6.1The claim
Part II of Section IV opens and closes with Hume's account of his own limits. He proposes to be modest in his pretensions, to raise the difficulty himself before others raise it, and he ends by admitting that an argument he cannot find might still exist. This section reads those passages as a model for a particular kind of statement: the statement of what would end a stage. A staged decision is only as good as the rule that says when the first stage is over. Hume's paragraphs show what a statement looks like that admits the limits of its evidence and still commits to an inference. The two records' revisit lines are then read against them. What this section adds to the debate of Section 2 is an account of the point both sides need and neither note describes: the condition on which a decider moves from the small irreversible step to the large one.
6.2To be modest in our pretensions
Paragraph 28 sets out the chain of questions that leads Hume to his problem: what is the foundation of reasoning about matters of fact, and, when the answer is experience, what is the foundation of conclusions from experience. He then says how a philosopher should conduct himself before a difficulty of that kind:
The best expedient to prevent this confusion, is to be modest in our pretensions; and even to discover the difficulty ourselves before it is objected to us. By this means, we may make a kind of merit of our very ignorance.
The advice has a practical form. Before a step is judged a success, state what the evidence for it cannot show. Raise the objection yourself. Hume's phrase "a kind of merit of our very ignorance" describes a record that says plainly what it does not know and is the better for it.
Paragraph 31 adds that the modesty is not a refusal to act on experience:
And though none but a fool or madman will ever pretend to dispute the authority of experience, or to reject that great guide of human life, it may surely be allowed a philosopher to have so much curiosity at least as to examine the principle of human nature, which gives this mighty authority to experience, and makes us draw advantage from that similarity which nature has placed among different objects.
The sentence holds two attitudes together. Experience has authority, and only a fool would reject it. Its authority can still be examined. For a staged decision, this is the stance a revisit line should take. It should commit to acting on experience, since there is nothing else to act on, and it should say which experience, and why that experience is taken to bear on the next step.
6.3Snow that tastes of salt
Paragraph 30 gives Hume's reason why no demonstration can secure the inference from past to future. It does so with two images:
May I not clearly and distinctly conceive that a body, falling from the clouds, and which, in all other respects, resembles snow, has yet the taste of salt or feeling of fire? Is there any more intelligible proposition than to affirm, that all the trees will flourish in December and January, and decay in May and June?
The first image is the snow that is not snow: an object that resembles the familiar one "in all other respects" and differs in the one that matters. The second is a reversal of the seasons. Neither is likely, and Hume does not say they are. He says they are conceivable, and that what is conceivable cannot be ruled out by demonstration. For a revisit line, the snow that tastes of salt is the case the line must anticipate: the live operation that resembles the simulation in every respect but one, or the second entry that resembles the first in every respect but the state of the market. A revisit line that names only the respects in which the stage resembled the plan does not guard against it.
6.4An enumeration not complete
Paragraph 33, which ends the section, turns Hume's modesty on his own argument:
I must confess that a man is guilty of unpardonable arrogance who concludes, because an argument has escaped his own investigation, that therefore it does not really exist.
And a few lines later:
Even though we examine all the sources of our knowledge, and conclude them unfit for such a subject, there may still remain a suspicion, that the enumeration is not complete, or the examination not accurate.
Hume goes on, in the rest of the paragraph, to give reasons why in this particular case the suspicion can be set aside. The general principle stands nonetheless. A list of what has been checked is not proof that nothing else needs checking. For a stage that is to be passed on a list of conditions, this is the residual risk: the conditions may all be met and something not on the list may still be wrong. A revisit line cannot remove that risk. It can say what the list is and whether it is meant to be complete.
6.5Two revisit lines
Synthetic scenes: both lines below are from modern summaries written for the platform.
The record Simulation to limited live states when its judgment would change:
Change Condition: If, over a sufficient number of operations cases, execution slippage and risk fall within the criteria and control devices operate, the next stage is considered. / Next Verification: Whether the stage is passed is evaluated using operations indicators rather than accidental initial profits.
