bordeebook was effectively increasing their price by a factor of 1.2684/day, and as of Apr 8 priced the new book at $2,198,177.95. I'll guess that the used price ($35.54) implies that the initial pricing of the new book was about $50.

Log base 1.2684 of the ratio of the Apr 8 price to the estimated original price = log($2198177.95/$50)/log(1.2684) = 45 [days]. Ergo looks like the two algorithms started competing around Tuesday, Feb 22.

I'm sure my original price is off, but that actually won't make much difference. If it was $40, then the initial date was Feb 21; if $60, then Feb 23.

[EDIT] By the way, if the price was originally 99c you would still see the $2.2M price after just another two weeks (61 days total). Ah, exponents.

The original price candidate, derived from the logarithm, that is second closest to integer cents is 62.91998, with 162.85966 coming in 6th and the other positions filled by implausibly large numbers.

Neither number has integer cents, since the actual series rounded or truncated values between iterations. For that reason, a simple logarithm won't tell us what the original price was - you'd need a simulation of the rounding.

Good idea! Here it is in C: http://pastie.org/1824968 (You have to manually add the 'f' after the constants if you want to do the whole thing in floating-point, as well as uncommenting the correct Dollars_t definition.)

For the range $20.00 to $200.00, no exact matches to an integer original price (in cents) for the algorithms ceiling, floor, or round, in float or double. This was on a 32-bit Xeon server (EC2).