The initial mesh contains five nodes. Running its local loop at aggressiveness four computes clarity as seventy five plus fifteen plus four, giving ninety four, and resilience as sixty five plus twenty, giving eighty five. At three pixels per authored percentage point the bars span two hundred eighty two and two hundred fifty five. These percentages are arithmetic fixtures, not measurements of student comprehension or model reasoning. The original works without its network graph library by retaining local controls and a fallback. It makes no language-model request in this replay. Raising aggressiveness from four to the native maximum five subtracts two clarity points in the formula, yielding ninety two, while resilience rises five to ninety. At three pixels per clarity point the bars shrink from two hundred eighty two to two hundred seventy six. Recall is eighty plus twice the five-node count, or ninety, and does not depend on aggressiveness. All three formulas clamp to specified bounds. The dialogue consists of four fixed text templates containing the selected topic and slider value. Neither their appearance nor a completed badge proves specialized agents actually debated, learned, or evaluated answers. Adding one node changes the mesh count from five to six. Running again changes the recall formula from ninety to ninety two. At three pixels per authored recall point the bars span two hundred seventy and two hundred seventy six. Simply adding more declared nodes is not evidence of improved memory. The downloadable Markdown blueprint describes configured roles, prompts, links, and learning telemetry. With the external graph library unavailable there are no edge pulses, but the same local fixture computation and export remain available. Treat the artifact as a reusable design sketch, not a tested model pipeline or measured educational result.