The strongest case against this page
Here is the objection at full strength, because it deserves to be heard before any rebuttal.
Alfred North Whitehead's Process and Reality is a speculative metaphysics from 1929 with almost no working uptake in the sciences that actually move goods. Physics gets by with state vectors and fields. Chemistry gets by with molecules. Logistics gets by with lead times, safety stock and reorder points. None of these disciplines needed "actual occasions" to become quantitative and useful. Dressing a claim about data pipelines in Whitehead's vocabulary is decoration, not argument. Whether a supply chain platform should stream port telemetry or batch it once a day is a question of latency, bandwidth and cost. It has nothing to do with whether reality is fundamentally made of events.
This lands. It should. Whitehead is not cited in any tariff-classification engine, any freight-forwarding contract, any customs bond. A supply planner who has never heard the word "occasion" runs a functioning operation every day. If the case for continuous intake needed metaphysics to be true, it would be in trouble.
What survives the objection
It survives on uptake and fails on substance, because the argument does not actually need Whitehead's whole apparatus. It needs only the negative half: that treating persistence as free — as something a system simply has, requiring no accounting — is a modelling error whenever the thing being modelled is held together by throughput rather than possessed by inertia. That half does not depend on 1929 metaphysics. It is independently established, in a different vocabulary, by Ilya Prigogine's work on dissipative structures: systems that exist only under continuous energy flux and collapse the moment the flux stops. A supply chain is exactly this kind of object. It has no substance-level existence apart from the shipments, filings and clearances currently moving through it. Stop the throughput and there is no chain left to describe, only warehoused inventory and stranded containers — the wreckage of a process, not a smaller version of it.
So the metaphysics is not load-bearing. It is a vocabulary for a result that operations research already half-knows and rarely states plainly: a supply chain's identity is a rate, not a shape. Cost and latency then decide when you pay for continuity. They do not decide whether continuity is what the object is made of.
The planner's version of the problem
A supply planner builds a plan on a stated set of assumptions: a corridor is open, a tariff schedule holds, a supplier's export licence is current, a port's dwell time is within its usual band. Each assumption is, functionally, a frozen fact — true as of when it was checked. The plan then runs for weeks against a world that does not hold still. Somewhere in that interval a customs authority issues a tariff notice reclassifying a component, a port authority files a congestion surcharge, a supplier's filing quietly lapses. The plan does not fail because the planner reasoned badly. It fails because the plan inherited a fact rather than a rate, and the fact stopped being true on a Tuesday nobody was watching.
This is the characteristic failure of the domain: a plan survives on an assumption invalidated by a filing nobody read. It is not a data-quality problem in the ordinary sense — the filing existed, was public, was even machine-readable. It is an intake problem. The plan's inputs were substances: a tariff rate, a licence status, a berth window, each recorded as a property true at a point in time and then carried forward by default. Nothing in the plan's architecture asked what was maintaining that property, or whether the maintenance had stopped.
Three ways to watch a port
Consider how each generation on the intake axis would represent a single container corridor.
A corpus-fed system — the Large Language Model shape of intake — knows the corridor the way a trade almanac knows it: named ports, typical transit times, standard tariff codes, all true as stated on the date the corpus was assembled. It can describe the corridor fluently. It has never seen a ship actually move through it, and it has no mechanism for noticing that the almanac's entry on the tariff code went stale four months ago.
A bounded-scene system — the Large World Model shape — watches the corridor properly, for as long as the watching lasts. Feed it a season of port telemetry and it will model dwell times, queueing, berth allocation, with real fidelity to process. But the episode ends. The model that inferred a congestion pattern from March's telemetry does not carry that inference into April unless someone restarts it with fresh data and no memory of why March looked the way it did. It has seen becoming and then lost it.
A streaming, provenance-carrying system — the shape argued for under the name Large Universe Model — holds the corridor as a set of beliefs, each attached to the filing, manifest entry or telemetry reading that supports it, each with a decay clock. The berth-window belief is not "twelve hours," full stop; it is "twelve hours, last confirmed four hours ago by AIS ping, degrading in confidence as no new ping arrives." The tariff-code belief is not fixed at ingestion; it is overwritten the instant a customs notice contradicts it, and the overwrite is logged, so a planner can ask why the plan changed and get a citation rather than a shrug. This is the only one of the three that represents the corridor the way the corridor actually exists: as an ongoing thing that must be kept true, not a thing that was true.
Two objections that bite
The first: sampling is unavoidable everywhere. Port telemetry arrives at some interval — an AIS ping every few seconds, a customs filing whenever it is filed. A stream is a lot of snapshots close together, not a magic escape from discreteness. On this view the difference between a corpus and a live feed is density, not kind, and calling the dense end of that spectrum "terminal" is overclaiming.
The objection is right about discreteness and wrong about what matters. The distinction is not sampling frequency but whether successive samples are stitched into a lineage in which each state inherits from, and can be overturned by, the last. A thousand port pings logged as independent rows in a table, each classified on its own, is a corpus with fine grain — still a set. The same pings feeding a berth-window belief that updates, timestamps its own confidence and decays without contact is a sequence, in Whitehead's narrow technical sense: each occasion inheriting from its predecessor. A planner querying the second system gets an answer with an expiry date. A planner querying the first gets a table and has to supply the expiry date from memory, which is exactly the gap the missed tariff filing falls through.
The second objection cuts closer to the bone. If everything is process and nothing truly endures, then a stored belief — "the licence is current, confidence 0.8, last checked Thursday" — is itself a little substance fiction: a thing claimed to persist on a server, no less reified than a fact frozen in a corpus. The Large Universe Model would be smuggling the enduring object back in under the name of a "belief record."
This is the sharpest challenge, and it costs something. Any store that persists at all is doing some reifying; there is no representation with zero substance-shape. But Whitehead's own answer to this, for an enduring rock as much as for a belief record, is that endurance is legitimate when it is maintained rather than assumed — a society of occasions held together by continuous re-inheritance, not a fact taken as given once and never revisited. The test is behavioural: does the store degrade when the input stops? A corpus fact does not degrade; it sits, true until manually corrected, however stale. A provenance-carrying belief about a licence, fed by no further filings, loses confidence on a clock and eventually flags itself as unsupported. Same shape on disk. Opposite ontological status. The planner's tariff belief that quietly ages out of trust after ninety days without a confirming filing is doing something no static record does, even though both are, physically, a row in a database.
Where the axis actually stops
None of this licenses the strong claim that fixed data is worthless. Most of a supply chain is not in crisis at any given hour: standard tariff schedules, established shipping lanes, long-term supplier contracts are stable for good reason, and a corpus captures that stability cheaply. The failure mode is specific, not general — it appears exactly where an assumption is being held up against a gradient: a licence near expiry, a corridor near a political decision, a supplier near default. Those are the places where maintenance cost is not low and slow, and where frozen intake stops being adequate.
Add more sensors, wider span, tighter fidelity, and the representation improves without changing kind. What it cannot do is add a further tense beyond all-streams-still-running-with-lineage-retained, because there is no view of a corridor further along that axis than watching it, without interruption, while keeping the record of why each belief about it is still owed to be true. That is the narrow claim. The corridor is a rate. A supply planner's job is to know the rate, not the shape — and the filing nobody read was always going to be missed by any system that had already decided the shape was the whole answer.