Pedagogical Engine

The Science of
Evaluation.

Most AI grading tools act as black boxes, emitting generic praise like "Good job!" or "Needs work." The RALE Engine is built differently. Every piece of feedback generated by our system is mapped to a strict, proprietary pedagogical taxonomy.

Hierarchical Mapping

Instead of a single holistic score, our pipeline dissects student input across distinct domains: Lexical Resource, Grammatical Range, Coherence, and Task Achievement. Errors aren't just corrected; they are strictly classified using standard pedagogical identifiers.

This enables institutions to run programmatic queries across thousands of assessments, identifying macro-trends like "Why are 8th graders consistently failing on GRA.AGREEMENT.SUBJ_VERB?"

Active Treev2.1.0_PROD
├── DOMAIN: Lexical Resource (LEX)
├── CATEGORY: Vocabulary Breadth
├── SKILL: Idiomatic Phrasing LEX.IDIOM
└── SKILL: Collocation Accuracy LEX.COLLOC
├── DOMAIN: Grammatical Range (GRA)
├── CATEGORY: Syntactic Control
└── SKILL: Subject-Verb Agreement GRA.AGREEMENT
The government should implementing
GRA.TENSE.MODAL
Modal verbs must be followed by base verbs.
stricter policies regarding environmental safe
LEX.FORM.NOUN
Requires noun form 'safety'.
.

Live Token Tagging

RALE doesn't rewrite student essays into a perfect, sterilized block of text. It acts like a digital red pen, isolating exact error tokens (start/end indices) and injecting the relevant pedagogical correction.

  • Strict string-index anchoring ensures no hallucinations.
  • Preserves the student's original voice.
  • Maps every error directly to the scoring rubric.

Defensible Scoring

By aggregating taxonomy codes, the final grade is calculated deterministically rather than generated probabilistically. If a student loses 1.5 bands in Grammatical Range, the system can instantly point to the 14 syntax errors that triggered the penalty. This makes AI grading entirely defensible to students, parents, and administrative auditors.