Figure 9.

Testing Grendel's response to static rules. a, Diagnostic belief network generated for diagnosing ribosomal data-model conflicts. b, Probabilities of error for each step in the process prior to the introduction of static evidence. c, Updated probabilities of error after simulated interpretation error (metadata values: default_parameters_changed=yes, user_experience=very_little, interpretation_basis=personal_experience). The interpretation step is correctly singled out as the step most likely to be problematic. Notice that most of the prior probabilities remain unchanged because they are unaffected by the metadata we introduced. d, Updated probabilities of error after simulated journal article error (metadata values: peer_reviewed=no, journal_quality=poor). Based on the new metadata evidence, the same belief network now identifies the journal article reporting the data as a potential source of error.