We document manipulation of a targeting system which used a poverty index score to determine eligibility for social welfare programs in Colombia, including health insurance. We show strategic behavior in the timing of the household interviews around local elections, and direct manipulation when some households had their eligibility scores lowered. Initially the number of interviews increased around local elections. After the algorithm was made public to local officials, the score density exhibited a sharp discontinuity exactly at the eligibility threshold. The discontinuity at the threshold is larger where mayoral elections are more competitive; and smaller in municipalities with less competitive elections, more community organizations and higher newspaper circulation.
