In an investment committee session, the greater part of the discussion tends to circle a single figure in the model — unit cost, occupancy rate, capacity factor, procurement price. Whether that figure sits too high or too low is debated at length, sensitivity cases are run, upper and lower bands are contested; the question rarely put to the room, by contrast, is where the figure originated in the first place. The presenting team defends the number, the committee interrogates the number, yet the interrogation is directed at its value rather than at its lineage. By the time the session closes the band may well have narrowed, but which source produced the figure, by what method, and on what date remains unrecorded — and that silence outlives the decision itself by a considerable margin.
The same pattern becomes sharper when successive versions of a feasibility file are laid side by side. A single line item may have been drawn in the first draft from a supplier's indicative quote, updated in the second from an engineering team's experiential estimate, and carried in the third from a prior project; the three sources differ entirely in reliability, in useful life, and in the direction of their likely bias, yet all three appear in the file in identical form, as an unattributed cell. Six months on, no one remains who recalls the ancestry of the number, while the number continues to sit in the model, growing marginally more solid with each internal repetition.
The mechanism underlying this behavior is designated in the decision literature as **provenance uncertainty** — the condition in which the actual source of an input, the method by which it was produced, and the date of its production cannot be verified — and it is not merely an information deficit. The mechanism operates as follows: the moment a figure enters a document, it inherits that document's institutional standing. An estimate placed into a board presentation is no longer read as an estimate but as data, because the formatting of the document — the table, the footnote, the currency unit, the decimal place — dresses the figure in a claim to precision that its source was never able to carry. This decoupling of the epistemic quality of an input from the presentational quality of its container is where provenance uncertainty is actually manufactured.
The second layer is repetition itself. A number whose origin was never recorded is treated, at each internal citation, as though it had passed a verification step; by the third presentation what stands behind the figure is no longer a supplier quote but the two presentations preceding it, and the chain collapses within a short interval into a closed, self-referential loop. Once that loop closes, questioning the number ceases to be a technical operation and becomes an institutional assertion — whoever raises the question has opened for debate not a single assumption but every intermediate decision that rested upon it, and the cost of doing so ordinarily suppresses the question altogether.
The conditions under which the mechanism is functional deserve naming as well, since this tendency is not an error but a shortcut that lowers cost under a specific set of circumstances. Re-verifying the origin of every figure at every decision point would slow decision velocity to an unacceptable degree; institutions therefore reuse previously validated inputs without reopening them, and under ordinary conditions that preference is entirely rational. The difficulty lies not in the shortcut but in its persistence after the conditions have moved: when procurement prices are repriced, when a regulatory regime shifts, when project scale changes an order of magnitude, or when the person who originally produced the input leaves the organization, the verification credit the figure once carried has already been exhausted — and the figure does not announce this.
The institutional cost is not, as first supposed, the inaccurate estimate itself. An inaccurate estimate whose source is known remains a correctable line item; an accurate estimate whose source is unknown is an uncorrectable structure, since when conditions change there is no way to determine what ought to be updated. Where a cost item is found to have drifted, the greater share of the team's time is spent not measuring the magnitude of the drift but tracing backward the stage and the document through which it entered the model; this exercise in archaeology typically consumes a meaningful portion of a budget cycle and rarely arrives at a complete answer.
The second cost surfaces directly on the valuation plane. In a diligence process, the buy-side adviser will ask for the chain of sources beneath the assumptions supporting the projection; every assumption for which that chain cannot be produced is rebuilt in the acquirer's model from the conservative end, because accepting an unverifiable input at face value is not a defensible position on the buy side. The practical consequence is that what gets priced is not the seller's optimism but the documentation gap — and that pricing appears in one of three places: a discount in the headline figure, an earn-out trigger deferred to the post-closing period, or a broadening of the representations and warranties package accompanied by an increased escrow ratio.
The third cost accumulates along the accountability line. An input whose provenance was never recorded belongs, when it fails, to no one; responsibility disperses across teams, and to the extent it disperses, institutional learning does not occur. The same drift reappears on the next project and is received as a fresh surprise. The most insidious form of founder dependency is manufactured here as well: when the real origin of inputs resides in the memory of a handful of individuals rather than in the documents, the model can be run without those individuals but cannot be defended, and that distinction emerges quickly at a review table.
Structural intervention begins not by raising individual skepticism but by attaching an identity to the input itself. It has four components: first, a source record carrying, for every numerical input, its origin, its production method, and its production date; second, a labeling regime that separates inputs by reliability class — contractually fixed, quotation-based, derived from a comparable project, expert estimate; third, a defined shelf life for each class together with a revalidation step triggered on expiry; fourth, a change trail showing, for every cell altered in the model, who altered it and on what stated ground. Constructed together, these four place the chain behind a number in the file itself rather than in institutional memory.
In the projects BEIREK manages, this architecture is established as a precondition of the model rather than as an annex to it. Every feasibility and financing model opens alongside a cell-level assumption ledger, in which each input is logged with its source, its reliability class, its date, and its responsible line — and the moment of logging is the moment the input first enters the model, not the moment it is approved, an ordering that is decisive precisely because retrospective reconstruction is never complete in practice. Each ledger entry, upon reaching the end of its shelf life, generates an agenda item of its own, and those items are attached to the milestones of the project calendar: pre-FID, pre-closing, pre-first-draw.
The same discipline produces direct leverage in counterparty negotiation. Where a credit committee or an acquirer's adviser can be shown, on a single page, the source of the assumption under discussion, its date, and what has changed since that date, the conversation moves from the plausibility of the assumption to the currency of the data; this narrows the negotiated band appreciably and weakens the conservative repricing reflex. By the same mechanism, when a variance emerges the point of intervention is located within hours, because what is being searched for is a record rather than a recollection — and an error that carries a record becomes, in most cases, an error the institution is able to learn from.
The decision quality of an institution is measured less by the accuracy of the decisions it takes than by the speed with which it can demonstrate where a decision that proved inaccurate first went wrong. That measure depends not on the sophistication of the model but on whether the chain standing behind each figure within it has been recorded; and where that chain is absent, the institution becomes not the owner of its own numbers but merely their carrier. The question worth putting is not whether the assumptions in the model are correct, but which of them could have their source produced tomorrow, should anyone ask.