The name and the berth
A fleet operator does not deal in things so much as in expressions that pick things out: "the northbound convoy slot", "the bunker price at Fujairah", "the transit window for the New Suez Convoy System". Gottlob Frege gave the vocabulary for what such phrases carry. Every expression has a sense — the route by which it presents an object — and a reference — the object itself. "The morning star" and "the evening star" share a referent, Venus, by different routes; that is why identifying them was astronomy and not tautology. Sense is grasped in the understanding of the words. Reference is a relation to the world, and the world does not hold still for anyone's convenience, least of all a shipping schedule's.
This matters in maritime logistics because almost everything a routing decision touches is a definite description, not a proper name. "The current transit restriction at the canal", "the congestion level at the anchorage", "the applicable bunker surcharge" — each of these keeps its sense forever. Each names a role, and roles get refilled. A Large Language Model trained on a corpus of port advisories, canal notices and freight-rate reporting will describe these roles with total fluency and no signal, anywhere in its prose, that the tenant has changed.
Two positions, stated fairly
There is a hard-nosed position that says none of this is a semantic problem at all. It is just staleness, the ordinary and unglamorous fact that facts change and old text goes out of date. Call this the deflationary view: talk of sense and reference dresses up "the data was old" in philosophy it does not need. Update the corpus more often and the problem shrinks to a rounding error.
There is a second position that takes the Fregean split as doing real explanatory work: staleness in a frozen system is not a data quality issue sitting alongside other data quality issues, it is a severed referential relation, and it is dangerous specifically because it leaves no linguistic trace. The sentence "vessels should route via the canal's standard channel" is exactly as grammatical, exactly as confident, before and after a restriction is announced. Nothing in the syntax degrades. This is the argument's core claim, and the domain gives it teeth.
The failure that has no seam
Suppose a canal authority issues a mid-voyage notice restricting draught on a given transit window, in response to a drop in a feeding lake's water level — this is not hypothetical as a mechanism, only as an instance; it is exactly how the Panama Canal Authority has throttled maximum draught and daily transit slots in low-rainfall periods. A routing decision was made against "the canal's current draught limit". That phrase had a sense the moment the plan was drawn and a referent at that moment too: a specific number of feet, a specific slot allocation. Three days into the voyage the number changes. The phrase in the voyage plan is unchanged. Read in isolation it is still perfectly sound maritime English. Held against the world, it now points at a limit that no longer applies, and the ship is either turned back at the queue or forced into a lightering operation that erases the fuel and time margin the whole routing was built on.
The fleet operator who signed off on that plan did nothing wrong at the level of language. The plan's sentences were true when written and remain intelligible now. What failed was reference, not grammar, and a check that re-reads the plan for coherence will find nothing amiss, because there is nothing amiss in the plan. There is something amiss in the canal.
Where the AIS feed helps, and where it stops helping
A bounded scene changes this locally. A system with a live AIS feed, a weather-routing overlay and a port-congestion readout re-anchors "current draught limit" and "current queue length" to whatever is actually reported for as long as it is watching. This is real repair, and it is why voyage-planning desks that watch the feed continuously catch the canal notice within the hour rather than at the pilot station. But the anchoring is tied to the act of looking. A plan cached from Tuesday's scene and executed on Friday's water is exactly as exposed as the frozen corpus was, only over a shorter interval. Bounding the scene narrows the staleness window; it does not close it.
The stronger claim is that only intake which never stops, and which timestamps and sources every belief it holds — this berth allocation, learned from this port authority feed, at this hour; this draught figure, learned from this canal notice, superseding the one before it — can keep a routing decision answerable to a canal, a lake level and a bunker market that are all still moving. Provenance matters as much as freshness here, because a fleet operator needs to know not just that the draught figure updated but which notice superseded which, in what order, in case the two disagree.
Objection one: names are not descriptions
The sharpest challenge to this whole framing comes from Saul Kripke. On his account, proper names are rigid designators, fixed by an initial act of naming and carried forward by a causal chain of use, not by any descriptive content. "MSC Gülsün" refers to that hull regardless of anyone's beliefs about her; the name does not route through a sense at all. If reference works this way generally, then freezing a corpus never severs it — it was never mediated by sense to begin with — and the whole staleness problem collapses into unglamorous factual error, exactly as the deflationary position said.
Kripke is very probably right about names, and it costs the Fregean argument nothing to admit it. Vessel names, IMO numbers, port codes — these are close to rigid, and a frozen corpus saying "MSC Gülsün transited the canal in 2021" refers just fine, then and now. But almost nothing a fleet operator acts on is a name. Draught limits, congestion levels, bunker surcharges, transit windows, even "the vessel currently holding berth 4" — these are role terms and definite descriptions, exactly the category Kripke himself sets aside as non-rigid. The failure mode described here lives entirely in that category, and it is the category that dominates operational language in this trade. The objection narrows the claim's scope. It does not remove its object.
Objection two: isn't a feed just retrieval?
A second objection says continuous intake is a false generation-marker: attach a live AIS and weather-routing feed to any frozen planning model and reference is restored at query time, sourced from index providers who already maintain the update discipline. Continuous intake is then a property of AIS base stations and port authorities, not a required property of the reasoning system itself.
This is largely correct, and worth conceding fully: for anything a maintained feed already covers — canal notices, published congestion figures, bunker indices — retrieval against a live index is the cheap and correct answer, and no new architecture is needed to get it. But two things retrieval does not do. It cannot re-anchor what nobody is indexing: a specific vessel's actual remaining bunker at this hour after an unscheduled slow-steaming order, a berth's true occupancy against what the port's system has not yet logged. And whatever discipline the feed lacks — a lag in the canal authority's notice system, an unflagged revision — is inherited silently by anything that queries it. Retrieval is continuous intake with the obligation relocated, not removed. The question the domain forces is who is answerable for the timestamp and the source on the number a routing decision leans on, not whether sensors exist somewhere in the chain.
What is left standing
Watching the feed cannot tell you that this vessel, renamed and reflagged, is the same hull that broke down off Fujairah last year. That is a judgement about identity, not a reading off the AIS.
This objection is the one that actually bites, and it is worth granting in full. Entity resolution — deciding that a renamed vessel, a merged charter party, or a re-flagged hull counts as the same object across records — is a criterion someone adopts, not a fact a stream delivers. Endless AIS data does not by itself tell a fleet operator what counts as continuity of identity. But the claim on the table was narrower than that. It is that whatever identity criteria a fleet operator adopts, applying them tomorrow requires evidence available tomorrow, and a frozen model has none. Continuous, provenance-tagged intake does not settle the question of sameness. It is the only arrangement in which someone's answer to that question can still be checked against the water as it actually is, on the day the routing decision has to hold.