How to interpret the evidence.

The fictional distributor’s invoices, payments and accounting entries were recorded in CRONUS, the sample company in Microsoft Dynamics 365 Business Central. Bank balances and transactions came through Plaid Sandbox, a simulated bank-connection service, using matching test data. No real company accounts or bank funds are represented.

Accounting entries balance, settlements match invoices, and recorded bank movements explain the closing cash balance. The bank comparison covers the declared statement period; it is not independent bank confirmation because both systems use the same generated transactions.

The current evidence contains four business comparisons and four no-discretion checks: twelve evaluated plans, with eight recorded approvals. The narrower-supplier-terms comparison and its control were corrected and rerun on 24 October with fresh approval. The other cases retain their 20 October results on the same frozen financial facts and planning forecast. This is not a new uninterrupted run of the entire matrix. Every plan passed the financial engine’s checks and a separate verification of its saved calculation. “Separate verification” means an additional software check, not review by an independent audit firm.

Approvals bind the exact planning inputs, permitted choices and engine version. The test also checked rejection before approval. The separate, subsequent liquidity-review test confirmed that a shortfall requires human review before rerun or reuse, and that edits invalidate the previous planning approval. Neither control authorizes a financial transaction.

The comparison policy pays on due dates or valid discount deadlines, then chooses the cheapest permitted funding. AGX may additionally move payments within the approved supplier terms. Both have the same information, obligations, financing access and $10 million weekly cash minimum.

• Borrowing: one bridge loan, in $5 million increments up to $30 million; a permitted draw in weeks 1–8 and full repayment in week 13. These increments can affect the amount needed.
• Interest: 0.12% per active week, including the draw and repayment weeks. The delayed-collection calendar borrows $10 million for twelve charged weeks: $144,000 interest. This is an assumed fixed rate, not a lender quote.
• Payments: each original invoice is paid in full in one permitted week. Discounts depend on the specified payment deadline. Contractual flexibility and discount prevalence are synthetic assumptions.
• Objective: minimize interest less supplier discounts while meeting the modeled constraints. Cash headroom and peak borrowing are reported separately; greater headroom is not the optimization objective.

The model does not cover intraday cash, partial supplier payments, late fees, facility fees, financial covenants, revolving borrowing limits linked to assets, foreign exchange or payment execution. A company evaluation must first establish whether those features matter and whether the supported model fits its decision.

The thirteen additional weeks check that the selected plan’s original invoices and loan are settled rather than deferred beyond the reporting period. Ongoing business obligations remain disclosed. The checks do not establish that the company has no further liabilities.

The forecast uses recorded customer payment history, including partial receipts, with shared history where customer evidence is insufficient. Known and planned receipts and outflows retain their stated sources. Expected collections estimate receipt timing; the planning forecast is a more conservative view used for the payment decisions.

Two different tests are reported. They must not be treated as interchangeable measures of accuracy:

• Historical development checks: across thirteen overlapping periods from the synthetic history, the latest planning forecast had a $707,882.91 average absolute weekly receipt error. Its largest cumulative cash overstatement was $3,019,575.97. These periods were reused while improving the model; they are not an independent test.
• New synthetic collection variations: after fixing the model, 1,000 receipt paths were tested. No real-company outcomes were available. Other future receipts and operating outflows were held fixed.

Average absolute error measures the size of weekly misses regardless of direction. Cash overstatement measures how far cumulative forecast receipts run ahead of simulated receipts at a weekly cutoff. It is a timing exposure, not an assumed permanent credit loss.

The planning forecast had lower average error across the test, but its weekly error was worse in 122 of the 1,000 paths. Those outcomes remain in the evidence. Neither the observed frequencies nor the historical timing ranges establish real-world probabilities or confidence levels.

For the normal-case schedules only, the operating calendar breached the cash minimum in 24 paths; the selected AGX schedule breached it in none. A further check of combinations of the collection delays present in the synthetic history found a maximum $3,778.57 shortfall for the calendar and minimum $13,097,004.78 headroom for the selected AGX plan. These figures are conditional on that history and unchanged other cash flows; unseen delays, defaults or additional outflows remain outside the check.

The cash-risk check is separate from verification that a plan satisfies its approved forecast. It does not make the forecast certain or expand what the financial engine has proved. Inspect the historical checks and the subsequent collection-risk results.

The financial evidence package contains case inputs, allowed choices, payment and funding selections, approval and engine identifiers, results, follow-through cash balances, source assumptions and forecast checks. It contains all current cases, exact-zero no-discretion checks and unfavorable forecast outcomes. The corrected narrower-terms case explicitly checks all fifteen fixed invoice dates in the submitted requests and verified schedules. The earlier unqualified submission is retained in a historical folder rather than overwritten.

All current comparisons use planning forecast version 7. Under narrower supplier terms, both policies capture $80,872.50 of discounts, rather than the $113,221.50 available in the other cases. Coordinated timing avoids a $5 million bridge draw and $72,000 interest in this corrected case. Its 26-week accounting follow-through also passed. The separate forecast-risk campaign distinguished expected collections from planning and added the cash-risk review. Retained checks confirm that the planning amounts and normal-case financial result remained unchanged. The later review-control test applies to that normal case; it is not represented as a new run of every commercial scenario.

The download is a financial evidence extract. Proprietary algorithms, executable code, development-service details, personal account identifiers and unrelated operational records are excluded. Its accompanying notes identify retained records and any removed nonfinancial fields; the financial amounts and outcomes are not rewritten.

The file-check list identifies this publication’s exact files. Those checks detect file changes; they do not themselves prove the finances correct. The engine is not distributed, so the package does not independently run AGX or reproduce its optimization. It supports inspection of the stated financial inputs, choices, arithmetic and recorded verification.

The evidence establishes a synthetic planning demonstration and local application-control tests. Independent company data, contractual review and production testing would be needed before making claims about a customer’s accuracy, savings or operating reliability.