On Time and Uncertainty

§ 1Two modes of one fact

The methodology essay ended on a boundary: the a priori core certifies what any actor must do, and the applied layer — data-fed, testable — forecasts what these actors will tend to do. Time and uncertainty are where that boundary is drawn most sharply, because they are the two features of action that make the future at once reachable by deduction and unknowable in its content.

Mises makes the connection categorial, not incidental. Action is a reaching-forward: it intervenes in a present for the sake of a state attained later. And precisely because that later state is not yet given, its outcome is uncertain. The two are not separate discoveries about action; they are one structure seen twice:

"That man acts and that the future is uncertain are by no means two independent matters. They are only two different modes of establishing one thing."
— Human Action, ch. VI §1

"The uncertainty of the future is already implied in the very notion of action."
— Human Action, ch. VI §1

The task of this essay is to make "two different modes of establishing one thing" executable — to show that time and uncertainty are carried by one kernel, and that the very fact they need no separate apparatus is the evidence that the kernel has the right shape. The move that does the work is a single identification, stated here and cashed out through the rest:

Uncertainty is defeasibility. The certainty that the future is uncertain is a strict theorem — a priori, and it cannot be defeated. Any particular outcome is only a defeasible conjecture — the actual event can retract it. Profit and loss are that residual. The two coexist without contradiction because they sit on different tiers of the grounding ladder, and that tier asymmetry is exactly "two modes of one thing" made mechanical.

§ 2Time as a praxeological category

Praxeological time is not the physicist's t. It is not measured duration, not a real number, not a coordinate. It is the ordinal order of before-and-after in which action is necessarily embedded — the bare fact that acting means intervening now for a state that comes after. Mises is emphatic that this order is directional and that its three regions are not symmetric:

"Action is always directed toward the future; it is essentially and necessarily always a planning and acting for a better future." — ch. V §2

Between the past and the future there is "a real extended present. Action is as such in the real present because it utilizes the instant…" — ch. V §2

Action "can influence only the future, never the present that with every infinitesimal fraction of a second sinks down into the past." — ch. V §2

Two consequences fall out immediately, and both are already expressible.

Action is in time, and aims at a later state — as theorems. Take the action axiom at the strongest tier; the temporal structure is derived, not posited, so the engine tags each conclusion a theorem with provenance back to the axiom (the deduction the methodology essay's §7 makes inspectable via why).

axiom acts_purposefully("actor")
acts_in_time(H)        <= acts_purposefully(H)
aims_at_later_state(H) <= acts_purposefully(H)
grounding("acts_in_time", "actor")          # -> "theorem"
grounding("aims_at_later_state", "actor")   # -> "theorem"

The order is ordinal — ranked, never measured. Earlier precedes later; that is a comparison, and comparison is licensed. But later - now is a category error: a moment is a position, not a magnitude, and subtracting one position from another asks a question praxeological time has no answer to. Axioma's :: Ordinal discipline enforces exactly this line as a static rule — comparison and selection pass, arithmetic is flagged under --typecheck — while remaining runtime-neutral (an ordinal rides an ordinary integer, so scripts run identically with or without the checker):

now   :: Ordinal: 1
later :: Ordinal: 2
now < later            # licensed — earlier precedes later
max([now, later])      # licensed — selection over the order
later - now            # --typecheck error: "arithmetic '-' on ordinal operand
                       #   'later' — ordinals are position labels, not magnitudes"

The past, in this order, is the region action cannot touch. A bygone is sunk — a datum, the lowest tier: given, not derived, and not actionable. "Bygones are bygones" is not a proverb bolted onto the theory; it is what it means for a moment to have sunk into the past, and the ladder records it as the flat, ungrounded given it is.

§ 3Uncertainty as constitutive, not added

The mainstream reflex is to model uncertainty as an epistemic deficit — a probability distribution standing in for knowledge we happen to lack, in principle removable by more data. Mises denies exactly this. The uncertainty of the future is not a fact about the limits of our information; it is implied in the very notion of action (ch. VI §1). An actor for whom the future held no openness would not be deliberating between alternatives at all — there would be nothing to deliberate about. Uncertainty is a condition of the possibility of action, not a contingency afflicting it.

