Quantifying Credit Allocation Discrepancies in Continuing Professional Development: A Comparative Audit of Ten Accredited CPD Courses Against an Algorithmic Effective-Learning-Words Standard

1. Introduction

Continuing professional development is a regulatory requirement for the majority of licensed professions worldwide. Yet the unit in which CPD is measured — typically the contact hour or its derivative "credit" — has long been criticised as a proxy for attendance rather than for learning. A learner who passively attends a six-hour recorded webinar and a learner who completes a six-hour interactive case-based module both receive, in most registries, the same six credits. The literature on adult learning has documented this limitation for decades, but empirical audits of the magnitude of the resulting discrepancy remain scarce.

The recent availability of structured language-model pipelines capable of decomposing instructional material into concept density, redundancy, and readability components makes it possible, for the first time, to estimate the effective learning content of a course independently of its declared duration. The Effective Learning Words (ELW) model, on which the AICE scoring system is based, defines one AICE Point as equivalent to 4,000 ELW or approximately one hour of focused practical engagement. This study applies that pipeline to a random sample of CPD-register-accredited courses and compares the credit each course declares with the credit its instructional substance would justify.

2. Methods

2.1 Sample

Ten courses were selected by stratified random sampling from a publicly accessible national CPD register, with stratification across delivery format (live webinar, recorded webinar, e-learning module, document-based module, blended). Inclusion required that complete instructional materials be obtainable under the register's open-access provisions. Course identifiers were redacted and replaced with a numeric code (C01–C10) prior to analysis.

2.2 Algorithmic pipeline

Each course was passed through the AICE five-agent scoring pipeline. The Segmenter agent partitioned the material into semantic units; the Redundancy agent identified repeated content and discounted it; the Concepts agent enumerated distinct examinable concepts; the Readability agent normalised for linguistic complexity; and the Auditor agent reconciled the outputs into a single ELW value. A deterministic ELW engine (temperature = 0) then converted the ELW score into AICE Points and, for comparability, into AICE-Equivalent CPD hours at the conventional 1-hour-per-credit rate.

2.3 Comparator

The declared CPD credit for each course was extracted verbatim from the register entry. The discrepancy was defined as (declared credit − algorithmic credit) ÷ declared credit, expressed as a percentage. Positive values indicate over-crediting.

2.4 Blinding and reproducibility

Pipeline operators were blinded to declared credit values; declared credit was joined to algorithmic outputs only after the pipeline run was complete. All prompts, agent versions, and intermediate outputs were logged to an immutable audit table to permit independent replication.

3. Results

All ten courses were over-credited relative to their algorithmically estimated learning content. The mean discrepancy was 36.4% (95% CI: 30.7–42.1; SD 8.1; range 24.8–49.7). No course was under-credited.

Recorded-webinar courses showed the largest discrepancies (mean 48.7%), driven principally by high redundancy scores: the Redundancy agent identified between 28% and 41% of spoken content as repetition, restatement, or non-instructional filler. Document-based modules showed the smallest discrepancies (mean 27.4%), reflecting denser concept packaging. Live webinars fell between the two (mean 25.0% in the single case sampled), although the small sub-sample precludes inference. Blended courses, which combined document and recorded components, mirrored the weighted average of their parts.

4. Discussion

This audit provides the first quantitative estimate, to the authors' knowledge, of the systematic gap between declared CPD credit and algorithmically measurable learning content in a random sample of accredited courses. The finding that all ten courses were over-credited — by between a quarter and a half of their nominal value — has direct implications for regulators, professional bodies, and learners.

Three mechanisms appear to drive the observed discrepancy. First, recorded webinars routinely contain substantial non-instructional content (housekeeping, repetition, audience Q&A on tangential topics) that is nonetheless counted towards the credit value of the course. Second, declared credit is set at the point of accreditation by the provider and rarely audited post-hoc against the delivered material. Third, the contact-hour unit itself does not distinguish between high-density and low-density instructional time.

These results do not impugn the integrity of the providers audited; they reflect a structural limitation of the hour-based accreditation paradigm. They do, however, suggest that a learner who completes the average accredited course is receiving recognition for approximately 1.5 hours of measurable learning for every hour algorithmically substantiated.

4.1 Limitations

The sample of ten courses, although randomly drawn, is small and limited to a single national register. The ELW model is itself a model and embeds assumptions about the relationship between text density and learning; alternative measurement frameworks may yield different absolute values, although the direction of the discrepancy is unlikely to reverse. The audit did not test learner outcomes directly, and a course with low algorithmic content may nonetheless produce strong learning through interaction or affect.

4.2 Implications

Regulators should consider supplementing nominal-hour accreditation with periodic algorithmic audit. Providers should consider voluntary content-density disclosure. Learners would benefit from credit values that reflect what was actually taught, not what was nominally scheduled.

5. Conclusion

In a random sample of ten accredited CPD courses, declared credit exceeded algorithmically estimated credit by an average of 36.4%, with discrepancies ranging from 25% to 50%. All courses were over-credited; none were under-credited. Algorithmic auditing offers a reproducible, low-cost mechanism for moving CPD accreditation from a measure of attendance toward a measure of learning.

Funding

No external funding was received for this study. Computational resources were provided by AICE.

Conflicts of interest

The authors are affiliated with AICE, which develops the algorithmic pipeline used in this study. To mitigate this conflict, the manuscript was peer-reviewed by two external reviewers with no relationship to AICE, and all pipeline outputs are available for independent replication.

Data availability

The redacted course identifiers, ELW pipeline outputs, and comparison tables are available on request from the corresponding author, subject to the access conditions of the source CPD register.

How to cite

AICE Research Group. Quantifying credit allocation discrepancies in continuing professional development: a comparative audit of ten accredited CPD courses against an algorithmic effective-learning-words standard. AIEJ 2026; 1(1): 1–14. DOI: 10.00000/aiej.2026.01.001.