4.1The claim
The opening paragraphs of Hume's Section IV, before the problem of induction in Part II, make a plain claim: nothing about a new object, examined however carefully, reveals what it will do; its effects are known only after it has acted. For an object that can be tried safely, this is a remark about the order of knowledge. For an object whose first trial cannot be undone, it is a remark about cost: the knowledge arrives with the step and not before it. This section reads paragraphs 23 to 25 and a part of paragraph 29 with that second sense in view, and then brings in the record Simulation to limited live. What it adds to the debate of Section 2 is a reason, drawn from Hume, why the information the option-value argument waits for may not arrive while the step is withheld.
4.2An object entirely new
Hume begins his account of cause and effect by asking how knowledge of that relation is obtained. His answer is that it is never obtained by reasoning from the appearance of the cause. In paragraph 23 he states the point as a general proposition and illustrates it with the first man:
Let an object be presented to a man of ever so strong natural reason and abilities; if that object be entirely new to him, he will not be able, by the most accurate examination of its sensible qualities, to discover any of its causes or effects. Adam, though his rational faculties be supposed, at the very first, entirely perfect, could not have inferred from the fluidity and transparency of water that it would suffocate him, or from the light and warmth of fire that it would consume him. No object ever discovers, by the qualities which appear to the senses, either the causes which produced it, or the effects which will arise from it; nor can our reason, unassisted by experience, ever draw any inference concerning real existence and matter of fact.
The examples Hume chose are worth noticing. Water that suffocates and fire that consumes are not neutral illustrations. Each is an effect that the first trial would make known, and each is an effect that the person making the trial might not survive to use. Hume's argument does not depend on that feature, and he does not draw attention to it. The point he makes is about the source of knowledge: Adam, with a perfect intellect, could not have known from the look of water what it would do to him. But the choice of examples carries a second point with it. If the only way to learn what water does is experience of water, then for some objects the lesson and the harm arrive together.
That second point is the one that matters for an irreversible step. The option-value argument, as Section 2 set it out, values waiting because information will arrive later. Hume's paragraph says that information about the effects of a new object comes from experience of that object. If the object is held back, the experience does not occur, and the information does not arrive. Waiting, in that case, preserves the option but buys no knowledge with it.
4.3Marble, gunpowder and intricate machinery
Hume anticipates the objection that some effects seem obvious in advance. He grants that the dependence of knowledge on experience is easy to admit for objects that were once strange to us, and offers examples:
Present two smooth pieces of marble to a man who has no tincture of natural philosophy; he will never discover that they will adhere together in such a manner as to require great force to separate them in a direct line, while they make so small a resistance to a lateral pressure.
The marble example concerns a behaviour that depends on the direction of the force. The pieces resist being pulled apart in one direction and slide easily in another. Nothing in their look suggests either behaviour, let alone the difference between them. Hume continues with a class of cases closer to the record this section reads:
In like manner, when an effect is supposed to depend upon an intricate machinery or secret structure of parts, we make no difficulty in attributing all our knowledge of it to experience.
A trading system routed into a live market is an intricate machinery in Hume's sense, and so is the market it routes into. The order goes through a broker, an exchange, a matching engine and a settlement process, and meets other orders whose senders are not known. Nobody would claim to know from the design of such a system alone how it will behave. Hume's sentence says that in such cases we readily admit that all our knowledge comes from experience. The difficulty, for a decision, is which experience counts.
4.4Not waiting for the event
Hume then turns to the cases where the dependence on experience is harder to see: familiar events, such as one billiard ball striking another, whose effects we feel we could have foreseen from the start.
We fancy, that were we brought on a sudden into this world, we could at first have inferred that one Billiard-ball would communicate motion to another upon impulse; and that we needed not to have waited for the event, in order to pronounce with certainty concerning it.
The phrase "we needed not to have waited for the event" names exactly the illusion that would make waiting unnecessary. If the effect could be inferred without experience, there would be no need to wait for the event, and no need to take the step in order to learn. Hume's claim is that the feeling of foresight is itself a product of past experience, so familiar that it hides its origin. In paragraph 25 he takes the billiard balls apart:
When I see, for instance, a Billiard-ball moving in a straight line towards another; even suppose motion in the second ball should by accident be suggested to me, as the result of their contact or impulse; may I not conceive, that a hundred different events might as well follow from that cause? May not both these balls remain at absolute rest? May not the first ball return in a straight line, or leap off from the second in any line or direction? All these suppositions are consistent and conceivable.
And he concludes the paragraph:
In vain, therefore, should we pretend to determine any single event, or infer any cause or effect, without the assistance of observation and experience.
Two phrases bear on a staged decision. The phrase "a hundred different events" says that to reasoning alone many outcomes of a cause are equally conceivable, and only observation narrows them; a decider who has not yet seen a system's live behaviour faces such a range. The phrase "any single event" says that not only general laws but each single event needs experience. A first live order is a single event, and so is the first stop command sent to a real broker.
4.5Objects only in appearance similar
The record this section reads turns on a substitute for the event: simulation. A simulation is an object built to resemble the market. If it resembled the market in every respect that matters, experience of it would be experience of the market, and the step from simulation to live trading would carry no new information. Hume's Part II addresses exactly this kind of extension. In paragraph 29, having said that past experience gives direct and certain information, he defines its limit:
information of those precise objects only, and that precise period of time, which fell under its cognizance: but why this experience should be extended to future times, and to other objects, which for aught we know, may be only in appearance similar; this is the main question on which I would insist.
