Orthogonal
competency
isolation.
Two axes. Neither influences the other. The research that follows shows what that architectural independence produces — and what it measures.
Each point is a classified student writing sample. The absence of correlation between axes is the design — not a byproduct.
Two independent experts reviewed the same set of student writing samples and classified each student's mathematical ability and language development stage using the XODA rubric. They agreed on 87.85% of classifications.
In educational and computational diagnostic systems, the threshold for reliable annotation is 80% — established as the minimum acceptable for trustworthy classification. XODA cleared it. On authentic student writing from North Texas Title I classrooms. On the population this system was built to serve.
That result means a teacher, a specialist, and an administrator can look at the same XODA output and reach the same conclusion about what a student knows mathematically — independently of how the student wrote it down.
κ = 0.8785 · 95% CI [0.802, 0.955] · Cohen's kappa · Offline, batch-processed · Artstein & Poesio (2008) minimum κ ≥ 0.80
κ = 0.8785
95% CI [0.802, 0.955]
Offline · Batch-processed
N=97 · Math/geometry stratum
North Texas Title I Grade 2–3 corpus
How the measurement works.
XODA produces two independent classifications. One does not influence the other's output.
The XODA system incorporates a proprietary deterministic ruleset.
Language Development Stage
Classifies where the student is in acquiring English academic vocabulary, including phonetic drift patterns, code-switching behavior, and morphological transfer from Spanish.
Mathematical Cognitive Signal
Classifies the student's demonstrated reasoning structure independently of the orthographic surface. This axis does not penalize abbreviated, phonetic, or mixed-language writing. It reads the math.
Status
NSF SBIR Phase I
Submission in preparation.
PI: Mario Oscar Pureco-Razo
Co-PI: Antonieta Ceron-Ponce
mar&mar ideas products & more, LLC · Fort Worth, Texas
BEA 2027 — Manuscript
Peer-reviewed manuscript under preparation for BEA 2027 (ACL Workshop on Innovative Use of NLP for Building Educational Applications). κ = 0.8785, 95% CI [0.802, 0.955].
Raw student text never reaches any AI model. FERPA and HIPAA Safe Harbor compliant.
No student-identifiable data is used in any published output. District identity is not disclosed in any published research output. District identification is available only to IRB reviewers and journal editors under sealed confidentiality.