Complete · Competition
DDSC: Data Structures Visual Novel
AI classifies free-text answers into four verdicts while authored lessons and a deterministic engine own affection, retries, and endings.
Current status The Stack route and self-produced interface and art form a complete contest demo. A real AI grading deployment and learning-outcome evaluation are not complete.
Verified scope
Contribution boundaries and evidence
- Direct role
- Individual product concept, learning scenario, engine, grader integration, UI implementation, and validation
- Team boundary
- Individual contest entry
- Evidence visibility
- Private sources, public-safe summary only
- Contribution evidence
- An authored eight-beat Stack scenario with questions and answer anchors
- 23 Node tests for the deterministic engine and provider contract
- A Playwright playthrough from title to demo ending
Problem to solve
Data-structure material often ends with explanation and multiple-choice checks. This project teaches Stack through a visual-novel flow and asks the learner to explain concepts in free text, combining retrieval practice and immediate feedback in one game mechanic.
Responsibility boundary between AI and code
Authored scenarios are the authority for questions, answer anchors, explanations, and story dialogue. The AI grader compares a free-text response with the anchor and returns wrong, partial, correct, or critical. A deterministic engine calculates affection, bands, reteaching, problem order, and endings.
Provider keys remain server-side, while the static demo uses a browser mock implementing the same contract. A playable static page is not presented as a deployed real-AI backend.
Implementation and validation
The completed Stack route has eight beats: five for Jisoo and three cameo beats. Self-produced character, background, and title assets are integrated, with fallback UI when an image is missing.
Twenty-three Node tests cover affection bounds, a single grader call per submission, reteaching, character gauges, bands, and ending invariants. They include a complete lap through the real stack.json, while Playwright drives a mock playthrough from title to ending.
Limitations and next steps
Real-provider grading quality and learning outcomes have not been evaluated. Queue and Tree routes are not implemented. The publishable static demo uses only the mock grader; production AI would require a separate server deployment with secret and abuse controls.