This is why the register matters. There are two things one can be certain or uncertain about, and they belong on different tiers:

The first row is a strict derivation from the action axiom, and it is logical in kind — deduced, not observed. The engine will let you query its provenance but will not let you defeat it: the grounding-aware cancel refuses to retract a theorem, pointing you instead at the premise it rests on. The second row is a defeasible conjecture, and cancel retracts it without protest. The same operator, two outcomes — and that asymmetry is the "two modes of one thing":

faces_uncertain_future(H) <= acts_purposefully(H)
grounding("faces_uncertain_future", "actor")   # -> "theorem"
truth_kind("faces_uncertain_future", "actor")  # -> "logical"  (a priori, not empirical)

cancel("faces_uncertain_future", "actor")           # the category…
canceled("faces_uncertain_future", "actor")         # -> false  — cancel REFUSED

This is where the essay pays a debt to the methodology essay's §4.5. Lachmann's radical subjectivism warned that once expectations are themselves subjective and divergent, the future is genuinely open, and any determinate result "smuggle[s] in assumptions that radical subjectivism denies" — so determinacy must be earned per-derivation, never assumed. Uncertainty-as-defeasibility is that maxim mechanized. A particular outcome is never a standing fact of the model; it is a conjecture that holds only until the world rules, and the ladder makes its provisional status visible in its very grounding.

There is a foil worth naming and not over-claiming. One can model a Laplacean omniscient observer — a demon for whom the future is closed — as a non-actor for whom nothing derives, illustrating the contrapositive of VI §1: no openness, no action. In the companion corpus this is written as an explicit control, not proved as an in-system theorem — it shows the shape of the entailment; it does not certify that no other derivation could reach the conclusion. Honesty about which claims are checked and which are illustrative is part of the discipline the methodology essay's §6 is about.

§ 4Class and case probability

Uncertainty admits of two utterly different treatments, and conflating them is the oldest error in the economics of risk. Mises separates them at the root:

"There are two entirely different instances of probability; we may call them class probability (or frequency probability) and case probability (or the specific understanding of the sciences of human action)." — ch. VI §2

The difference is not one of degree — not "more" versus "less" information about the same kind of thing. It is a difference of object and therefore of method:

Mises's own definitions anchor each column. Class probability: "We know or assume to know … everything about the behavior of a whole class of events or phenomena; but about the actual singular events or phenomena we know nothing but that they are elements of this class" (VI §3). Case probability: "We know, with regard to a particular event, some of the factors which determine its outcome; but there are other determining factors about which we know nothing" (VI §4). And the numerical line is categorical:

"Case probability is not open to any kind of numerical evaluation." — ch. VI §5

Axioma's kind axis holds the two apart exactly as Mises does — and this is the essay's second load-bearing identification, after uncertainty-as-defeasibility. The kind axis is orthogonal to the grounding tier: a fact's tier records how firmly it is held (axiom … datum); its kind records what sort of knowledge it is (logical, empirical, motive, transcendental). Class probability is empirical — a frequency, a measured number about a class. Case probability is motive — thymological understanding of a singular case, carrying no number at all:

postulate/empirical class_probability("roulette_spin")            # frequency of a class
axiom/motive        case_probability("this_venture") by entrepreneur  # understanding
truth_kind("class_probability", "roulette_spin")   # -> "empirical"
truth_kind("case_probability",  "this_venture")    # -> "motive"

This is precisely Mises's praxeology / thymology line drawn as a kind distinction — and it is why the action axiom itself is not tagged motive. The bare category of action is praxeological a priori: logical in kind, its strict conclusions deduced. Thymology — the specific understanding of this actor's appraisement of this unrepeatable venture — enters only at case probability. (The word thymology is Mises's own, coined later — in Theory and History, 1957 — for what Human Action ch. VI §2 calls "the specific understanding of the sciences of human action": ch. VI supplies the distinction, the later work supplies the label.) Reserving motive for the singular case, and keeping the deductive spine logical, is what makes the causality / teleology fault line of VI §2 executable rather than merely narrated. (The corpus's Rothbard scripts tag the action axiom motive for provenance bookkeeping; this pass deliberately draws the sharper line, so that logical names the a priori and motive names the thymological — the ledger in the book-pass notes.md records the divergence.)

