The saved commerce page simulates transactions in the browser. It selects a worker uniformly at random, rather than optimizing cost or availability despite the adjacent source comment. Token usage is an integer from eight hundred through nineteen ninety nine. Cost equals tokens divided by one thousand, multiplied by the selected worker rate. At an illustrative rate of two cents per thousand tokens, one thousand tokens cost two cents and two thousand tokens would cost four cents. Bars show one hundred sixty and three hundred twenty pixels at eight thousand pixels per dollar. The second example explains the linear formula, although two thousand is outside the actual random generator range. A daily budget input is exported but the transaction branch checks only the maximum unit bid. No real agent purchase, payment, worker request or cryptographic settlement is performed by this local simulator. The transaction first checks worker uptime failure, then timeout, then whether cost exceeds the unit bid. Only the first matching reason is reported. Timeout occurs if actual latency exceeds either the timeout setting or the maximum service latency. With illustrative caps of one thousand and fifteen hundred milliseconds, eight hundred passes the latency test and twelve hundred fails because the lower cap already applies. Bars compare two hundred and three hundred pixels at one quarter pixel per millisecond. Equality with a cap passes because the comparison is strictly greater. Jitter spans about minus twenty to plus twenty percent of the worker latency, followed by integer rounding. A quoted amount for any rejected transaction is added to the displayed refund total, even a rejection before any modeled payment. The lock type and arbitration fee appear in configuration and export but do not change these transaction decisions. A batch resets history and telemetry, then calculates success percentage from successful transactions divided by the requested count, rounded to one decimal place. Settled volume and refund totals are rounded to cents. Batch escrow holdings equal five point two percent of settled volume, also rounded to cents. Ten dollars therefore gives fifty two cents and twenty dollars gives one dollar four cents. Bars show one hundred thirty and two hundred sixty pixels at two hundred fifty pixels per displayed dollar. This is a hard coded illustration, not a balance from actual locked funds. A single transaction step instead sets escrow holdings to four cents regardless of settled volume. The export includes buyer settings, escrow rules, seller pool, aggregate telemetry and only the most recent twenty receipts. The signature scheme is a literal descriptive string, and the transaction hash is generated locally; these fields do not prove signatures or settlement. The explainer describes saved source behavior without claiming real commerce security.