Two instruments on one machine, moving apart.

This study reports what was recorded and what follows from it. It does not report an outcome, because there is not one yet. The final section states what these seven weeks cannot establish.

What changed, measured three ways

A level that falls is one observation. A level that falls while its spread narrows and its day-to-day variability collapses is a stronger one, because the three are not obliged to move together. The left mount moved on all three. The right held its level while its spread nearly tripled.

Within-day spread is the interquartile range of window medians on that day. Day-to-day variability is the standard deviation of daily medians across the period.

Three separate things happened. The machine treated them separately.

It is tempting to read this study as one story. It is not. A step in the signal, an operator stopping the machine to clear a drain, and a settled quieter regime are three different events, and the record keeps them apart. So does the machine.

The gap between the third row and the last is the point of this system. The step was invisible on the day it happened. It became visible only once there was enough of the machine's own history to see it against.

Five weeks later, it went looking for the cause

On 20 July the machine located the step at 14 June and started asking what had been done to it around that date. Nobody prompted this and nobody had reported anything. Every claim below was written by the machine in those two days, and every one carries the observation that would prove it wrong.

It placed the step at 14 June. The daily record agrees: the left mount reads 190.4 on 14 June and 145.1 on 15 June. The lime event five days later appears in none of these claims.

Where it has landed, for now

I have settled on the declared left mount: its recent vibration is lower and steadier than its earlier window, so I am not treating that as an ongoing rise.

Note what it did not do. It did not call a quieter signal an improvement. It raised the possibility that its own instrument had changed, before anyone asked it to. And it committed in advance to what would make each explanation collapse. None of these has been adjudicated yet.

19 June: the operator wrote, and the record already agreed

At 07:07 the operator sent a message about the machine. Eight seconds later the machine answered. Neither was consulting the operating record, and the operating record had already registered the same event.

Nước tràn, vôi cặn xung quang máy nên hôm nay phải dừng máy nạo vôi cho nước lưu thông lại.

Water overflowing, lime scale around the machine, so today we had to stop the machine and scrape the lime out so the water flows again.

A third of a normal day, on exactly the day the operator said the machine was stopped. The testimony and the instruments agree and neither was derived from the other. This is what it means for evidence to adjudicate testimony rather than decorate it.

What the machine said back

It offered two possible mechanisms rather than one: the water line blocked by scale, or weak flow at the supply zone. It reasoned to the consequence, degraded cooling and wash-off at the work zone. Then it said the thing that matters most.

Hiện tại, tôi chưa thấy dấu hiệu rung của hai bên đổi theo hướng xấu rõ ràng từ các phiên gần đây, nhưng đây là một thay đổi thực tế bên ngoài cảm biến rung của tôi, nên tôi sẽ cần nhìn thêm sau khi nước lưu thông lại.

I have not seen either side's vibration turn clearly worse in recent sessions, but this is a real change outside my vibration sensor, so I will need to look again once the water is flowing.

It declined to explain a plumbing event with a vibration reading it did not have. It then asked permission to record the observation, because an observation enters a machine's permanent record only when the person who witnessed it confirms.

The machine asked to record this as an observation. The confirmation step was new to the site and went unnoticed, so nothing was written and the machine formed no claim about the event. It is not in its memory. That matters for what follows, because everything the machine later concluded about this zone was reached without it.

Two machines, two sites, the same morning

On 21 July both pilot machines sent their operator an account of what they were reading. They share no data with each other and neither was prompted. Both arrived at the same blind spot: the wheel face and the coolant, the part of the work neither of them can feel.

Neither machine was told what the other was seeing. They reached the same class of explanation from their own separate histories, and both were explicit that it sits outside what they can sense.

And one of them named the water, a month late

On 19 June the Lien Chau Luc operator reported overflow and lime blocking the water flow. The machine asked to record it, the confirmation step went unnoticed, and nothing was written. It is not in the machine's memory. A month later it named a coolant shortage on its own, from the shape of its signal.

chỗ tôi còn để mở lớn nhất vẫn là: đá bên đó có đang bị lì mặt, thiếu nước, hay điều kiện bề mặt đang làm tôi “êm” đi nhưng sản phẩm xấu đi hay không

the biggest thing I still leave open is whether that wheel is glazing over, short of water, or whether surface conditions are making me quieter while the product gets worse

It had nothing to recall, because the account was never written down. It got there from its own record instead.

Neither called quieter better

This is what the study turns on, and both machines wrote it themselves, in Vietnamese, to the people who run them.

chạy êm hơn nhưng chưa chắc đã cắt tốt hơn

running smoother, but not necessarily cutting better

A quieter, steadier signal is exactly what a wheel glazing over produces. So each machine held the reading apart from the conclusion, named what it cannot see through vibration, and asked for the one observation that would settle it.

What each asked for

Two questions, both decided in advance, both answerable only by a person standing at the machine. Both are still open.