And here the :: Ordinal discipline of §2 does double duty. The entrepreneur does not compute a probability for his venture — VI §5 forbids it. He ranks it: this opportunity above that one, this venture over its runner-up. Ranking is comparison and selection — licensed. Attaching a number to the singular case — venture_rank * 0.5 — is the same category error as subtracting two moments, and the same static rule catches it. One typecheck rule enforces two Misesian doctrines at once: that praxeological time is ordinal (ch. V) and that case probability admits no numerical evaluation (ch. VI §5). The unity is not a coincidence of implementation; it is that both are the ordinal discipline — positions to be ranked, not magnitudes to be measured — showing up in two doctrines.

§ 5Entrepreneurship: profit and loss as the residual of expectation

If every action reaches into an unknown future, then every actor is, in the relevant sense, a speculator:

"Every action refers to an unknown future. It is in this sense always a risky speculation." — ch. VI §1 · "Every action is speculation." — ch. VI §4

The entrepreneur is that function made a vocation — the actor who bears case-probability uncertainty as his business. That he speculates is a theorem (it follows from acting at all). But any particular anticipation of profit is a conjecture: the plan defeasibly warrants success, and the world then rules. Profit and loss are not two different mechanisms; they are the two residuals of one defeasible structure — the conjecture the world vindicates (profit), and the one it defeats (loss):

speculates(H) <= acts_purposefully(H)             # a theorem — every actor speculates
expects_profit(H, V) <~~ plans(H, V)              # defeasible — the plan CONJECTURES profit

cancel("expects_profit", "bob",   "venture_y")    # the world defeats Bob's venture…
canceled("expects_profit", "bob",   "venture_y")  # -> true   — this residual is his LOSS
canceled("expects_profit", "alice", "venture_x")  # -> false  — hers stands: PROFIT

Note what defeasible does and does not mean here. It models only the retractable status of the expectation — standing, or canceled — never a credence, a degree, or a number; a conjecture is all-or-nothing, which is exactly what keeps it on the right side of the non-numerical case probability of §4. The entrepreneur's appraisement itself — the thymological understanding that formed the plan — is what the motive kind records, and it too carries no number. Defeasibility is the status of the expectation; the appraisement is the understanding behind it. So "defeasible conjecture" names a qualitative epistemic flag, not a graded forecast — no probability slips in through the back door.

There is a deeper mechanism than a bare cancel, and it is what makes the treatment a genuine theory of error rather than a bookkeeping of outcomes. An entrepreneurial expectation rests on a premise about the future — that demand will hold, that a price will move. When the world falsifies that premise, everything built on it must go with it. This is retraction, not contradiction: the actor was not wrong in logic; his premise was overtaken by events. Axioma's forget_cascade is the truth-maintenance step — it withdraws the falsified premise and every conclusion that named it, and lazy re-derivation keeps only what still has support:

venture_pays(H, V) <~~ plans(H, V) and demand_stays_high(V)
# … while the premise holds, venture_pays(carol, venture_z) is a conjecture …
forget_cascade("demand_stays_high", "venture_z")   # demand collapses — retract the premise
# venture_pays(carol, venture_z) is now gone: the expectation resting on it withdrawn

Profit and loss, on this reading, are the epistemic residue of acting under case probability — the difference between the future the entrepreneur appraised and the one that arrived. The engine does not forecast which; it certifies the form — that the anticipation was always defeasible, that its defeat is retraction, and that what survives a falsified premise is exactly what still has independent support.

