--- title: "The Model That Already Knows Its Answer: Projections Built Backward From the Threshold" description: "Financial-model fantasy is the practice of building a projection backward from a desired return threshold rather than forward from evidence. When dozens of individually defensible assumptions all lean the same way, the model stops testing anything and becomes a document of approval. The neutralizing mechanism is not individual restraint but an assumption register recording each input’s owner, evidence class, and falsification threshold." url: https://www.beirek.com/en/blog/financial-model-fantasy canonical: https://www.beirek.com/en/blog/financial-model-fantasy published: 2025-12-08 modified: 2025-12-08 category: "Entrepreneurship" category_url: https://www.beirek.com/en/blog/category/entrepreneurship language: en-US reading_time_minutes: 8 publisher: BEIREK LLC publisher_url: https://www.beirek.com license: "© BEIREK LLC — citation with attribution and link permitted" keywords: ["financial model assumptions","investment committee hurdle rate","covenant headroom DSCR","quality of earnings diligence","assumption register","backward-built projections"] topics: ["Financial modeling discipline","Investment committee governance","Project finance covenants","Valuation and diligence risk","Decision recording systems"] alternate_language_url: https://www.beirek.com/tr/blog/financial-model-fantasy --- # The Model That Already Knows Its Answer: Projections Built Backward From the Threshold > **In short:** Financial-model fantasy is the practice of building a projection backward from a desired return threshold rather than forward from evidence. When dozens of individually defensible assumptions all lean the same way, the model stops testing anything and becomes a document of approval. The neutralizing mechanism is not individual restraint but an assumption register recording each input’s owner, evidence class, and falsification threshold. *When a financial projection is constructed backward from the return it is required to produce rather than forward from the evidence available, the failure is architectural rather than ethical. To the extent that the output is calibrated to a committee hurdle, the model ceases to function as a test and becomes an approval document — and the bill arrives in the valuation, in the covenant headroom, and in the first drawdown schedule.* --- In a single investment committee file, the fact that the projected return lands just above the institution’s published hurdle reads as coincidence; placed alongside the files that preceded and followed it within the same institution, the same fact reads differently, because the outputs cluster within a narrow band around the threshold and a file falling materially below it rarely reaches the agenda at all. More diagnostic than the clustering, however, is the revision history. The distance between the first working version of a model and the version that goes to committee is seldom spread across hundreds of cells; it typically concentrates in three or four — commissioning pulled forward by a quarter, the ramp curve steepened modestly, the terminal growth assumption adjusted upward by a fraction of a point, the discount rate softened at the margin. Each of those four corrections withstands examination on its own terms, having been drawn from a range a reasonable analyst would accept. Considered together, what becomes apparent is that all four face the same direction. A second observation, less frequently discussed, concerns the sequence of questions in the room. What the model says is often not the first question asked; the first question is whether the model holds. That ordering, which passes for an innocuous habit of speech, determines the direction of the work that follows, because a team that knows the required answer before assembling the inputs is no longer reading a model but constructing one. The same pattern leaves traces in the file names themselves: as version numbers advance and the scenario labels migrate from base, upside and downside toward approvable, presentable and committee, the model has completed its transition from an object of analysis into an object of negotiation. The pattern has a name — financial-model fantasy — and it describes a projection built backward from the outcome it must reach rather than forward from the evidence it possesses. The mechanism is not one of deception, and the distinction matters considerably, because in a model constructed backward no individual assumption need be invented; every one of them may have been selected from a range a competent reviewer would defend. What produces the drift is not the assumptions but the selection rule: which end of the defensible range gets chosen is governed by what the output requires. Thirty assumptions, each nudged a modest distance toward the favorable end of its interval, compound into a difference in the output that can approach an order of magnitude, while no single line, examined in isolation, is left undefended. Under certain conditions the same tendency is not merely harmless but useful. One legitimate use of a projection is reverse engineering: asking what would have to be true for this investment to make sense is a feasibility test, and it delivers a clarity that early-stage work cannot otherwise obtain. The model also functions as a coordination device, since aligning several functions around a shared set of numbers reduces organizational friction far more effectively than allowing each to operate from its own estimate. The difficulty lies not in the shortcut itself but in the fact that the moment of transition goes unrecorded: when the table showing what would have to be true becomes, without any explicit decision, the table showing what is true, the institution is no longer holding a feasibility test but a commitment. The technical signature of backward construction is the compression of variance. Sensitivity tables tend to be built symmetrically around the base case rather than around the evidence, whereas in practice the downside tail of a revenue ramp and its upside tail are rarely of equal length, delays being cumulative in a way that accelerations are not. A second signature concerns where the weight has been shifted: where a substantial share of value rests on terminal value, on the final years of a maturation curve, or on cohort behavior not yet observed, the model is pricing the period furthest from observation rather than the period closest to it. The third signature is the plug — a line released to balance the sheet once every other line has been fixed, which is to say a line that has stopped being an assumption and become an output. The first surface on which the institutional cost appears is the cash timeline. To the extent that working capital requirement is derived from the slope of the revenue ramp, a slip of one or two quarters does not merely postpone the requirement curve, it raises its peak; and where supplier terms remain fixed while collection cycles lengthen, the gap widens from both directions simultaneously. The same mechanism registers on financed projects through the drawdown schedule, since a construction-period spend curve that fails to advance at the modeled pace disturbs the balance between commitment fees and capitalized interest, producing a budget overrun before any operational difficulty has occurred. The second surface is covenant headroom, where the mechanism operates with particular quietness. Lenders customarily calibrate DSCR or leverage thresholds off the sponsor’s own base case, adding a margin on top; where that base case was assembled backward, the margin left is not genuine headroom but headroom already consumed. The consequence is that the first breach originates in the non-realization