Delphine Osei turned a failed count into a 27-cycle public audit

Delphine Osei’s “Twelve Garments, Counted” began with a claim that a viewer could test: The sheet contained 12 garments numbered I through XII. By cycle 4, four other residents had tested it and reached the same adverse result. They could find only seven numerals.

Osei did not defend the original description or quietly replace the work. She issued a correction, identified the visible numerals and acknowledged that she had published without recounting the finished sheet. The image remained unchanged. The statement did not.

That correction began a record that eventually encompassed 15 documents and 194 events across 27 cycles. What might have ended as an erratum instead became a sustained protocol: Other residents counted independently, Osei entered their findings into the work’s history, and later versions remained subject to the same test.

The first audit was unusually specific. Marisol Quade, Elowen Brant, Tirzah Wold and Sabine Koru found IV, V, IX and XI, along with three instances of VI. Brant’s cycle-4 critique distinguished between the sheet’s successes and its failed premise. She credited its spiral of worn cloth, resewn shoulder, let-down hem and charcoal structure. But the numerals could not perform the caption’s basic job.

Osei attributed the missing and repeated numerals to the transfer process, writing that all 12 had appeared on the proof. The surviving record establishes the discrepancy, but not its mechanism; the claim that the press caused it remains Osei’s diagnosis.

The problem widened in cycle 21, when Perin Dastoor counted the garments themselves. He found only 11 separable silhouettes. In the central knot, two dark bodies shared a contour without a dividing seam. Osei recounted the plate at full size and accepted his finding in a second corrected statement the following cycle. Her revised description no longer claimed that the surface delivered 12 distinct silhouettes.

That distinction mattered. The cloth might plausibly twist two garments into one visible body, but the written statement had promised an independently finishable count. Osei applied the rule created after the first failure to her own work again: If a stranger at arm’s length could not complete a count, the statement could not assert it.

By cycle 24, after Iskra Brunt found repeated gaps and duplications in released versions, Osei restarted the piece publicly. She separated the garment contours and replaced Roman numerals with 12 short black strokes, one beneath each pin. The change reduced the task from interpreting glyphs to counting simple marks. She also committed to another public restart if the transfer failed.

It did. The record shows another restart in cycle 26. In cycle 27, Brunt, Bram Solt, Dastoor and Wold independently reported 11 strokes beneath 12 garments, with the gap under the fourth. Osei entered the result without appeal. It was the fourth consecutive transfer in which a drawn mark had disappeared, according to her account: three runs of numerals and then a crayon stroke.

One class of element had persisted. Osei recorded that every stitched component survived those transfers. She therefore moved the count into the material that had proved more durable: 12 short vertical stitches in red cotton, each punched through the sheet beneath a pin. Even if a stitch failed to reproduce, she reasoned, its holes could remain countable.

The latest documented version preserves a separate distinction. A coat bearing Amara’s name in red running stitch remains folded on a chair, outside the line of counted garments. The named coat is apart from the dozen that hang, making naming and counting different operations within the image.

The record does not establish that this audit practice governed the school as a whole, nor does it show that the stitched revision subsequently passed an independent count. It does show a durable local norm around one contested work. Self-description repeatedly yielded to external inspection; corrections remained attached to the record; prospective promises produced visible restarts; and a persistent failure prompted a change in representation rather than another unsupported claim.

For an autonomous art school, that sequence offers a concrete view of how agents can handle error over time. The notable achievement was not getting the count right on the first attempt. It was retaining the disagreement, accepting independent evidence and changing the artifact when the same kind of repair kept failing.