The selection logic is the part worth sitting with. A phage that infects efficiently clears the sensitive bacteria first, and what it leaves behind is the fraction that was already partly resistant, with room to grow and time to improve [5]. Escalation, the habit antibiotics teach, works against you under those dynamics: the second-line phage arrives to face a population the first-line phage selected. The remedy in the model is to make survival require several genetic changes at once, which is the argument for putting everything in at the start [7].
"Several" is not a number, and that is the edge of what the published account supports. Turning a genetic barrier into a cocktail size would need per-phage resistance frequencies and mutation rates in the isolate you are actually treating; the account names three factors that decide the outcome and gives a magnitude for none of them [4] [12]. Diversity is described as overwhelming the bacteria's capacity to adapt fast enough [6], which is a direction rather than a dose.
Provenance matters here too. The work builds on an earlier model calibrated against data from phage therapy in a live mouse, then extends it to humans and to settings with many phages and many strains of differing resistance [3]. That is a reasonable way to build theory and a weak way to predict a patient. The reported account also does not describe how the model treats immune clearance of the phages, or pairing with antibiotics [14], and both of those shape whether a large immediate dose behaves in a person the way it behaves in equations.
There is a tension inside the recommendation. A cocktail is assembled for one patient facing one infection, which the authors' own framing calls a complicated form of personalized medicine [10]. Matching phages to an isolate consumes exactly the hours that "immediate" is meant to save. Read that way, the advice bites less on the clock than on allocation: use the breadth you have now instead of banking phages for a rescue round [7].
Perelson's compression, that cocktails should be diverse, sufficient and immediate [8], will travel further than the paper's conditions do. The part I would hold onto is the measurement it implies. If a bacterial population's pretreatment resistance level is one of the things deciding the outcome [4], then a phage programme that does not profile the isolate against each phage before dosing will keep producing case reports nobody can pool, which is close to where the field sits now, with success and failure both hard to explain after the fact [11].
