When nutrients run out, cells reprogram gene expression within minutes. mRNA stability, poly(A) tail length, chemical modifications and translation each contribute, yet have almost always been studied one axis at a time, leaving unresolved whether they act in parallel, redundantly or convergently on individual transcripts. We exploit a unique strength of long-read nanopore direct RNA sequencing: transcript abundance, degradation rate, poly(A) length and m5C/m6A/pseudouridine status are read simultaneously from the same single molecules, measurements that each normally need a dedicated protocol. Profiling light and heavy polysome fractions of fed and 10-minute glucose-starved yeast, we fused these layers with Stochastic Translation Efficiency (STE), an AI readout of absolute protein synthesis independent of mRNA level, in a joint model whose feature effects are allowed to be program-specific. This resolves the transcriptome into five translational programs invisible to any single layer: translation machinery, ribosome biogenesis and amino-acid/catabolic metabolism induced, secretory glycosylation sharply repressed, and a near-neutral housekeeping bulk. The same feature can predict activation in one program and repression in another, and effects negligible transcriptome-wide become dominant once the right transcripts are isolated. The repressed glycosylation program is specified by a convergent stack of independent cis-signatures: weakest start-codon context, heaviest uORF load, most structured AUG, poorest codon optimality and depletion of stabilising poly(A)-binding-protein targets, a genome-wide uORF repression opposite to canonical GCN4 control. Modification remodelling concentrates in the translating pool, converging on Ty retrotransposon transcripts, while multi-hit transcripts reach the highest translation rates. Transcripts are thus not merely regulated by isolated events but pre-encoded, in their sequence, for the translational fate that starvation enacts.