Enthusiasts know that solving a crossword puzzle is a messy business. You don't solve the clues in numerical order and write them in their proper place, converging in an orderly manner to a correct -solution, unless, perhaps, it's a quick crossword and you're an expert. Instead, you attack the clues rather randomly, guided only by a vague feeling of which ones look easiest to solve (some people find anagrams easy, others hate them). You cross-check proposed answers against others, to make sure everything fits. You detect mistakes, rub them out, write in corrections.

It may not sound like a rational process, but the end result is entirely rational, and the checks and balances - do the answers fit the clues, do the letters all fit together? - are stringent. A few mistakes may still survive, where alternative words fit both the clue and the words that intersect them, but such errors are rare (and arguably aren't really errors, just ambiguity on the part of the compiler).

The process of scientific research, says Haack, is rather like solving a crossword puzzle. Solutions to nature's riddles arrive erratically and piecemeal. When they are cross-checked against other solutions to other riddles, sometimes the answers don't fit, and then something has to be changed. Theories that were once thought to be correct turn out to be nonsense and are thrown out. A few years ago, the best explanation of the origin of stars had one small flaw: it implied that the stars were older than the universe that contained them. At any given time, some of science's answers appear to be very solid, some less so, some are dubious ... and some are missing entirely.

Again, it doesn't sound like a rational process, but it leads to a rational result. Indeed, all that cross-checking, backtracking, and revision increases our confidence in the result. Remembering, always, that nothing is proved to the hilt, nothing is final.

Critics often use this confused and confusing process of discovery as a reason to discredit science.