Reliability is derived from approved ownership reports. Each report contributes its worst reported problem, weighted by severity, and the resulting rate is shrunk toward the category baseline so products with few reports are not ranked on noise. Scores are withheld entirely below eight effective reports. Age-based failure rates count only owners who have owned the product long enough to have reached the year in question.
Methodology version 1.0.0
How reliability is measured
A reliability archive that will not explain its own numbers is an opinion site. This page describes exactly how every score is produced, and where it should not be trusted.
How a score is built
- 01
Reports record ownership, not just problems
Every submission is an ownership record: what someone owns, how long they have owned it, how hard they use it, and whether anything went wrong. Problems are optional details inside that record. This matters more than any part of the formula. If the archive collected only problem reports it would have a count of failures with no idea how many working units exist, and every rate it published would be meaningless.
- 02
Each report contributes its worst problem
A report is penalised according to its most severe problem rather than the sum of its problems, weighted from cosmetic through to critical. Someone who describes one failure in careful detail should not count more heavily than someone who briefly mentions a catastrophic one.
- 03
Sparse products are pulled toward their category baseline
A raw failure rate from twelve reports is mostly noise. Each product's rate is blended with the average for its category, weighted so the baseline dominates when evidence is thin and fades as real reports accumulate. This is why a product with few reports will sit near the middle rather than at either extreme.
- 04
Reports are weighted by how much they can be trusted
Data from rental houses and repair shops, who see many units of the same product under hard use, carries more weight than an anonymous form submission. Reports with supporting evidence carry more weight than those without.
- 05
No score is published below a minimum sample
Below 8 effective reports no score is shown at all. The underlying reports are still published, because individual accounts are useful even when an average is not.
- 06
Failure rates are reported against time
Shutters fail in year four, not week one. Age-based failure rates count only owners who have owned a product long enough to have reached the year in question, so a recent release cannot look flawless simply because nobody has had time to break it yet.
What confidence means
Confidence describes how much the score itself can be trusted. It is always shown next to the number.
| Not enough data | Too few ownership reports to publish a score. The reports below are shown as individual accounts, not as a rate. |
|---|---|
| Low confidence | Based on a small number of reports, or on reports covering short ownership periods. Treat as indicative only. |
| Moderate confidence | Based on a reasonable sample, though not yet enough to be precise about small differences. |
| High confidence | Based on a large sample with substantial ownership history. |
| Very high confidence | Based on an extensive sample with long ownership history across many owners. |
Current parameters
These values are stored as data rather than code, so a change to the methodology is recorded and versioned.
| Minimum reports before a score is published | 8 |
|---|---|
| Category baseline strengthHow much evidence the category average is treated as being worth | 12 reports |
| Weight on failure rate | 65% |
| Weight on owner satisfaction | 20% |
| Weight on whether problems were resolved | 15% |
Category baselines
The average reported problem rate per category, and how many reports it is derived from. Sparse products are pulled toward these figures.
| CamerasFrom 0 reports | 15.0% |
|---|---|
| LensesFrom 0 reports | 15.0% |
Known limitations
Stated plainly, because a reliability archive that hides its weaknesses does not deserve to be trusted with the strong claims.
Self-reported data is biased toward problems
People are more motivated to report a failure than to report that nothing happened. The archive counters this by making no-problem ownership records easy and explicitly valuable, but the bias never fully disappears. Treat reported rates as an upper bound rather than a true failure rate.
Popular equipment attracts more reports of everything
A product with a large, engaged community generates more reports of both failures and satisfaction. Rates are expressed as a share of that product's own reports, which controls for this, but comparisons between a mainstream body and a niche one are still less reliable than the numbers suggest.
These are not manufacturer failure statistics
No manufacturer publishes real failure rates, and nothing here is derived from warranty data. Every figure describes what owners told this archive. Service advisories are the exception: those are documented manufacturer statements and are labelled as such.
Usage varies enormously and is only partly captured
A lens used for studio work and one used in coastal wind live very different lives. Usage intensity and environment are recorded, but they cannot fully explain away differences between products.
The methodology will change
Every score is stored with the methodology version that produced it, and raw reports are never overwritten. When the formula improves, history is recomputed rather than discarded, and the previous version stays on record.