Opportunity cost of delayed noticing in mining operations

The objection that should win

A geotechnical engineer reviewing a pit wall does not want a number every second. Slope instability is a slow accumulation of strain expressed through noisy proxies — piezometric pressure, extensometer creep, radar interferometry returns that jitter with rain and temperature before they mean anything. Acting on five minutes of data is acting on noise. Acting on a week of data, aggregated and filtered, is acting on a trend. Optimal stopping theory says exactly this: sometimes the correct decision is to wait, because the value of the next observation exceeds the cost of delay. Weekly geotechnical review exists because someone worked out, empirically, that daily review mostly added false alarms and Sunday overtime.

So the claim that a Large Universe Model — a system that keeps every stream live, with beliefs about the world continuously revised and stamped with when they were last touched — closes an economically important gap looks, on this ground, like an argument for panic dashboards. Mines have tried those. Alarm fatigue is a documented failure mode, not a hypothetical one. If the answer to "how often should we look at the wall" is "constantly," the honest reply is that constant looking has already been tried and mostly produces ignored alerts.

That is the strongest form of the objection, and it deserves to be taken seriously before any of the lineage argument gets to speak.

Where the objection is right

The objection is right that delay is often optimal, and it is right that the cost of shortening an observation interval is real and usually falls on people, not on sensors. A geotechnical engineer who wants hourly slope-stability updates instead of weekly ones is asking for more radar passes, more piezometer channels logged and validated, more model runs reconciling InSAR displacement against borehole pore pressure, and — the part that never shows up in a sensor budget — more hours of a scarce specialist's attention to decide whether a wiggle in the data means anything. Grade control assays, equipment telemetry from haul trucks and shovels, and the copper or gold price curve against which the whole pit is economically justified are already competing for that attention. Continuous intake without continuous judgement is just more noise, faster.

This is the correct half of the objection and it should be conceded without qualification: the weak reading of the thesis — sample everything, always, because faster is better — is wrong, and easy to show wrong. A mine that instruments every joint in every bench and pipes it to a dashboard nobody screens is not managing risk. It is generating an audit trail for the inquiry after the failure.

Where the objection changes shape under pressure

But the objection, examined closely, is not actually about frequency of looking. It is about frequency of deciding. Those are different clocks, and the difference is the whole argument.

A slope that is reviewed weekly and moves daily has an opportunity cost that has nothing to do with alarm fatigue. The movement — say, a bench face creeping from 2mm/day toward the 10mm/day threshold that geotechnical practice treats as the boundary between "watch" and "evacuate the pushback" — becomes true on a Tuesday. It crosses into concerning territory on a Thursday. It is noticed at the Monday review. In the intervening six days, every action that would have been optimal against the true state of the slope was unavailable: rescheduling the shovel out of the pushback, halting blasting on the adjacent bench, pulling the haul road, even simply increasing radar cadence to confirm the trend before it becomes an emergency. None of those actions were foregone because anyone chose badly. They were foregone because the decision-maker was reasoning against a six-day-old world while standing in a live one.

That is the distinction the optimal-stopping objection collapses. Optimal stopping says: given a live feed, choose when to act on it. It does not say: go a week without the feed and call that choice. A geotechnical engineer who watches the extensometer trend tighten in near-real time and decides, deliberately, to wait two more shifts before recommending an exclusion zone — because the confidence interval on the acceleration is still wide — is exercising exactly the judgement optimal stopping describes. A geotechnical engineer who finds out on Monday that the trend tightened on Wednesday has not exercised judgement at all. The option to defer was never on the table, because deferral requires knowing there is something to defer on.

"You're describing a scheduling problem, not a category of intake. Review the wall twice a week instead of once and the gap shrinks. No new architecture required."

This is worth answering directly, because it is the natural next move. Shrinking the review interval from weekly to twice-weekly does shrink the gap — and it shrinks the pool of foregone actions proportionally, at the cost of the same specialist attention spread thinner. It does not change what kind of gap it is. The blind interval, whether six days or three, still opens and closes on a schedule set by human bandwidth rather than by the physical process being watched. A rockmass does not creep on a review cadence. Ore grade does not stay stable between assay batches out of courtesy to the sampling schedule. The commodity curve against which the whole pushback is valued does not pause overnight because the site's mine-planning software runs its price feed weekly. What a continuously live model — beliefs about slope displacement, ore grade, equipment health and commodity value all carrying a timestamp for when each was last actually touched, decaying in confidence as that timestamp ages — buys is not more looking. It is the collapse of the gap between "the fact became true" and "someone is choosing with it in view," so that the engineer's deliberate wait, when she chooses one, is a wait against a live posterior rather than a wait imposed by the calendar.

Cost, conceded properly

The second objection with real force in this domain is economic, not conceptual: instrumenting every stream continuously is expensive, and most decisions at a mine do not need it. A weekly grade-control assay cycle is often correct. Reconciling shovel-level ore tonnage against the plan daily rather than hourly is often correct. Nobody is arguing that a haul truck's engine telemetry needs the same intake discipline as a slope fifty metres from an active pushback. This should be conceded in full: the economics of continuous observation are a cost curve, and for the majority of decisions on a mine site the curve says wait.

Where it does not say wait is exactly where the value at risk is asymmetric and fast-moving — active benches above haul roads, tailings dams, anything where the foregone action is not "recalculate the plan a day later" but "get people and equipment out of the way before the ground moves." Sensor and telemetry costs have fallen enough over the past decade that continuous radar and automated piezometric logging are now standard practice at major operations precisely because the industry, independently, priced the same trade-off the argument describes and came out on the side of closing the interval where the tail risk is thick.

The lineage, applied

What actually survives

The narrow claim is this, and no wider: a geotechnical engineer who defers action deliberately, watching a live signal tighten toward or away from a threshold, has not incurred an opportunity cost of delayed noticing. She is doing her job. The engineer who finds out at the Monday review that the wall crossed the threshold on Wednesday has incurred exactly that cost, whether or not the eventual call was correct. Continuous intake does not tell the mine to act faster. It tells the mine the difference between the two engineers, and makes the second one avoidable — not by working harder at the review, but by removing the calendar as the thing that decides when the fact becomes visible.