can be a failure rate. FAILDATE is complainant-reported, and check (2) shows that at least the same-day portion of it is conventional. Complaints are self-selected reports rather than an inspection sample, so publicity, recall mail and model volume all affect how many exist. NHTSA republishes these files continuously, so a reader re-pulling them will not reproduce these counts exactly; the two frozen CSVs below are the reproducible artefact.
Frozen datasets.
- Completion curve, pooled and per cohort, both bases:
https://zilocar-news.github.io/zilocar-rss/data/how-long-car-defects-take-to-reach-nhtsa-complaint-record-vin-check-2026-10-11.csv - Every robustness check with its result, including the two that killed a planned finding:
https://zilocar-news.github.io/zilocar-rss/data/how-long-car-defects-take-to-reach-nhtsa-complaint-record-vin-check-robustness-2026-10-11.csv
Sources
- NHTSA Office of Defects Investigation, consumer complaint period files,
https://static.nhtsa.gov/odi/ffdd/cmpl/—COMPLAINTS_RECEIVED_2015-2019.zip,COMPLAINTS_RECEIVED_2020-2024.zip,COMPLAINTS_RECEIVED_2025-2026.zip. Retrieved and analysed 2026-10-11. Last verified 2026-10-11. - NHTSA complaints API,
https://api.nhtsa.gov/complaints/complaintsByVehicle— used as the independent corroboration path for the 2017 Honda CR-V and 2016 Toyota Tacoma. Last verified 2026-10-11. - NHTSA ODI flat-file record layout, field positions for
FAILDATE(8),LDATE(17),COMPDESC(12),STATE(14) andPROD_TYPE(46), verified empirically against the files themselves on 2026-10-11. Last verified 2026-10-11. - National Motor Vehicle Title Information System,
vehiclehistory.gov— the authority on legal title brands, cited here because this analysis does not and cannot speak to them. Last verified 2026-10-11.
