Known Limitations
Every dataset has a shape, and knowing ours is the difference between using it well and being misled by it.
Last updated 2026-08-17
Sampling bias is real
People who complain in public are not a random sample. They skew technical, English-speaking, and toward industries with active online communities. A quiet industry is not a problem-free industry — it is an under-sampled one.
Cross-source agreement partly compensates, which is why source span is weighted heavily in evidence strength.
Recency and staleness
A complaint from three years ago may describe a problem that has since been solved. Dates are shown on every piece of evidence for exactly this reason — check them before concluding a gap still exists.
What a model can get wrong
- Summarising a complaint more confidently than the original warranted
- Naming a competitor that does not exist, which is why we prefer an empty list to a speculative one
- Estimating revenue for a category where we hold no comparables
The honest bottom line
This narrows a field of infinite possibilities to a few dozen well-evidenced ones. It does not tell you what to build, and it is not a substitute for talking to the people who have the problem.