The mantra that policy and management should be `evidence-based' is well
established. Less so are the implications that follow from `evidence'
being predictions of the future (forecasts, scenarios, horizons) even
though such futures define the actions taken today to make the future
sustainable. Here, we consider the tension between `evidence', reliable
because it is observed, and predictions of the future, unobservable in
conventional terms. For flood risk management in England and Wales, we
show that futures are actively constituted, and so imagined, through
`suites of practices' entwining policy, management and scientific
analysis. Management has to constrain analysis because of the many ways
in which flood futures can be constructed, but also because of
commitment to an accounting calculus, which requires risk to be
expressed in monetary terms. It is grounded in numerical simulation,
undertaken by scientific consultants who follow policy/management
guidelines that define the futures to be considered. Historical evidence
is needed to deal with process and parameter uncertainties and the
futures imagined are tied to pasts experienced. Reliance on past events
is a challenge for prediction, given changing probability (e. g. climate
change) and consequence (e. g. development on floodplains). So, risk
management allows some elements of risk analysis to become unstable
(notably in relation to climate change) but forces others to remain
stable (e. g. invoking regulation to prevent inappropriate floodplain
development). We conclude that the assumed separation of risk assessment
and management is false because the risk calculation has to be defined
by management. Making this process accountable requires openness about
the procedures that make flood risk analysis more (or less) reliable to
those we entrust to produce and act upon them such that, unlike the
`pseudosciences', they can be put to the test of public interrogation by
those who have to live with their consequences.