7.1Where the information comes from
The three cases suggest a way of sorting irreversible steps that neither side of the debate, as the notes give it, makes explicit. Some steps face information that will arrive whether or not they are taken: the further movement of a price, for instance. For those, the option-value argument of Arrow and Fisher (1974) and Dixit and Pindyck (1994) fits, and waiting is a real option. Other steps face information that only the step can produce: how a system executes in a market it has never touched. For those, Collingridge's dilemma (Collingridge 1980) applies, and waiting preserves the option without filling it. Hume's Section IV supplies the reason the second class exists. The effects of a new object are known only from experience of it, and experience of a similar object informs only about that object.
The two records fall on either side of this line and answer in a similar way. The simulation record faces information only operation can give, and divides the step. The averaging record faces information the market will give anyway, and still does not simply wait; it sets a plan and a calculation in front of the step. In both, the record's preferred course is neither full commitment nor delay, but a bounded commitment with a stated condition for going further. That is the form, in the terms of Section 2, that a decision takes when it accepts both the irreversibility the option-value literature stresses and the dependence on experience that Hume describes.
7.2Who learns and who pays
A staged decision has a cost that is easy to overlook: the losses of the first stage are borne by someone, and that someone may not be the person who learns from them. The parent study's hypothesis H4 asks whether, the harder a decision is to undo, the further apart are those who decide and those who bear the cost. For the averaging record both coders read the trader who decides as the one who bears the loss, and under the lenient reading the record counts against H4. For the simulation record the coders did not agree on who bears the cost of waiting, and the record is unclear. Neither record names the bearer in its own words. In the simulation record a "separately approved" budget implies someone who approves and someone whose capital is at stake, but the record does not say whether they are the same.
7.3What the reversal figures cannot carry
The parent study's coding of reversibility is the weakest part of its ledger, and the figures should be read with that in mind. The coders agreed on the reversal constraint for 11 of 53 records (21%) and on the reversal action for 13 of 53 (25%). For the two records read here they agreed that undoing the harm would be costly, and disagreed on how hard it would be to lift the restriction.
Read strictly, H4 applies to none of the 53 records. Read leniently, it goes against, with supports 1, against 3, unclear 3 and not applicable 46. These figures cannot carry a claim that the records separate deciders from bearers as undoing gets harder, nor the opposite claim. The lenient verdict rests on very few records, and on cells most of which the coders did not settle. What the figures can carry is narrower: WSTiger's records rarely say in their own words how hard a step is to undo, and the two coders, given the same words, often read the difficulty differently. That is consistent with the case readings above. Both records describe the irreversible step indirectly, through limits on loss and through the calculation of total risk, and neither states in so many words that a fill cannot be recalled.
7.4An alternative reading
A sceptical reader could say that the records need no Hume. Staging a new trading system and capping total risk before adding to a position are ordinary prudence, and Lindblom (1959) already defends deciding by successive limited comparisons as the method that works. On that view, the records are muddling through, and Hume adds a philosophical gloss to common sense.
The objection has force, and the records give it some support. Neither cites any theory. Both describe their stages in the plain terms of practice. But Lindblom's method, as the note records it, compares options that differ little from present policy. The step from simulation to live trading is not a small departure from present policy; it is a change in the kind of object being acted on. Hume's account explains why that difference matters: more of the same experience, which incremental comparison would supply, does not inform about the new object. Kay and King (2020) recommend strategies robust and resilient to what cannot be foreseen rather than optimising on invented numbers. The records' refusal to treat early profit or an improved average as a guide fits that recommendation. Whether it reflects it is not something the records can show.