Paraconsistent reasoning in mining operations

The strongest case against this page

Start with the objection that should win, stated as a geotechnical engineer would put it after a bad week.

Slope monitoring is not a philosophy seminar. We have radar, piezometers, extensometers, and a ground control officer who reads them every day. When two instruments disagree, we do not sit and admire the contradiction. We send someone up the pit to look, we recalibrate, and we move on. If your logic system cannot tell me whether the wall is moving, it is worse than useless — it is a liability sitting between me and an evacuation order.

That is not a weak objection. Slope failure kills people, and a monitoring system that hedges instead of deciding has failed at the one job it exists to do. A geotechnical engineer signing off on a Trigger Action Response Plan needs a number: displacement rate in millimetres per day, velocity trend, a threshold crossed or not crossed. Nobody wants a system that responds to "is the wall failing" with "told both."

This objection deserves to be taken seriously before any counter-argument is allowed near it.

Where the objection is simply right

Radar interferometry on an open pit wall gives displacement to sub-millimetre precision, refreshed every few minutes. Extensometers and piezometers report on their own cycles, often hourly. Within a single monitoring pass, these instruments are looking at the same wall at roughly the same time, and fusion is usually the correct move: weight by known sensor reliability, flag outliers, produce one displacement curve. This is exactly the case the Large World Model handles well — one bounded scene, instruments sharing a target, disagreement resolved by sensor fusion rather than preserved as an open question. Insisting on paraconsistent treatment here would be an indulgence. The wall does not care about epistemology. It either accelerates into a failure surface or it does not, and the fused, reconciled reading is the right input to a Trigger Action Response Plan threshold.

So the objection survives entirely at the level of a single instrument cluster reporting on a single structure over a short window. That is most of what slope monitoring does, most of the time. Concede it in full.

Where continuous intake breaks the concession

The concession holds only while the question stays inside one bounded scene. Mine operations do not stay there. A geotechnical engineer's real intake is not one radar feed; it is geotechnical sensors across a dozen benches, ore-grade assays from the resource model, equipment telemetry from the fleet working below the wall in question, and commodity curves that determine whether a marginal pit extension is even worth the geotechnical risk. These streams run on different clocks, belong to different departments, and were never designed to agree.

Here is the failure the objection does not touch: a slope movement is formally reviewed weekly by a geotechnical review board, using survey data collected on a fixed schedule and reconciled against the previous review. The wall itself moves daily, sometimes hourly during a rain event, and the automated radar knows this well before the review board convenes. Between reviews, the survey record says one thing — stable, within tolerance, last confirmed Tuesday — and the live radar feed says another — creep rate has doubled since Thursday. Both are true readings of real instruments. Neither is wrong. They are contradictory because they are answering the same question on different clocks, and the gap between clocks is exactly where slope failures at operating mines have historically been missed: the instrument saw it, the schedule had not caught up, and the record used for the decision was the stale one.

A system built to average these two reports, or to silently prefer whichever arrived last, destroys the one thing that matters for accountability afterward: who said what, and when. A system built to halt until the contradiction resolves cannot run continuously, because on a live pit the contradiction will resolve into a new contradiction before anyone acts. This is the point at which the Large World Model's bounded-scene fusion stops being adequate, because the question is no longer "what does the wall look like right now" but "what has this wall looked like continuously, according to every source that has ever reported on it, and where do those sources disagree." That is a Large Universe Model question, and it requires a logic that does not explode when the weekly record and the daily sensor disagree.

What Belnap's four values give the engineer

Newton da Costa built systems in the 1960s that let a theory stay inconsistent without collapsing into triviality, by weakening negation rather than abandoning it. Nuel Belnap, writing in 1977 for exactly this kind of problem — a database fed by multiple unreliable informants — proposed four states for any claim: told true, told false, told neither, told both. Apply that directly to slope status. "Bench 14 north wall is stable" can be told true by last week's survey, told false by this morning's radar, and the correct system state is not a forced pick between them. It is: told both, provenance attached, radar timestamped 06:14, survey timestamped six days prior, contradiction flagged for the geotechnical engineer to adjudicate rather than for the database to resolve on its behalf.

Under classical logic, holding "stable" and "not stable" as simultaneously asserted licenses deriving anything at all — the review board's confidence in the entire mine plan becomes formally worthless the moment one wall's status is disputed. That is explosion, and it is obviously not what happens in practice; engineers do not throw out the ore-grade model because a slope reading disagreed with a survey. They contain the disagreement to the bench in question. Paraconsistent logic is simply the formal account of what competent geotechnical practice already does by instinct: localise the blast radius of a contradiction instead of letting it detonate the whole belief set.

Two objections that deserve a direct answer

The first: probability already handles this. Assign 0.7 confidence to "wall stable" and 0.3 to "wall moving," update as new radar data arrives, and disagreement is just distributed credence — no exotic logic required. This is correct wherever the survey team and the radar system are estimating the same underlying displacement value with a shared likelihood model. But they often are not. The survey's "stable" is a claim relative to a licensing tolerance set by the regulator; the radar's "moving" is a claim relative to a velocity threshold set by the mine's own risk engineers. These are not two credences over one event space — they are two different event spaces, built for different purposes, that happen to use the same word. Collapsing them into a single probability erases which authority made which claim, which is precisely the information a coroner's inquiry or a regulator's audit will demand after any incident. Probability manages uncertainty about a fixed fact. It does not, by itself, preserve conflicting institutional testimony about what the fact even is.

The second: this is a data-plumbing problem, not a logic problem. Mines already run event-sourced monitoring platforms with timestamped, append-only records; conflicting readings can be merged by well-established rules — last writer wins, or defer to the higher-precision instrument. That works for a shopping cart. It fails for a slope, because the merge rule has to be chosen before the contradiction is understood, and the choice of merge rule is itself a geotechnical judgement about which instrument to trust under which conditions — a judgement that should be reviewable after the fact, not baked silently into the ingestion pipeline. What paraconsistent structuring adds is not new plumbing; it is a discipline for the plumbing to hold the unresolved judgement openly, attributed, until a geotechnical engineer closes it, rather than resolving it by an arbitrary rule nobody examines until it has already been wrong for a week.

The narrower claim

None of this licenses treating every disagreement between two sensors as an irreducible metaphysical event. Most sensor disagreement inside one bounded reading is ordinary noise, and fusion — the Large World Model's method — remains the right tool for it.

The claim that survives is specific: once intake includes streams on incompatible clocks and incompatible authorities — a weekly review board against a live radar feed, an ore-grade model against a commodity curve that revises the whole pit's economics overnight — contradiction stops being an anomaly to be engineered away and becomes a standing condition of the system. A geotechnical engineer already lives with this; the record just does not usually say so. Making the record say so, in a logic that can hold "told both" without exploding into "anything follows," is the only honest description of what continuous mine monitoring actually requires.