A single proficiency score routinely conceals which component of second language competence is actually limiting a learner's progress. Two learners placed at the same overall band can be held back for entirely different reasons: one carries a stable, rule-governed error pattern into otherwise fluent speech, the other has intact grammar but a narrow phonological repertoire that blocks intelligibility. Instruction pitched at the aggregate score treats both learners identically and helps neither. This paper develops a diagnostic framework for identifying which specific listening, speaking, reading, writing, vocabulary, grammar, pronunciation, fluency, and pragmatic barriers are driving a stalled proficiency trajectory, and for distinguishing a genuine component deficit from a measurement artifact, a motivational barrier, or a contextual constraint. Building on interlanguage theory, error analysis, CEFR-referenced assessment, and individual-differences research on aptitude, working memory, motivation, and anxiety, the framework organizes thirty-six diagnosable constructs into nine domains nested within a five-layer evidence architecture spanning outcome, process, context, implementation, and governance evidence. It specifies falsifiable propositions linking domain-specific diagnostic signals to proficiency growth, a measurement matrix defining how each domain should be operationalized and reported, and a staged pathway for moving diagnostic practice from a single classroom pilot toward validated, transferable use. The framework does not claim an empirically tested intervention: no dataset is analyzed here, and every mechanism is stated as a testable proposition rather than a finding. Its contribution is a shared, falsifiable vocabulary for locating the specific barrier behind a stalled score, so that remediation can be targeted rather than generic.