Represented Pathway Legibility: A Graph-Based Scoring Framework with Cross-Vendor Synthetic Mechanical Reproducibility

Abstract

Existing measures of postsecondary transition address important neighboring constructs, including factual college knowledge, college-planning behavior, and college-going self-efficacy. This measurement-development report targets a different object: the structure of a represented pathway from a current state to a declared domain target. The proposed task elicits an open directed pathway map and scores a non-collapsed Path Legibility Vector (PLV): Coverage (COV), Relation Accuracy (REL), Sequence Resolution (SEQ), Contingency Resolution (CONT), Next Action Validity (NAV), Unsupported Edge Rate (UER), and confidence/correctness Calibration (CAL). Mechanical development was conducted in a closed fictional Northbridge domain with a frozen reference graph and predeclared synthetic gates. A cross-vendor v0.1.1 calibration achieved exact synthetic reproducibility but exposed a route-family attribution artifact: unsupported cross-route edge endpoints could inflate coverage in a competing route family. A frozen stress test confirmed the defect on three predeclared maps. Version 0.2.1 repaired route-specific node attribution and added deterministic final-output round-half-even reporting. Two isolated blind scorers from distinct vendor families (Anthropic Claude Code and Google Gemini) independently reproduced all 12 stress maps and all 8 repair guard cases; all predeclared Gates A-F passed. The result supports synthetic mechanical reproducibility of the frozen scoring procedure and the specific route-attribution repair. It does not establish human inter-rater reliability, expert-reference validity, construct validity, predictive validity, educational validity, or real participant-map reliability.