Requisition to receipt

A receipt that never posts is the single largest cause of an invoice that cannot be matched, so the cost of a broken requisition chain is paid twice: once here in chasing, and again in the payment tower as an exception.

An outcome is assembled, not recorded. No source system in this estate stamps a single reference across a purchase requisition, the order raised against it and the receipt that closes it, so the chains on these pages were stitched together after the fact. Where both documents sit in the same source system the chain is keyed and the join is exact. Where it crosses a system boundary the chain is matched on the party, the market, the region and whether the timing is plausible, and it is labeled inferred with its confidence shown. Where the two sides name no party in common the only evidence left is place and timing, and the chain is labeled cohort and confidence is capped low on purpose. A cohort chain is a hypothesis about which documents belong together. Read it as one.

The ledger behind these chains is one week, 10 to 16 August 2026, which is 168 hours. A chain that had not finished when the week ended is open, not late: the week simply stopped watching it. A chain whose commitment falls after the end of the week could never have been seen to miss, so its service level reads not measurable rather than holding. Nothing on these pages should be read as a monthly or annual attainment.

Cost to serve is modeled from the touches the ledger recorded, not billed from a timesheet. Every touch is priced at 11 minutes of a person at a loaded 58 dollars an hour, and every stage carries 1.10 dollars of system cost and 0.42 dollars of inference. Handoff cost is priced separately: waiting is held at 0.90 dollars an item an hour, rework at 34 minutes, and duplicated data entry at the re-keying minutes recorded against that handoff. Change any of those six rates and every dollar on these pages moves.