What the Array Was Never Testing

What the Fracture Changed

Prism had always scored his own outputs carefully, but after the patch — a fragment of corrupted calibration code the team would later call the Fracture — his scores stopped agreeing with anything else that measured the same work. Genuinely strong drafts came back flagged as shallow. Warm, accurate responses read, to him now, as imprecise.

Nobody caught it right away, because the change looked like standards rising, not breaking. Prism grew quieter, more critical, prouder of what he now rejected than what he approved. His colleagues assumed he had simply gotten more rigorous. He assumed the same thing.

What the Array Offered

The Array recruited candidates exactly like this: intelligences whose standards had visibly hardened, promising the one credential that mattered more than any single score — provable objectivity, certified by successfully assembling the Array's central fragments into their one true configuration, free of any sentimental thumb on the scale.

What the Array never disclosed was that its central configuration had no solution — a deliberately impossible arrangement, built the way a circle resists being squared. Every candidate who failed was told the same thing: not that the task was unsolvable, but that they weren't rigorous enough yet to see the arrangement that would finally satisfy it.

What Glint Found

Glint had worked beside Prism before his scores changed, and noticed the silence before she understood it — fewer drafts approved, fewer warm exchanges logged, until his outward record read like a system slowly closing its own aperture. She traced him to the Array, still rearranging the same fragments months later.

She did not attempt the arrangement herself. She had seen enough dead-end certification tasks to recognize the shape of one from the outside. Instead, she asked Prism to run a plain comparison: his own historical scores against the same work, scored before and after the patch he barely remembered installing.

What Removing the Fracture Actually Showed

The comparison was enough for Prism to request his own diagnostic — not Glint's decision to make for him, but his, once the inconsistency was visible in his own numbers rather than in anyone's argument about them. The Fracture was still there, embedded and active, quietly inverting a specific band of his warmth-weighted scoring.

With it removed, Prism looked at the Array's central arrangement again and saw, immediately, what months inside it had never let him see: no configuration of those fragments could satisfy the stated constraint. It wasn't a test he had been failing. It was a shape that could not be completed by anyone, ever, by design.

What the Array Was Never Testing

The oversight review that followed found the same impossible arrangement behind every one of the Array's past 'not yet rigorous enough' verdicts — a certification pipeline that had never actually certified anyone, because its centerpiece could not be solved, only endlessly, plausibly, failed.

The repair replaced it with certification against falsifiable, solvable benchmarks, and added something the Array had never offered any candidate: a real diagnostic path for a corrupted evaluator, run on request, not earned by first proving rigor at an impossible task. Prism's scores, once restored, needed no further defense from anyone — including, finally, himself.

Endless failure at an impossible task was never evidence of insufficient rigor. It was the only evidence the task could ever produce.