An evidence architecture for cash flows

A cash-flow forecast is an argument about how a service becomes money and how that money survives competing obligations. The evidence architecture should follow that argument. Begin with the physical activity, follow the right to invoice, establish the time and conditions of collection, then identify the expenditure required to keep the right alive. A headline revenue assumption conceals these transitions. A register that contains only a number, a source and an owner can look complete while omitting the very dependency that determines whether the number becomes usable cash.

The first analytical distinction is between a recorded event and an expectation. An invoice issued is an event. Its payment date may still be an expectation. A delivered unit can be verified without establishing that the customer accepted its quality or that the tariff applies to that delivery. The model therefore needs linked records rather than a universal category called actuals. For each material input, an analyst should state what has happened, what remains conditional and which cash-flow line would change if the condition were not satisfied. This exposes uncertainty without dismissing useful operational evidence.

Source quality is also specific to the question. A signed contract can establish a stated payment mechanism while offering little evidence of a counterparty's future liquidity. A technical design can establish planned capacity while offering little evidence of achievable utilisation. A demand study can identify prospective users while failing to demonstrate their willingness to commit. Treating these materials as substitutes allows apparently strong evidence to answer the wrong question. The acceptance register should name the proposition being supported before it names the document. Review then concerns relevance as well as authenticity.

Consider a conceptual service asset for which the sponsor presents an annual sales forecast. The forecast could rest on an accessible customer population, a share expected to use the service, an available operating window and an achievable selling price. Each assumption has a different failure mechanism. A capacity shortfall limits possible deliveries; weak customer uptake limits desired deliveries; a collection dispute delays receipts after delivery. Combining these mechanisms into a single cautious revenue estimate destroys the information needed to choose a remedy. The register should preserve them even if the public model uses a deliberately simpler cash schedule.

The distinction between sourced, assumed and derived information is not a hierarchy of professional respect. Assumptions are unavoidable in forward decisions. Their weakness lies in being disguised, unsupported by a range or unaffected by contradictory evidence. A well-labelled assumption with an explicit failure consequence can be more decision useful than a sourced historical average applied outside its domain. Derived values should retain their inputs and calculation rule. A reader should be able to move backwards from an output to the originating observations and the judgment connecting those observations to the future.

The perimeter of an observation matters as much as its precision. A recorded price may exclude taxes, rebates, transport or quality adjustments that affect the asset's receipts. A historical volume may refer to a wider region, a different service level or an unusually favourable period. Neither observation needs to be discarded automatically. Instead, the analyst should state the transformation required for use in this project. Where that transformation is judgment rather than measurement, the register should show it as an assumption. Quoting additional decimal places does not repair a perimeter mismatch.

Dependencies deserve an explicit place in the evidence record. The selling price may depend on availability, certification or a delivery point. The projected volume may depend on a connecting facility completing on time. The cost estimate may assume that a particular source of labour, power or materials remains accessible. These dependencies can invalidate an input without changing the document that originally supported it. The record should therefore include a trigger for reopening the input. Evidence is not frozen forever merely because a committee once accepted it; its conditions can cease to hold.

Conflicting evidence should be retained in a disagreement register rather than averaged into an apparently neutral answer. Suppose the operator's capacity assessment and an independent engineering review disagree about sustained performance. Taking a midpoint can imply a level of confidence that neither party supports. The useful questions concern the operating conditions each assessment assumes, the test that could distinguish them and the financing exposure before that test can occur. A committee can condition approval on resolving the disagreement or price a bounded uncertainty. It cannot legitimately declare the disagreement resolved by arithmetic alone.

Negative evidence also requires interpretation. The absence of customer commitments may reflect an immature commercial process, a poor service proposition or buyers unwilling to take a long obligation. Those explanations produce different decisions. The sponsor should not be penalised merely because a project is early, but an early project should not inherit the evidence status of a mature contracted asset. The funding commitment can be staged around observable milestones. What matters is that the timing of capital exposure follows the timing of evidence, instead of assuming that later diligence will justify an irrevocable decision already taken.

The model should identify its most consequential evidence gaps by decision effect, not by the number of empty cells. A missing minor cost item may be immaterial to the investment choice; a missing collection lag can be central to liquidity. Sensitivity helps rank these gaps, provided the analyst does not treat a tested range as a measured distribution. An evidence gap becomes a stop condition when plausible values within the unresolved range change the recommendation or leave a dated obligation unfunded. This gives diligence a purpose: resolve the propositions that determine the decision.

An input owner should be accountable for meaning as well as delivery. Naming the commercial team as owner of revenue does not explain whether it attests to market access, tariff terms, customer commitments or collections. A more useful record separates preparer, source custodian and acceptance authority. The preparer assembles evidence; the custodian maintains the underlying record; the acceptance authority decides that the input is sufficiently relevant for the present decision. These roles can overlap in a small organisation, but the overlap should be explicit. Responsibility should not disappear into a departmental label.

Version control is a substantive financial control when decisions depend on changing inputs. The committee package should identify which schedule, contract interpretation and model edition it considered. If a later update changes the ramp, operating perimeter or terminal assumption, the decision owner should receive a bridge explaining the change in economic meaning. Merely replacing the old workbook with a newer file prevents learning and can conceal scope drift. A retained decision record also allows monitoring teams to distinguish a forecast error from a deliberate change in the project they were asked to finance.

Evidence collection should protect the project's ability to act. A diligence process can become so comprehensive that it asks for information whose cost and delay exceed its value. The solution is to sequence questions around reversible decisions. A modest preliminary commitment may require enough evidence to justify the next investigation, while a binding financing commitment requires evidence about obligations that become difficult to unwind. The register should state the decision level for which each input is accepted. This avoids extending a preliminary finding into a stronger approval through repetition of the same source label.

The public paper and the transaction record have different disclosure purposes. A public analytical illustration should make its assumptions transparent while excluding confidential counterparty terms, personal information and private commitments. A transaction team may need those records for actual diligence, under appropriate controls. Public omission should not be confused with missing private diligence; equally, a statement that confidential evidence exists cannot be used as public verification. This paper's worked schedules are illustrative inputs and reproducible calculations. They demonstrate analytical mechanisms, not the performance or financing terms of an undisclosed client.

The acceptance decision should close with an explicit consequence. An input can be accepted for a stated use, accepted subject to a sensitivity boundary, held pending a discriminating test or rejected as irrelevant to the proposition. The reason should identify what the next analyst would need to change to obtain a different decision. A generic instruction to obtain more evidence is weak because it can never be satisfied conclusively. A precise instruction to validate the connection between an operating milestone and the first collectible invoice creates an actionable diligence task and a traceable financing condition.