About
Search built for things that read
Most search APIs were designed for a person scanning a page of blue links. Agents do not scan. They read.
A model handed ten links and ten snippets has to fetch each page itself, strip the navigation and cookie banners, and guess which source to trust. That work is slow, it is repeated by everyone, and the guessing is where answers go wrong.
So Truscan does it once, properly. A query goes to many independent sources at the same time. Results are merged on a canonical URL, which turns agreement between sources into a ranking signal rather than a pile of duplicates. Then we fetch the pages that matter and return the readable prose, not the snippet.
The part we care most about is the score. Every result says why it placed where it did, broken into the signals that produced it. If a result looks wrong, you can see which signal carried it and decide whether to believe it. A number with no explanation is just a claim.
The difference
The same query, two answers
Read the rows as pairs. Each line on the left is the work your code has to do, and the line beside it is the work we did instead.
What most search APIs hand back
- Ten links and a snippet clipped at 160 characters
- One relevance number with nothing behind it
- The same page three times, once per mirror
- A 200 that quietly dropped half the sources
What Truscan hands back
- The readable prose of the page, boilerplate stripped
- Every signal that produced the score, itemised
- One entry per canonical URL, agreement folded in
- The failed URLs, named, with the reason each failed
Principles
How we decide things
Three rules that settle most arguments about what to build and how to charge for it.
Show the working
A ranking you cannot inspect is a ranking you cannot trust. Every result carries the signals that produced it, so you can audit a placement rather than take it on faith.
Charge for what ran
Money is counted in millionths of a dollar as whole numbers, never floats. A cached answer costs nothing, because it cost us nothing. A page that failed to load is not billed, and neither is a page we read only to rank the ones you asked for.
Fail honestly
Partial results say which URLs failed and why. Errors carry a stable code and a sentence written for a person. A useful failure beats a confident guess.
Try it on a query you already know the answer to
That is the fastest way to judge a ranking. Start with the monthly credit and see whether the score agrees with you.