🧠 Agent Memory Cost Simulator

The default has five runs and three discovery steps per run. Without memory each run adds three assigned token units, totaling fifteen. With memory the first adds three plus zero point two five and later runs add only zero point two five. Total is three plus five times zero point two five, or four point two five. At twenty pixels per unit the bars measure three hundred and eighty five; savings ten point seven five measures two hundred fifteen. These units are teaching constants, not actual model token counts or measured retrieval costs. For one run at three steps, without memory is three and with memory three point two five, so reuse costs an extra quarter. At two runs the totals are six and three point five, saving two point five. At forty pixels per unit these two-run bars measure two hundred forty, one hundred forty and one hundred. In general savings are steps times runs minus steps minus one quarter times runs. This ignores invalidated memory, wrong notes, storage overhead and changing tasks; it is a simplified cumulative cost model rather than a reliable production savings guarantee. At twenty runs and three steps, without memory reaches sixty and with memory eight, giving fifty two saved units and eighty seven percent after rounding. At five pixels per unit the bars are three hundred, forty and two hundred sixty. Chart.js construction occurs before displayed summary updates, so missing Chart locally leaves stale statistics even though native sliders and reset listeners exist. Export uses html2canvas for a twelve hundred by six hundred thirty PNG, but missing that function throws before its promise fallback. Preserve the original positive canvas buffer and disclose absent chart and export operation; no actual PNG or memory benchmark is claimed.