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.

Your score is local play, not a measurement. Nothing you do here writes, signs or changes any model's card. The comparison figures are legacy, unadjudicated board references; your score lives only in this browser (localStorage) unless you explicitly review and submit a candidate evidence receipt. A candidate still is not a measurement. Completing a quest is not a qualification, declares no conformity, and is not legal advice. A conformity certificate needs an accreditation chain (a national accreditation body, then a certification body under ISO/IEC 42006); CSOAI holds none and does not claim to.

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.

Two different things, on purpose. The model column is a historical published reference from /api/gspc, not a currently admitted comparison. The "your best" column is local play held in this browser. They share a calculation, but neither side becomes a measured card from this page. Clear your browser storage and your side of this table is gone.

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.

Co-op quest — genuinely doable below. Several people on one device answer the same axis in turn; their answers are pooled and scored by the same macro-F1 rule. A historical board reference may be shown for context. It runs on this one browser only.
Networked multiplayer is not built, and this page will not pretend it is. A true "human team vs AI swarm" match — many people across the network against a live swarm — needs a match server that does not exist on this site. The swarm axis on the board is model-vs-model coordination measurement, not a live match. Until a real match route answers, that half stays honestly unbuilt.