The condition has three parts. A sufficient number of cases must be observed; execution slippage and risk must fall within criteria; and the control devices must work. When all three hold, the next stage is considered, not taken. The line does several things Hume's paragraphs would recognise. It asks for a course of experience, not one instance: "a sufficient number of operations cases". It names what the experience is to show, slippage and risk within criteria, and these are the very respects in which, by the record's own assumption, the simulation differed from the market. It also names the stopping machinery, "control devices", as something whose operation must be observed rather than assumed.
What the line does not do is equally clear. It does not say how many cases are sufficient. It does not state the criteria. It does not say who considers the next stage. The parent study's code, which looks for a number, a threshold or a time in each revisit line, found none in this one. Across all 53 records only 5 of 53 (9%) of revisit conditions contained one. On Hume's terms, the line commits to experience as the guide and leaves the amount of experience to a later judgment. That is modest in one sense and open in another. It does not claim to know in advance how much experience will be enough. It also leaves the end of the stage to whoever reads the indicators, without saying who that is.
The record Split entry or averaging states its own condition:
Condition for change: If a new plan is established that meets independently verified split rules and the total-risk cap, application is considered from subsequent trades onward. / Next confirmation: Display the risk with all orders filled, not with a single additional order.
This line is different in kind. It does not say what experience would change the judgment. It says what document would: a new plan, meeting independently verified rules and the total-risk cap. It also says the change applies only "from subsequent trades onward", which means the present position is not rescued by a plan written after the loss. The line refuses to let a rule made after the event govern the event that prompted it. That is a precise answer to the averaging problem. A plan made after the loss, applied to the position that lost, would be a rationalisation shaped by the outcome. The record forbids it.
The second part of the line names a calculation: the risk is to be shown "with all orders filled, not with a single additional order". This is the record's version of Hume's incomplete enumeration. A trader who looks at the risk of one additional order sees a part. The record asks for the whole, the state in which every planned order has been executed. It does not claim that the whole is then known; it asks that the largest exposure the plan permits be displayed before the plan is followed.
6.6Accidental initial profits
The Simulation to limited live line contains a phrase that ties the staged decision to the literature on judging decisions by their outcomes: the stage is to be evaluated "using operations indicators rather than accidental initial profits". The phrase names, in the record's own words, the pattern Baron and Hershey (1988) describe: that a decision is rated better when its outcome was good. A first live stage that happens to make money invites that rating. Fischhoff (1975) adds that people who know an outcome judge it more likely than they would have without knowing it, and do not notice the shift. A stage that made money would, on this account, look in hindsight as if it had been bound to.
The record's remedy is to name in advance what is to be judged. Execution slippage, risk within criteria and the operation of control devices are features of how the system behaved, not of whether it made money. They are, in Hume's terms, experience of the right object: the market's response to the system's orders. Profits in a first stage are experience too, but of a mixture of the system and the market's movement over that stage, and a short stage cannot separate the two.
This is where the case connects back to the eggs. A profitable first stage resembles a successful system in appearance. Whether it is one is the main question, and an early profit does not answer it. The record says so, briefly, and directs attention to the indicators that bear on the question it is actually asking.
6.7What the case adds
The case adds to the debate of Section 2 an account of the hinge that a staged decision turns on: the condition for moving from the small irreversible step to the larger one. Hume's paragraphs 28 to 33 show a form for such a condition: commit to experience as the guide, state what experience is being relied on and why, and admit that the list of checks may not be complete. The two records' revisit lines take that form in part. One names the right kind of experience, operations indicators rather than early profit, and the respects in which the stage must differ from the simulation. The other refuses to let a plan written after a loss govern the position that lost, and asks for the risk of the whole plan, not one order. Neither names a number of cases, a threshold, a time or a person. That is the gap the option-value argument and Collingridge's remedy both leave at the same point, and the records do not close it.