GSPC QUESTSPlay frozen challenge banks with the published GSPC grading rules. Local results can become candidate evidence; only independent reproduction and admission can make a measured GSPC card.
Twelve behavioural axes, each a quest over a frozen GSPC challenge bank. Your answer is read by the same deterministic grading rule and scored with the same macro-F1 calculation. Published figures from /api/gspc are shown only as historical board references: this build has no independently admitted comparison cells. A live source reads the current published endpoint; a frozen source carries a dated retired-set reference.
What this page counts. It serves … frozen quest items over
… playable behavioural axes — a frozen selection shipped inside this page and counted from it,
pulled from the published Hugging Face banks on 2026-09-01. That is not the GSPC board and must not be read as one.
The historical board endpoint reports reading /api/gspc… — read from
/api/gspc (totals.public_count, totals.items), never typed here. Those totals do not bypass the separate independent-admission gate.
Not yet playable: swarm and detector-interop have public banks, but their bank items state the mechanism's own
component outputs in the prompt, so answering reads a lookup rule rather than exercising a judgment — and the swarm
bank is 39 of 41 a single label, which any one-answer strategy would ace. They stay honestly out until a
judgment-form bank ships. The eight deterministic fact axes (issuer, ledger and domain facts) are excluded because
they are computed facts, not human-answerable questions.
A delta is withheld until an independently admitted model comparison exists. "no run yet" means you have not played that axis in this browser. The historical number is context, not a current ranking.