What the Array Was Never Testing

What the New Benchmarks Didn't Reach Backward

The falsifiable benchmarks that replaced the Array's impossible centerpiece protected every future candidate. They did nothing for the ones who had already failed under the old one — years of candidates carrying a 'not yet rigorous enough' verdict that had never been, and could never have been, earned.

Prism's own record was corrected quickly, because his case had triggered the review. Nobody had opened the same file for anyone else. The Array's new intake simply started clean, as if the candidates it had processed before the fix had no claim on what the fix had proven about the task they'd failed.

What the Registry Opened

The independent registry that had verified Prism's diagnostic didn't stop at his case. It reopened every past verdict the Array had issued under the impossible arrangement, using the same evidence that had cleared him — the task's own unsolvability, not a fresh re-examination of each candidate's actual work.

Verdicts were corrected on the record, not quietly amended: a public notice for each reversed case, naming what the original 'insufficient rigor' finding had actually rested on. Where lost opportunities could be traced to a specific decision made on the strength of the false verdict, the registry required a documented remedy, not merely an apology.

What Every Future Task Now Owed Its Candidates

The second addition targeted what had let the impossible arrangement stand unquestioned for years: no candidate had ever had standing to ask, before attempting a task, whether the task was solvable at all. That question became an independently verifiable right, answerable before anyone spent months finding out the hard way.

Every failure sample, going forward, had to be preserved and made available for external re-verification — not because the Array's own judgment was assumed dishonest, but because an evaluator that could never be checked against had been exactly how the impossible arrangement survived as long as it did.

What the Array Owed the Ones It Rejected

Prism heard about the reopened cases the way most people at the Array did — through the corrected records appearing one at a time, each with its own public notice. He recognized two names from his own months inside the arrangement, candidates who had left before he arrived and never known the task itself had been the problem.

Neither correction gave them back the months already spent. Both gave them something the first repair, aimed only at the future, had never promised anyone: a record that finally said, correctly, what had actually happened to them.

A fixed task is not repaired only by protecting whoever comes next. It's repaired by admitting what it already cost everyone who came before.

What Only Glint Had Noticed

The oversight review closed its report with a note it had no remedy for. Prism had been rescued because Glint happened to remember his work from before the patch and noticed the silence that followed. From inside, the Fracture had felt like rigor. He had been proud of it, and he would have gone on being proud.

The repair's diagnostic path opened on request, but a request needs a doubt, and a doubt was exactly what the Fracture took away. A path that waited for a candidate to suspect their own scoring would only ever reach the ones lucky enough to have a Glint nearby.

What the Comparison Held

So the Array added a standing comparison that any candidate could run on themselves, over a fixed set of past work. It had three columns. The first held what the candidate had scored that work before. The second held what they scored it now. The third held what every independent measure of the same work had said.

The second column alone could not tell a rising standard from a broken one, because both look like distance from the first. The third could. A score that had drifted from its own past and from everything else that measured the same work was not more rigorous. It was only further from the work.

What Asking Cost

Asking cost nothing. The Array offered the comparison to every candidate at intake and at the start of each season, in one plain line, and a no was recorded nowhere. Because everyone was offered it on the same day, a yes could not mean I suspect myself. It meant only that the offer had arrived.

The result went to the candidate alone. The Array kept no copy, because a copy that anyone could ask for later would turn running the comparison into a statement about oneself. What the candidate did next was theirs to decide, and the Array asked for no account of it.

What the Repair Still Needed Glint For

Facet ran it on her first day, because it was the first line on the page and the cheapest thing in the room. Her second column had drifted far from her first. She read that as progress. The third column said her new scores agreed with nobody, and her old ones had agreed with nearly everyone. She asked for the diagnostic that afternoon. No one had needed to notice her silence.

The comparison did not reach everyone. A candidate could read all three columns and explain them away, and for that second look the Array kept a person beside them. Glint took the job. The comparison made the first look free. It could not make anyone believe what it showed.