§ 6Praxeological prediction: apodictic on direction, silent on magnitude

The final doctrine is the one that most sharply distinguishes praxeology from econometrics, and it is the natural test of everything above. Praxeology does predict — but only qualitatively, and with a certainty no statistical model can claim:

"Praxeological knowledge makes it possible to predict with apodictic certainty the outcome of various modes of action. But, of course, such prediction can never imply anything regarding quantitative matters." — ch. VI §7

"We can predict … that — other things being equal — a fall in the demand for a will result in a drop in the price of a. But we cannot predict the extent of this drop." — ch. VI §7

"there are no constant relations between what are called economic dimensions." — ch. VI §7

The two halves land on two different tiers, and the engine keeps them apart without being told to. The direction is a strict theorem — take the antecedent as given and the qualitative conclusion derives, a priori, logical in kind. The extent is not a praxeological theorem at all: no rule derives the magnitude, because there are no constant relations to derive it from. The relation that would hold the number simply stays empty — and its emptiness is the doctrine, not a gap:

price_drops(G) <= demand_falls(G)          # the qualitative law — a strict rule
axiom demand_falls("good_a")
grounding("price_drops", "good_a")          # -> "theorem"   (apodictic direction)
truth_kind("price_drops", "good_a")         # -> "logical"   (a priori)

relation price_drop_extent(g, pct)          # the magnitude…
# … no rule populates it. len(price_drop_extent) == 0 — praxeology is silent on extent.

This is the methodology essay's §7.5 and §8 in miniature. There the point was general: Axioma formalizes the qualitative categories and the deductive steps, introducing "no numbers, no measurement, no functional forms," because "there are no constant relations" is Mises's own complaint against quantification. Here it is executed: the direction carries the strongest grounding the engine has, and the magnitude is left, by construction, underived — the boundary between the a priori core and the applied, data-fed layer where "verifiable prediction belongs" (methodology essay §8) drawn as the presence and absence of a derivation.

§ 7Why no new machinery was needed

Every mechanism in this essay was already in the engine before time and uncertainty were attempted. That is not an incidental convenience; it is the essay's strongest claim, and it deserves to be stated as one.

• Uncertainty needed no probability monad, no modal ◇, no belief measure. It needed defeasibility — the conjecture tier and the grounding-aware cancel — which the ladder already had. The certainty-of-uncertainty is a theorem cancel refuses; each outcome is a conjecture it defeats.
• Time needed no temporal logic, no LTL, no clock type. It needed the ordinal discipline — comparison and selection licensed, arithmetic a static category error — which :: Ordinal already enforced for wholly unrelated reasons.
• Class versus case probability needed no new type and no numeric machinery. It needed the kind axis — empirical versus motive — already present to mark the causality / teleology line.
• Profit and loss needed no market simulator. They are the two residuals of one <~~ conjecture under cancel, and error is forget_cascade — the truth-maintenance step the engine already used for retraction.

The methodology essay's §7.5 framed mechanization as an auditor: if a purported praxeological law derives mechanically, the apriorist claim is vindicated in the strongest way; if it cannot without smuggling in an empirical premise, the failure is diagnostic. Turn that auditor on time and uncertainty and it returns a clean bill: the hardest categorial doctrines in Human Action fall out of the existing concept-and-grounding kernel with zero engine changes. A kernel that carries the action axiom, scarcity, and exchange (methodology essay §7) and also carries time, uncertainty, class/case probability, entrepreneurship, and the qualitative-prediction boundary — without extension — is a kernel cut to the shape of the theory. The feature gate never fired, and its not firing is the result.

§ 8What it captures, and what it brackets

Stated plainly, so the boundaries are not mistaken for oversights.

It captures, each as executed, self-checking claims:

• Time as a praxeological category — ordinal, directional, with the past as sunk datum — not measured duration (§2).
• Uncertainty as constitutive of action, a priori certain in the category and defeasible in every particular (§3).
• The class / case distinction as a kind difference — empirical frequency versus thymological understanding, the causality / teleology line (§4).
• Profit and loss as the two residuals of one defeasible structure, with error as retraction (forget_cascade), not contradiction (§5).
• Praxeological prediction as apodictic on direction and, by construction, silent on magnitude (§6).

