If it is not recognisable as an instance of one genre, then it must be another genre.
I disagree with this position.
indeed; this would be genre à la old-style literary criticism (probably being ungenerous even to them). The story can be 'made to be true' trivially by defining genre in such a way that it is a unique label that can be attributed to any text, but that is an unnecessary modelling move that weakens any account and has a dubious relationship to properties of the relevant semiotic system(s).
Since genre is a discourse phenomenon (if it's anything), attributions of genres are attributions of hypotheses concerning relevant conventions of interpretation. The lexicogrammar and semantics are well capable of carrying patterns attributable to multiple sources of conventional interpretation. Any instance of language use can draw on and signal relevance of a variety of genres simultaneously or across logogenesis. Models are required that do not promote 'mixing' as something special, but as the norm. Otherwise genre change over time becomes more difficult to account for than necessary, rather than falling out of the model as a prediction. Surprised that one should consider anything else these days...
 It is Bateman's de novo theorising on genre that is unnecessary. In SFL theory, a genre is Hasan's (1985) term for a text type — that is, a genre is a register viewed from the instance pole of the cline of instantiation. Every text, by definition, is a type of text, even if it is the only token of its type.
 On the SFL model, where genre variation is a point on the cline of instantiation, rather than misconstrued as a stratum of context, genres vary by the frequencies of feature selection at the stratum of semantics, together with those of the strata below that realise semantic selections.
If genre variation is viewed from the system pole of the cline, then it appears as register variation. Registers vary by the probabilities of feature selection at the stratum of semantics, together with those of the strata below that realise them. This means that registers shade into one another, since they may share some probabilities, and it means that a given text (instance) can be attributed to different registers — and so: different genres — depending on which features are focused on as criterial. Halliday & Matthiessen (2014: 73):
In general, like any other language, English needs to be interpreted and described as an assemblage of varieties – varieties that are differentiated along different dimensions, with fuzzy boundaries.It is the probabilistic interpretation of variation that provides a window on the evolution of language and its subpotentials, as Halliday & Matthiessen (2014: 73-4) point out, taking the perspective of lexicogrammar:
the system of lexicogrammar is probabilistic in nature, and probabilities vary across varieties of English – dialectal, codal and registerial varieties. If we include probabilistic information in the description of the lexicogrammar, we also pave the way for interpreting the system as one that is always in the process of becoming, not one that is in a frozen state of being: the evolution of language involves gradual changes in probabilities, over long periods of time but also over much shorter periods.