Dead metaphors can no longer be unpacked; they no longer embody semantic junction.
Halliday & Matthiessen (1999: 261):
Contrast this now with failure in a technical expression such as heart failure; in origin this was no doubt a grammatical metaphor for the heart fails, but the metaphorical quality has since been lost, or at least significantly weakened (the metaphor is "dead"), and heart failure is now the only congruent form. Likewise contrast he regretted his failure to act, agnate to that he had failed to act (or that he had not acted), where failure is a grammatical metaphor, with he always felt that he was a failure, where failure is now the congruent form and this is not a metaphorical agnate of he always felt that he had failed.
Halliday & Matthiessen (1999: 283):
In our interpretation, the text as it stands, with the grammatical metaphor left in, embodies semantic junction: it is not just a variant form, identical in meaning with its congruent agnate — it also incorporates semantic features from the categories that its own form would congruently construe. Thus engine failure is not synonymous with engines fail; it is both a figure consisting of participant ('engine’) and process ('fail') and an element (participant) consisting of thing ('failure') + classifier ('engine'). In other words, we need both analyses in order to represent it adequately.
This will always be true whenever the metaphor can be unpacked to yield a plausible more congruent form. And this is what distinguishes a grammatical metaphor from a technical term. Almost all technical terms start out as grammatical metaphors; but they are grammatical metaphors which can no longer be unpacked. When a wording becomes technicalised, a new meaning has been construed — almost always, in our present-day construction of knowledge, a new thing (participating entity); and the junction with any more congruent agnates is (more or less quickly) dissolved. If for example we said that engine failure had now become a technical term, what would we mean by this? We would mean that the semantic bond with a figure an/the engine fails had been ruptured (it could no longer be 'unpacked'); and that a new meaning, an abstract participating entity or thing 'engine failure' had come into being which had the full semantic freedom — to participate in figures, to admit of classes and properties, and the like.