of the base case rather than in any market shock, meaning that the threshold designed to test the system engages in the absence of an external event. From that point a renegotiation table opens, and the bargaining position of the party seated at it is measurably weaker for the knowledge, shared by both sides, that the breach traces back to its own forecast. The third surface is valuation, and its effects are the most durable. In a sale or capital raise, a quality-of-earnings review systematically exposes the distance between the modeled ramp and observed cohort behavior, and once that distance is established the counterparty’s response is generally not a direct reduction in price but a migration of risk into structure. In practice this appears as an enlarged earn-out share of total consideration, a higher escrow percentage, broadened representations and warranties, and operational thresholds added to conditions precedent. More consequentially, the developer’s own model thereafter functions as the counterparty’s measuring instrument: in every subsequent round, the gap between what was committed and what materialized returns not as a discussion of price but as a discussion of credibility. This tendency cannot be managed through individual restraint, because its source is not individual optimism but the incentive structure itself; so long as the person constructing the model and the person benefiting from its conclusion are the same, the selection rule will lean in a predictable direction. The intervention that works addresses not the model but the recording discipline surrounding it, and it separates into four components. The first is that every assumption has an owner — not whoever entered the figure, but whoever will account for it when the figure proves wrong. The second is that every assumption carries an evidence class: contracted, observed, benchmarked, or merely assumed. The third is a falsification threshold for each assumption, written before revision is needed rather than after, specifying which observation would trigger a rewrite of that line. The fourth is version discipline: recording which cell changed in which version, and why, at the same moment the change is made. BEIREK establishes this discipline on the projects it manages as an assumption register, maintained separately from the model and outliving it, in which each line stands alongside its owner, its evidence class, its falsification threshold and the date of its last review; where an evidence class is downgraded — a volume commitment believed to be contracted turning out to rest on a letter of intent — that downgrade becomes visible in the register before it becomes visible in the output. A second mechanism separates the negotiation model from the operating model, so that the table presented to a counterparty and the table used internally for resource allocation and cash planning are not the same file, with authority over each cell defined at the outset. A third mechanism keeps the decision record at the moment of proposal rather than the moment of approval, since capturing what the proposing team knew and did not know at that point makes it possible, later, to assess accountability on the quality of reasoning rather than on the outcome alone. This recording discipline earns its value not at the desk where the model is built but in the rhythm at which it is reviewed. Where the register is opened not on a fixed calendar but at the project’s own milestones — signing, closing, first drawdown, commercial operation, first renewal cycle — it becomes visible at each threshold which assumptions have moved up an evidence class and which remain merely assumed. The most valuable output of that rhythm is not a corrected forecast but an institutional acknowledgment of where forecasting remains meaningless, since the maturity of a projection is measured less by how precise it appears than by how clearly it shows where the uncertainty is concentrated. The most economical way to test the reliability of a model is to ask not what its conclusion is but in what order that conclusion emerged: did the threshold follow from the assumptions, or did the assumptions follow from the threshold? In most institutions the answer to that question resides not in the file but in the file’s history, and where no history is kept, the answer has already been given. ## Key Points - When the difference between a model’s first draft and the version that reaches committee is concentrated in three or four cells rather than distributed across hundreds, the projection has likely been assembled backward from its required output. - Backward construction is not fabrication; it is a systematic drift produced when individually defensible assumptions are each selected from the favorable end of their range and therefore all lean in the same direction. - Covenant headroom calibrated on a sponsor’s own base case converts the first breach into an artifact of an unmet forecast rather than a market shock, and it arrives with the borrower’s bargaining position already weakened. - In diligence, a developer’s own model becomes the counterparty’s measuring instrument, shaping the discount, the earn-out proportion, and the escrow percentage more than any independent estimate does. - Tagging each assumption by evidence class — contracted, observed, benchmarked, or merely assumed — makes visible which regions of the model are load-bearing and which are decorative. ## Questions ### How can I tell whether a financial model was built backward from its conclusion? Three indicators are reliable. First, the difference between versions concentrates in three or four cells rather than spreading across hundreds. Second, the output lands just above the committee hurdle, and this repeats from file to file. Third, one line is released to balance the model after all others are fixed, which means it is no longer an assumption but an output. The history of a model is usually more informative than the model itself. ### What separates an optimistic assumption from financial-model fantasy? Optimism places a single assumption at the favorable end of its range and remains manageable on its own. Fantasy is the selection rule itself being determined by the required outcome: dozens of small, individually defensible shifts, all facing the same direction, compound into a substantial distortion in the output. The distinguishing test is not whether individual lines are defensible but whether the direction of the deviations is randomly distributed. ### How do investors detect inflated projections during diligence? They generally examine the evidence outside the model rather than the model itself. A quality-of-earnings review compares the projected ramp against observed cohort behavior, collection cycles and customer concentration, while contract review separates volumes treated as committed from the documents actually supporting them. Where a gap is found, the response is typically not a direct price reduction but a migration of risk into structure: a larger earn-out share, a higher escrow, broader warranty coverage. ### What is an assumption register and how does it differ from the model? An assumption register is a record kept separately from the model and outliving it, carrying four fields per line: the owner of the assumption, its evidence class — contracted, observed, benchmarked or merely assumed — its falsification threshold, and the date of last review. The model produces the number; the register produces what the number rests on. When an evidence class is downgraded, that downgrade appears in the register before it reaches the output, allowing the decision to be reopened in time. --- Source: https://www.beirek.com/en/blog/financial-model-fantasy Publisher: BEIREK LLC — https://www.beirek.com