This paper studies an episode of dissemination of wrong stock index values in real time due to a software bug in the Indian Nifty index futures market on the morning of January 18, 2006. The episode provides an opportunity to test various models of cognitive biases and bounded rationality highlighted in behavioural finance. The paper provides strong evidence against cognitive biases like "anchoring and adjustment" (Tversky and Kahneman, 1974) that one might expect under such situations even though the cognitive task involved is quite simple. The futures market tracked the true Nifty index which it could not see while completely ignoring the wrong Nifty index that it could see. However, the paper demonstrates that market efficiency failed in more subtle ways. There is evidence of a partial breakdown of price discovery in the futures markets and a weakening of the bonds linking futures and cash markets. This evidence is consistent with the centrality of "market devices" as argued in "actor network theory" in economic sociology (Muniesa, Millo and Callon, 2007 and Preda, 2006). Well functioning markets today depend critically on a whole set of information and communication technologies. Any failures in these material, socio-technical aspects of markets can make markets quite fragile even if behavioural biases are largely absent.