In an investment committee session, an asymmetry is built into the procedure itself, sitting between the act of rejecting a proposal and the act of approving one. An approved project acquires a life that can be observed: its budget is tracked, its variance reported, its delay converted into an agenda item, and when it eventually fails, the failure has an owner, a date and a quantified cost. A rejected project, by contrast, closes with a single line in the minutes and is never reopened; whether it was in fact a sound project is never established, because no reality in which it could have been measured was ever brought into being. The same asymmetry operates, without modification, in a hiring panel, in a supplier prequalification, in the decision to discontinue a product line, and in the assessment of a market entry.
The practical consequence is that institutions are able to see only one half of their own error portfolio. A review of an organisation's decisions over the preceding five years can produce a list of the transactions that were wrongly approved within a matter of hours, whereas no amount of time will produce a list of the sound proposals that were wrongly declined, since the counterfactual data required for the second list was never generated. What an institution believes about the quality of its own judgement therefore rests on a sample that is systematically incomplete, and the incompleteness runs in one direction rather than being randomly distributed. Over time, every control introduced to reduce the errors that can be observed exerts quiet pressure in favour of the errors that cannot.
The statistical name for this pattern is **type-II error** — the failure to detect an effect that genuinely exists, under the evidence threshold being applied, with the consequence that it is treated as absent. Its counterpart, type-I error, is the treatment of a non-existent effect as real; translated into institutional language, the first corresponds to the opportunity forgone and the second to the investment written off. The two are not independent: as the evidence threshold rises, the probability of a false positive declines, while the same movement raises the probability of a false negative in direct proportion. Making an approval process "more disciplined" is, in most cases, a shift in the balance between these two errors, effected without the trade-off ever having been discussed.
For the mechanism to operate, no one needs to be inattentive; three structural conditions are sufficient on their own. The first is a measurement window shorter than the interval over which the effect emerges — an intervention with an eighteen-month return horizon will appear inert when examined on a two-quarter review rhythm. The second is a shortage of observations, a small pilot lacking, from the outset, the resolution required to distinguish a genuine effect from coincidence. The third is a signal submerged in noise, since in a period during which price, seasonality, currency and demand are moving simultaneously, the contribution of a single intervention cannot be isolated. Under any of these three conditions, the statement "no effect was found" does not mean that no effect exists; it reports only that the apparatus assembled was never capable of seeing one.
Recognising the conditions under which this tendency is functional matters, since otherwise the intervention is built at the wrong point. Where capital is scarce, managerial attention limited and reversal difficult, a high evidence threshold is entirely rational; an organisation that acts on weak signals distributes its resources across a large number of half-finished initiatives and carries none of them to the scale at which they would matter. The difficulty lies not in the height of the threshold but in the threshold being held **identical across every category of decision**. When a reversible, low-cost decision that generates learning is subjected to the same evidentiary standard as an irreversible, capital-intensive and one-time commitment, the type-II error accumulating in the first category will typically cost far more than the protection purchased in the second.
The first surface on which the institutional cost becomes visible is the age distribution of the product and process portfolio. An organisation operating under a high-threshold approval architecture will, within a few years, find its revenue base resting increasingly on older lines and more mature customer cohorts; the shift registers as a break in no single quarter, because the existing lines continue to perform. Its counterpart in the accounts appears not in the absolute level of the research and development line, but in the number of distinct initiatives across which that line is distributed. The same pattern surfaces commercially as a high win rate on a low volume of bids submitted, and a high win rate is, more frequently than is acknowledged, evidence of excessive selectivity rather than good selection.
The second surface emerges in a sale or transfer process. Among the questions that determine valuation at the diligence table is whether the growth narrative rests on the natural expansion of existing lines or on a demonstrated capacity to originate new ones. In an organisation that has long operated at a high evidence threshold, the decision record yields no evidence bearing on growth optionality, and the buyer will typically price the absence through an adjustment to the terminal growth assumption or by attaching a portion of the earn-out to the performance of newly established lines. Founder-led companies carry an additional layer, in that the threshold is not a documented procedure but the founder's personal standard of persuasion, and because it cannot be evidenced in writing, it feeds the founder-dependence discount directly.
The third surface becomes apparent in the workforce. Rejections closed without stated reasoning produce a predictable behavioural adjustment among those who originate proposals; after the second or third such closure, the volume of proposals declines and those that continue to arrive have already been pruned in anticipation of the threshold. From that point the organisation no longer receives a flow capable of testing its own evidentiary standard, and the height of that standard becomes unmeasurable, since the material that would press against it never enters the system at all. The portion of staff turnover attributable to this mechanism is rarely reported accurately in exit conversations, given that the departing individual will also tend to attribute the decision to a general atmosphere rather than to any single ruling.
Structural intervention begins not with a request that decision-makers exercise greater care, but with making the record of the decision symmetrical. It has four components: first, a **rejection record**, in which each declined proposal is logged together with the grounds for declining it and the specific evidence that would reverse the decision; second, a **threshold statement**, setting out which evidentiary standard was applied and why that standard was selected; third, a **reversibility classification**, dividing decisions into two threshold groups according to how far they can be undone rather than how much they cost; and fourth, a **reopening rhythm**, under which rejection records are revisited at fixed intervals against the single question of whether the reversing evidence has since materialised. Taken together, these four components generate the only observable trace this error will ever leave.
BEIREK's intervention in capital-intensive and financed projects is constructed around precisely this record architecture. On assuming a development or investment pipeline, one of the first tasks is to set out in writing the threshold applied at each gate decision — pre-FID screening, supplier prequalification, technology selection, site elimination — and to define, for every option eliminated, the condition that would reverse the elimination; that condition is then attached to a date and carried as a distinct item in the project review rhythm. In stakeholder pre-mortem work, the scenario in which the eliminated option proves to have been the correct one is constructed as explicitly as the scenarios in which the project fails, the purpose being not to alter the decision but to make visible the assumption on which it rests.
The second line of intervention is the separation of the evidence threshold by decision type. For reversible decisions — pilot-scale deployments, time-limited supplier trials, technical validation on a single site — the threshold is deliberately lowered to increase the rate of learning, while for irreversible commitments it is raised and the evidence chain deepened, the two operating as consciously distinct regimes within a single governance framework. Where the distinction is not drawn, one uniform threshold delivers both slow learning and inadequate protection at the same time; where it is drawn, the organisation's error portfolio ceases to run in a single direction and both classes of error become capable of being priced.
The quality of an organisation's judgement is measured not by the accuracy of the approvals it grants but by its capacity to hold both classes of error in view simultaneously. In a system that tracks only what was approved, every backward-looking assessment will necessarily flatter the system, and that flattery generates the argument for raising the threshold further still. The question worth putting is not which decisions turned out to have been wrong last year, but which proposal declined last year would be most valuable to hold today; a decision record incapable of answering it is documenting the institution's own blind spot.