It brackets — deliberately, and these are ledgered as residuals, not modeled:

• Time preference, originary interest, periods of production (HA ch. XVIII, "Action in the Passing of Time"). This pass is the category of time (ch. V) and uncertainty (ch. VI), not the interest theory. The methodology essay's §7.5 lists "action → time-preference" among the qualitative arrows; developing that arrow is a separate pass, partially covered already in rothbard-mes/04_time_preference.ax.
• Equilibrium constructs — the plain and final states of rest as market constructs (HA ch. XIV). Script 01 stays with the pure temporal category and does not reach for them.
• The full theory of the market process — Kirznerian alertness, the equalization of the rate of profit, appraisement versus anticipation, the entrepreneur-promoter. Entrepreneurship is modeled here only as the defeasible-outcome residual, not as a process theory.
• A real frequentist model of class probability. Class probability is modeled as a bare empirical postulate with an illustrative frequency; the point is only the kind contrast with case probability, not a theory of insurance.
• The omniscience control is a control, not a theorem (§3) — it exhibits the contrapositive of VI §1; it does not prove it in-system.

The bracketing is the same discipline as the capturing. An executable praxeology earns its authority by being exact about which claims it checks — and equally exact about which it declines to.

§ 9In this project: the runnable corpus

This essay is the conceptual companion to a body of runnable material. Paths are in the axiomalang repository.

• tests/axioma/showcase/executable-praxeology/time_and_uncertainty.ax — the verified companion to this essay: the action axiom, the time-as-category theorems of §2, the uncertainty-as-defeasibility asymmetry of §3, the class/case kind axis of §4, and the speculation/cancel residual of §5 — each asserted with expect(...) (13/13 pass).
• tests/axioma/showcase/books/mises-human-action/ — the six-script book-pass over ch. I, V, and VI: 01_time_as_category.ax, 02_uncertainty_constitutive.ax, 03_class_and_case_probability.ax, 04_entrepreneurship_profit_loss.ax, 05_praxeological_prediction.ax, and 06_action_inaction_condition.ax — the act/refrain condition (the three prerequisites, cannot-act vs. choose-not-to-act, the defeasible outcome), which also holds the cardinal expected-utility/behavioral model to the empirical layer beside the a-priori core (44 assertions), with README.md and a notes.md ledger of verified quotes, modeling decisions, and residuals.
• .../mises-human-action/test_time_uncertainty_typecheck.sh — the static half: pins later - now (time), venture_rank * 0.5 (the singular case), and the expected-utility formula's arithmetic on ordinal preference ranks as --typecheck category errors, and certifies the comparison/selection-only program clean. One rule, the two doctrines of §2 and §4 — and the executable proof that the EU model is not praxeology.
• "Executable Praxeology in Axioma.md" (this directory) — the methodology essay this one presupposes: the machine (its §6), the grounding ladder (§7), the quantification-versus-logical-formalization line (§7.5), and the internal-Austrian limits of Hayek and Lachmann (§4.5) that §3 here operationalizes.
• tests/axioma/showcase/books/rothbard-mes/04_time_preference.ax — the adjacent interest-theory material this pass deliberately brackets (§8).

Together they realize the §1 claim in practice: time and uncertainty are two modes of one fact, and the fact that one kernel carries both — checked, not simulated — is the evidence that the kernel is the right one. Calculemus.

§Sources

• Ludwig von Mises, Human Action: A Treatise on Economics (1949), ch. V ("Time") and ch. VI ("Uncertainty") — §§1–7. Quotations are character-exact against the Mises Institute online edition.
• Ludwig M. Lachmann, The Market as an Economic Process (1986) — the radical subjectivism of expectations that §3 operationalizes.
• For the wider methodological apparatus — Hoppe, Rothbard, Machlup, Hutchison, Caldwell, Hayek, Kotarbiński, von Wright — see the Sources of the companion essay, "Executable Praxeology in Axioma.md", which this one presupposes rather than repeats.