A recall ends with a repair. Sometimes the repair does not hold, and the manufacturer has to call the same vehicles back a second time. When that happens the new notice says so, in a sentence, naming the recall it replaces.
Those sentences are in the public file. They are not a column, they are not indexed, and nobody has collected them. 453 of them join one Canadian recall to another, across 17,844 recalls and fifty years.
| Recalls that replace or expand an earlier one | 453 |
| Typical gap between the first recall and the next | 404 days |
| Vehicles recalled three times or more | 27 chains |
| The longest chain | 4 recalls |
The second figure is the one worth sitting with. 404 days is more than a year. That is the time an owner spent believing the problem had been dealt with.
One chain, start to finish
This is the longest chain in the record, chosen by the engine rather than by us: MAZDA, 4 recalls, 1,222 days from the first to the last.
| Recall | Date | Vehicles | After |
|---|---|---|---|
| 2016259 | 2016-05-26 | 7,283 | the first |
| 2018038 | 2018-01-19 | 5,251 | 603 days later |
| 2018584 | 2018-10-25 | 328 | 279 days later |
| 2019496 | 2019-09-30 | 4,460 | 340 days later |
Recall 2018038 says: Additional Note: This recall supersedes recalls 2016-259 (Mazda Recall 9416E) and 2017-011 (Mazda Recall 0617A) for RX-8 models only. RX-8
Each notice names the one before it. That is what makes this a record rather than an inference: nothing here is two recalls guessed to be related because they share a brand and a month.
What a superseding recall is, and is not
It is not always a repair that failed. A manufacturer may replace a notice to widen it to more vehicles, to correct a detail, or to merge two campaigns into one. Those are housekeeping, and they sit in this dataset alongside the ones that matter.
So the verb is kept with every link. Supersedes and is an expansion of are different claims, and a reader who wants only the first can filter on it at the links route.
Two checks run on every link before it is published here. The recall named must exist in the file, which removed 5. And the ancestor must be older than the recall naming it, which removed 3. A link that fails either is counted and not shown, because a check that quietly drops what it cannot explain is not a check.
Where this comes from
- Transport Canada, Vehicle Recalls Database, the full monthly open file. The links are sentences inside the notice text, in English and in French, not a field.
- 17,844 recalls read end to end, four megabytes at a time, re-read weekly.
- The phrases searched for are published at the engine status route. If one is missing, this page is undercounting and we would like to be told.
- The filings belong to Transport Canada. The reading, the counting and any mistake in them are ours.
Built 15 September 2026. Corrections go on the corrections page.
The French half of the record · Canada and America · Download the record · How we work