The invoice fixture assumes twenty five thousand units, twelve manual minutes each and forty five dollars per hour. That is five thousand labor hours and two hundred twenty five thousand dollars per month. Assigned automated unit cost combines one tenth of a cent trigger, zero OCR fee, zero point six two five cents input, four tenths of a cent output, point three two five cents vector lookup and point two cents sync: one point six five cents total. Multiplying by volume gives four hundred twelve dollars fifty cents. At one pixel per thousand dollars manual cost measures two hundred twenty five, automated cost point four one two five and difference two hundred twenty four point five eight seven five. These are assumptions, not measured invoice processing. The human review node assigns a six percent routing rate and five minutes per review. Of twenty five thousand units, fifteen hundred reviews would require seventy five hundred minutes, or one hundred twenty five hours. At forty five dollars an hour that is fifty six hundred twenty five dollars. At one pixel per thirty dollars review cost measures one hundred eighty seven point five, API cost four hundred twelve point five measures thirteen point seven five and combined cost six thousand thirty seven point five measures two hundred one point two five. The implemented automated cost omits review wages entirely. It sums review trigger rates for automation rate but does not subtract their human labor from claimed savings. No financial return is independently verified. Nonhuman assigned latencies total one hundred fifty plus nine hundred plus twenty one hundred plus three hundred plus two hundred fifty milliseconds, giving three point seven seconds. Adding the five minute review node produces three hundred three point seven seconds. At one pixel per second base latency measures three point seven, review addition three hundred and tail three hundred three point seven. The page calls these median and ninety ninth percentile, but computes sums rather than observing a latency distribution. Cytoscape initializes before native event bindings. With the missing saved external engine, inputs, reset and export stay inert. Math.js is optional and unused by the ordinary arithmetic. No actual pipeline, invoice, JSON download or percentile measurement is claimed.