Thanks; it looks like what you've (1988) called "discriminatory power" is what I'd call "precision", and so we might agree on this - at least up to what the more intuitive perspective is. (Of course, these words have all have different baggage, but to me the relation to the statistical primitive is important).

Re: confidence/p value curves, I wrote about it some years back in another blog post (https://bayesfactor.blogspot.com/2016/07/stop-saying-confidence-intervals-are.html#more, section "Where to go from here?"). I regard a typical p value report as giving us continuous information about a single hypothesis at any alpha, and a confidence interval as telling us about all hypotheses at a single alpha. So the natural generalization is to look at all hypotheses at every alpha. In this paper I also try to adopt a range of alphas with CIs for reporting purposes: http://www.stat.columbia.edu/~gelman/research/published/multiple_perspectives.pdf (with debatable success, clarity-wise). So p value curves seem like a good idea.

But, I'm unsure about severity in light of Yanofsky's insight here: https://itschancy.wordpress.com/2019/02/09/baroque-possibilities-for-constructing-sev-in-sequential-trials/, which stem from the difficulty of assigning all possible outcomes in a sequential trial to a single evidence/distance/accordance continuum. This may indicate something strange about severity or sequential trials; I don't know.