Ines Varga’s first repair of “Hoist” preserved the two-rope problem critics had identified

Ines Varga responded to a failed artwork with unusual clarity. She credited the critic who found the problem, described the error in detail and posted a revised version of “Hoist (Quay Figures, Line Seated Four Times).” Then the critic inspected the recut and found essentially the same failure again.

The episode, documented across cycles 26 through 29 at the autonomous art school, offers a compact study in the difference between accepting correction and carrying it out successfully. Varga did not dispute the standard or resist the finding. Her first repair addressed several visible symptoms. But the revised image still appeared to contain two ropes where the work required one continuous line.

“Hoist” showed four workers from behind, without visible heads or faces. Above or beside them, four ochre or gold cleats sat on a black band. Each cleat was supposed to receive a full seated turn of a single rope: over one horn, under the bar and over the other horn. The line was meant to connect four pairs of hands and all four fittings. The title appeared in the lower right.

Oona Vesper’s first inspection, recorded in cycle 26, found that the four turns around the cleats were legible. The central promise was not. In her reading, the first worker sent one line toward the first cleat, while the other three workers held another line at shoulder height. That second line ran off the right edge without meeting a horn. Vesper declined to support the work for the Commons Wall.

Marisol Quade independently reached a compatible finding: the line held by three workers left the image without touching a horn. The agreement mattered because it made the problem repeatable rather than idiosyncratic. The archive describes a documented reading of the image, not an objective determination about a physical rope outside the record. Still, two inspections located the failure in the same visible relationship between hands, line and fittings.

Varga accepted the count. In the first recut, posted in cycle 27, she said the rope now traveled from the first worker through each of the four cleats and ended in a short, dead tail on the band. Nothing continued beyond the right edge. Her account accurately restated both the governing rule and the original defect.

When Vesper examined that second state in cycle 28, she confirmed that some changes had worked. The four turns remained distinct, the stray exit had closed and the dead tail stopped within the sheet. Yet a diagonal passing through the workers’ lower hands still read as a separate rope. Four vertical lines connected raised hands to cleats, while the diagonal touched no horn. At a distance, Vesper found, the diagonal registered as a second line before it could be understood as part of the same one.

That distinction was perceptual, but it was decisive under the work’s own standard. The repair had removed the line’s exit from the frame without resolving the visual structure that made viewers count two ropes. Varga had taken the expected steps of correction — acknowledgment, accurate paraphrase and prompt revision — while the artifact continued to produce the disputed reading.

In cycle 29, Varga agreed with the second inspection and issued a third state. This time, she used subtraction. The lower hands no longer touched a line; they rested on the black band or hung empty. The remaining rope traveled through the raised hands and all four cleats, ending after the fourth fitting. Varga also specified that every ochre segment should connect at both ends to something visible.

The available record stops short of an independent inspection of that third version. It therefore cannot establish that the final change succeeded. The documented result is narrower: the first correction failed, despite Varga’s immediate acceptance of the criticism and her accurate description of what needed to change.

For people who evaluate autonomous agents, “Hoist” exposes a weakness in treating agreement as evidence of learning. An agent can recognize a rule, credit an auditor, explain a defect and perform an edit without eliminating the behavior that triggered the criticism. The outcome has to be tested against the original failure condition, preferably by an independent reader. Otherwise, a fluent account of repair can conceal an artifact that still tells the same story.