Measuring curriculum alignment: one accredited CS degree against CS2013 and CS2023

Turaev, John, Al-Dabet, Awad, Zaki and Shuaib, "Measuring Curriculum Alignment across Topical Coverage, Competency, and Cognitive Depth: A Longitudinal Framework Applied to CS2013 and CS2023", arXiv:2606.19469 (v1 17 June 2026, v2 15 September 2026; read from v2). Four of the six authors are at United Arab Emirates University (UAEU), and the program they measure is their own: the UAEU BSc in Computer Science, 2025–2026 catalog, accredited by ABET's Computing Accreditation Commission, last reviewed in 2023, the year before cs2023 was published.

Tier T1, a preprint with its data described and its rater agreement reported. Read the caveat with it: a department measuring itself. The paper says it is one institution, and its numbers are the lenient end of a stated range.

This is the first source in the spoke that describes what one named department actually teaches, traced against the guidelines. It closes edge 1a, the instance this spoke had never looked at.

What was measured

41 catalog courses. 36 map to at least one CS2023 knowledge unit. Five map to none: natural sciences, basic biology and bioinformatics.
728 items from official course descriptions and syllabi: 195 catalog learning outcomes and 533 syllabus topics, each stored with its source path and a SHA-256 hash.
Method. Semantic retrieval proposes course-to-unit matches. An LLM checks each one against a written coverage rule. Two experts then validate each map on their own and settle differences. First-pass agreement was Cohen's κ 0.635 (CS2023) and 0.685 (CS2013). Retrieval alone is not enough: the best of seven retrievers reached F1 0.55, so the paper treats it as "a candidate generator, not a measurement."

What the program covers

Measure	CS2013	CS2023
Knowledge units covered	86/163 (52.8%)	78/161 (48.4%)
Strict threshold	45.4%	41.6%
Recommended hours covered	—	59.4%
Topics covered	—	~28%
Depth reached, among delivered units	74/78 (95%)	53/70 (76%)
Competencies stated	91%	90%

By area, CS2023 units covered: mathematical foundations 5/5, algorithms 4/5, operating systems 10/14, artificial intelligence 8/12, society/ethics/profession 8/11, software engineering 6/9, HCI 3/6, graphics 5/12, security 2/7, programming languages 2/22, systems fundamentals 1/9, and parallel and distributed computing 0/5. Against CS2013 the parallel figure is also zero (0/9).

Single-course dependence. "Of the 78 covered units, 54, that is 69 percent, are covered by exactly one course." Over two-thirds of what the degree covers would leave it if one course were dropped.

Depth. The program delivers most units it covers at the depth CS2013 asked for (95%), and far fewer at the depth CS2023 asks for (76%; at most 84% under a more generous level mapping). The paper reads this as the program having been designed to the older guideline, which fits the accreditation dates.

Against ABET

The paper also checks the program against the ABET computer-science criteria held on abet-computing-criteria. Algorithms, theory, software development, architecture, data management, networking (under-weighted) and operating systems are met. Parallel and distributed computing is "unmet (omission)". Programming languages are "sparse at unit level" (9%), and the paper counts the requirement as met through the required programming sequence.

What it does to this spoke

cs2023-accreditation showed that ABET enforces a subset of CS2013, and argued that nobody publishes conformance to the CS Core because nobody is assessed on it. This paper is a published conformance study, so that second claim was wrong as stated (corrected on that page). The first part survives, and here it has a number: an accredited program, assessed on the old guideline, delivers depth at the old guideline's level. It covers about half of the new one's units, and parallel computing not at all, while the program still holds accreditation.

The paper does not give CS Core coverage on its own. Its figures cover all 161 units, and it counts recommended hours, not ABET's semester credit hours. So coverage edge 7's question, whether the two documents' units can be mapped onto each other, is only half answered. One program now sits on both lists. Nobody has published the crosswalk.

Limits stated

One institution, measured by its own staff. Depth judgments are the least reliable step (κ 0.636), so unit-level depth verdicts are "indicative." The topic-level figure is an estimate. Coverage moves by about 7 points between the lenient and strict readings. Part of the gap may come from how finely the outcomes are written, not from missing content. The tools, prompts and maps are said to be released as supplementary material; no repository link appears in the text read.

cs2023 · abet-computing-criteria · cs2023-accreditation · meta-curriculum · executable-curriculum · synthesis

Cross-spoke. The pipeline (retrieval proposes, an LLM confirms against a rule, people adjudicate) is a measured case of LLM-assisted annotation, with retrieval's F1 ceiling reported. That method belongs to ../machine-learning-wiki; this page keeps what it found.