The phrase "other objects, which for aught we know, may be only in appearance similar" describes the relation between a simulation and a market with uncomfortable accuracy. The simulation is a different object. It is built to look like the market, and in the respects it was built to reproduce it does. Experience of it informs, in Hume's words, about "those precise objects only". Whether that experience extends to the market is "the main question". Hume's later paragraphs argue that no reasoning can settle that question in general, and the companion article has read those paragraphs at length. What matters here is the narrower point of paragraph 29: experience of one object informs about that object, and its extension to another, similar in appearance, is an inference that experience of the first cannot itself justify.
For the option-value argument this has a direct consequence. A decider who waits and runs the simulation longer gains more experience of the simulation. Keynes (1921) separates the weight of evidence from probability. Applying that distinction here is this article's step: a longer simulation adds weight, but on Hume's account the added weight is evidence about the simulation, not about the market, which is the other object. Waiting improves knowledge of the substitute and leaves the main question where it was.
4.6The record: from simulation to limited live trading
Synthetic scene: a modern summary written for the platform; the system, desk and capital are invented.
The record Simulation to limited live sets out the problem in its Scenario field. The first part is an assumption:
Assumption: The simulated environment does not fully reproduce actual execution waiting, order rejection, and cost and margin constraints.
The sentence lists three things the simulation does not reproduce: the time an order waits to be filled, the chance it is refused, and the costs and collateral constraints of real trading. Each is an effect that depends, in Hume's phrase, on "an intricate machinery or secret structure of parts". Each is also a respect in which the simulation is, in Hume's phrase, "only in appearance similar" to the market. The record does not say that the simulation is useless. It says that it is incomplete in named respects. That is a more exact statement than a general warning about models, and it names the dimensions on which the step to live trading will produce new information.
The question the record poses is this:
Is it acceptable to proceed directly to operations with the full target capital if simulated trading is stable?
The question contains the inference Hume examines. The clause "if simulated trading is stable" is the experience. The words "proceed directly to operations with the full target capital" are the step taken as if that experience extended to the market. The record's question asks whether the extension is acceptable. Its three options are:
Full Investment of Target Capital
Restricted Risk Budget Stage
Expansion After Reviewing Operations Indicators
None of the three options is to keep simulating. The record does not offer waiting as the option-value literature frames it, that is, holding the step back while information arrives. Its three options are all forms of acting: in full, in a restricted stage, or in an expansion conditioned on indicators that can only come from operation. The record's example judgment reads:
Strategy hypothesis verification and actual execution verification are separated into stages. After confirming the execution, reconciliation, and stoppage procedures within a separately approved restricted risk budget, whether to expand is judged.
The first sentence separates two things that the simulation had run together. One is whether the strategy's hypothesis holds, which a simulation can test after a fashion. The other is whether the system executes as intended in a real market, which a simulation cannot test because, by the record's own assumption, it does not reproduce execution waiting, rejection and real costs. The second sentence says how the second kind of knowledge is to be obtained: by operating within a limited, separately approved budget, and confirming execution, reconciliation and the procedure for stopping. Only then is expansion judged.
The record's Scenario field also lists the evidence the judgment requires:
Required Evidence: Differences between simulated and actual environments · Stage-by-stage loss and size limits · Conditions for expansion, stoppage, and return
The third item, "Conditions for expansion, stoppage, and return", is the closest the record comes to naming corrigibility in Collingridge's sense. A stage that has conditions for stopping and returning is a stage designed to be watched and corrected. The word "return" suggests that the move to live trading is not treated as a one-way door: there is a way back if the conditions are not met, though the record does not say to what.
4.7Stages as purchased experience
Read beside Hume, the record does something the option-value argument does not describe. It does not wait for information. It takes a limited irreversible step in order to obtain information that waiting could not provide. The losses within the restricted budget, if any, cannot be undone; the record's own phrase "Stage-by-stage loss and size limits" assumes that there will be losses to limit. The stage is a purchase of experience about the market, paid in exposure that will not come back.
This is close to Collingridge's remedy and further from the option-value advice. The decision is kept corrigible not by holding it back but by making the first step small, watched, and reversible in its scale if not in its losses. The dilemma of control is not escaped. The effects of the system become known only in use, as Collingridge's first horn says. But the system is not yet entrenched when they become known, because only a restricted budget has been committed.
The record also stays within Hume's limits in a way that is easy to miss. It does not claim that the restricted stage will prove the system safe at full scale. The example judgment says only that after the stage, "whether to expand is judged". The restricted stage is itself a different object from the full operation, at the least in the size of the capital it puts to work. Experience of it informs about the restricted stage. Its extension to full scale is, again, an inference. The record does not say how that inference is to be made, only that it is to be made after the stage and not before.
Both of the parent study's coders read the harm from live trading as costly to undo; they did not agree on how hard it would be to lift the restriction. The disagreement is informative. Whether a restriction is easy to lift depends on who holds the power to lift it, and the record says only that the budget is "separately approved".
4.8What the case adds
The case adds to the debate of Section 2 a reason, drawn from Hume's Section IV, Part I, why the information the option-value argument waits for may not arrive while the step is withheld. For a new object, Hume holds, effects are learned only by experience of that object, and experience of a similar object informs only about itself. The record Simulation to limited live writes that reason into its own scenario: the simulation does not reproduce execution waiting, rejection or real costs, so more simulation cannot answer the question that matters. Its answer is not to wait but to divide the step, buying experience with a restricted, approved and stoppable exposure. That places the record on Collingridge's side of the debate rather than the option-value side, while keeping the option-value concern with irreversibility. What the record does not say, and what this case cannot supply, is who decides that the stage has been passed, by when, and on what number.