As computational thinking becomes central to an expanding array of disciplines, a core barrier to high-quality K-12 computer science (CS) education is insufficient instructional capacity: there are not enough qualified CS teachers, they are not distributed equitably, and those who are teaching often lack the tools to assess whether students are developing computational thinking skills. This dissertation examines both dimensions of that capacity problem.

To examine the first dimension, this work introduces the Equity-Driven Proportional Allocation of Resources (EDPAR) framework, which quantifies equitable access through three complementary metrics: deficit (each school's distance from a student-weighted ideal), disparity (the correlation between those deficits and school demographics), and allocation (the variability of deficits across a region). Applied to Georgia's public school districts, EDPAR revealed that 30 of 76 multi-school districts had no AP CS teachers at any school, and that schools serving predominantly Black communities were disproportionately likely to have none.

To examine the second dimension, this work introduces Surface Concept Analysis by Notation (SCAN), a curriculum-grounded methodology for extracting programming concepts from student source code through surface pattern analysis. SCAN produces a conceptual fingerprint: a structured record of the presence and placement of programming concepts in a submission, designed for teacher-facing formative interpretation rather than grading. Validated through the ARCHIE prototype against human annotation of 312 Java snippets, SCAN achieved substantial agreement for 12 of 14 AP CSA Unit 2 concepts, with primitive-class exact match rates averaging 93.5%. Two participatory design studies with K-12 CS teachers confirmed demand for concept-level visibility tools, and teachers identified three primary use cases: understanding individual student conceptions, supporting administrators evaluating CS course quality, and enabling computational thinking analysis.

Together, EDPAR and SCAN offer scalable computational approaches to both dimensions of the instructional capacity problem: EDPAR provides the means to answer the question of who has access with data rather than assumptions, while SCAN makes it possible to answer the question of what students are learning at the concept level.