Make consistency judgement is a standard practice of teaching/learning injury. However, what benchmarks and criteria support teachers? This is the question our study is asking about teachers’ assessment of a series of narrative texts written by students in 3. The assumption I make is that teachers show, through their annotations and comments, implicit definitions of consistency which are often far removed from the definitions declared on that concept. Our goal is dobserver how their “proto-theories” are linked to the conscious definitions of coherence. Indeed, I believe that this is the comparison of these two levels, which will enable us (1) to analyse judgments (2) to reflect on a possible formation of consistency. We have built a theoretical framework consisting of operating concepts from targeted research fields: textual analysis of speeches; textual semantic; textual psycholingualism. They represent the successive filters applied to the same data, i.e. fifty three judgements from the same text sequence. This allows us to analyse the differences and convergences of judgements and criteria regarding text and global effects in the text. We show that teachers have non-homogeneous reading patterns, asking the question of the relationship between the local and the overall level in the construction of coherence. All the data analysed suggest that the theoretical approaches we have selected seem to be problematic in the context of training. The role of training in judging the consistency of scriptor apprentice texts could be to help the reader-proofreader to raise awareness of the way he manages the various levels of textual constraint.