8. Conclusion

An agent can be given accurate data and still answer wrongly, because a value's meaning is lost on its way to the decision. In the cases studied, that loss was silent: the value stays well formed, every call succeeds, and the answer arrives fluent and confident. Component tests and evaluation of answers do not follow meaning along the path, and the model at the end of it cannot restore a qualifier that never arrived.

Actor-Network Analysis and Design addresses this one decision at a time. From ANT it takes the practice of following the actors, the assumption that every hop may transform what passes through it, and the view of names and descriptions as actors that script what others do. From NFE it takes qualifiers, loss modes, control patterns and a test against a right answer computed from the rules. It turns these into eleven documents, a chart notation, a lifecycle with signed gates, and cycles that send work back when the analysis was wrong, when a referred finding is decided, or when the network changes.

Applied to two cases, the framework placed the hidden mediators the case accounts describe, classified every loss by mode, and added findings the accounts had not recorded: a missing decision owner, four inscription actors, a persuasive field and a rule gap. It stopped at analysis where the findings belonged to people rather than to tools, and it refused to certify a network on illustrative evidence. Applying it also found 38 problems in the specification itself; 20 fixes resolve 30 of them, and 8 remain undecided. What remains is to test it on recorded runs, before outcomes are known, by more than one analyst. Done, in this framework, means a network whose answers to one decision are tested against a computed right answer, with the records that say why.