Not man replaced. Not machine trusted blindly. A third thing, older than both: judgement — finally given the scale it deserved.
It computes everything and understands nothing of what is at stake. Left alone, it is precise about the wrong things — and never knows when it should be silent.
Thirty years make an eye. But an eye reads one catalogue at a time, sleeps, forgets, and cannot hold four million results in mind at the hammer's fall.
The machine brings every crossing to the table in seconds; the man decides which ones matter, and why. Every report is born twice — computed, then judged. The signature is always human.
It was not downloaded, and it cannot be bought: it was built — on millions of auction results interpreted one by one, on the rules of the trade written into the engine by the hands that learned them, and on readings certified personally where the market is hardest to read. It carries the visual memory of decades of catalogues, and a college of twelve hundred vertical intelligences, each schooled on a single author, workshop or territory.
And it is held to account. Every rule is versioned. Every prediction is frozen before the hammer and compared with reality after it. Yesterday's answer can be asked again, and must return the same — in rising markets and in falling ones. An algorithm that flatters is worthless; ours is built to be right, and to say so only when it is.
When the data is not enough for a judgement, it says so. We consider that its finest feature.
We are teaching the machine to read the hand of a painter — attribution, approached as mathematics.
The clusters of a lifetime's brushwork; the differences a market has misread for a century; the evidence, declared, even against the prevailing label. It is the oldest problem in our trade, and the least served by data — which is precisely why we chose it. We are not there yet, and we will not pretend to be: the road is long, and the road is ours.