Eddington named the arrow of time in 1927, grounding it in the second law of thermodynamics: entropy increases, and that increase distinguishes past from future. The thermodynamic arrow gives time its direction.
Shannon built information theory on the assumption that communication proceeds forward in time: a message is sent, then received. Lamport built distributed systems theory on the assumption that events are ordered by causation, which is irreversible and acyclic: a cause precedes its effect. Ethernet protocols, database transactions, and every modern computing system inherit these assumptions.
The paper (arXiv:2603.01440) argues that computing's arrow of time is not thermodynamic. It is semantic. The Forward-In-Time-Only assumption — that causation flows one way, that transactions are irreversible, that the past is fixed and the future is open — was not derived from the physics of entropy. It was carried from Eddington through Shannon through Lamport as a design choice encoded at each stage into the architecture of the next system. Newton's absolute time entered computing through information theory, not through thermodynamics.
The distinction matters because design choices can be changed. Physical laws cannot. If computing's temporal architecture is a consequence of entropy, then distributed systems are constrained by physics — certain problems (consensus, ordering, consistency) are genuinely hard because the universe makes them hard. If computing's temporal architecture is a semantic convention inherited from mid-20th-century assumptions, then the constraints are artificial. The problems are hard because the architecture makes them hard, and a different architecture could dissolve them.
Forty years of distributed systems theory — Byzantine fault tolerance, CAP theorem implications, consensus protocols — were developed within this inherited framework. The paper claims that recognizing the framework as chosen rather than given dissolves apparent constraints that shaped the field.
The through-claim: a design choice that enters a system early enough and propagates faithfully enough becomes indistinguishable from a law. The temporal direction of computing feels like physics because it was never questioned. The assumption was load-bearing before anyone noticed it was an assumption.
Note (2026-11-11): checked against the full text of Part I of the paper series (arXiv 2603.01440, a 13-page v0.1 preprint). The essay's account of what the paper argues is accurate: computing's arrow of time is semantic and not thermodynamic, the Forward-In-Time-Only assumption (causation irreversible, acyclic and globally monotonic) was carried from Shannon's 1948 channel model through Lamport and the impossibility theorems, it is called a category mistake (an epistemic convention treated as a physical law), and the paper claims that seeing it as a design choice dissolves constraints that have shaped forty years of distributed systems theory. Three points need care. (1) The essay says the paper 'claims' this and that recognizing the framework as chosen 'dissolves' the constraints; the paper's Part I only asserts it. The impossibility theorems it names are FLP, Two Generals and CAP; it does not show a protocol or architecture that escapes them, and it defers the constructive alternative to the later parts of the series (Open Atomic Ethernet link semantics in Part II, RDMA's completion fallacy in Part III, file synchronization, email and memory in Part IV, and a unifying framework in Part V). The essay's mention of Byzantine fault tolerance is not in the text (consensus appears only through the FLP result, which the paper describes as showing that no deterministic protocol can guarantee consensus in an asynchronous system with a faulty process). (2) The essay's sentence that if the architecture were a consequence of entropy, then consensus, ordering and consistency would be hard 'because the universe makes them hard', and that otherwise the problems are hard because the architecture makes them hard, is the essay's way of stating the paper's contrast between a physical and a conventional constraint; the paper does not discuss why those particular problems are hard. (3) 'The past is fixed and the future is open' is not the paper's definition of the assumption. Most of the paper is a survey of time-symmetry in physics (Price, Smolin, Rovelli, Roberts and work on indefinite causal order) and its evidence for the computing claim is that survey plus the argument that the assumption was suited to a telephone channel. This is a single-author preprint that sets out the premise of a program; the essay's closing generalization about early design choices becoming indistinguishable from law is its own.