GenXis Model-l: the intelligence race.

One question, two racers: the bare model you pick, and the Model-l stack in nohands-cascade or answer-lane mode - same input, racing to the right text answer.
Give an expected answer and each lane is graded right or wrong beside its clock. One Gavel key opens and meters both; the stack lane is receipted.

Step 1 - Unlock with your Gavel key (same key as Compare)

The key stays in this browser tab and rides only over HTTPS. The stack lane bills your wallet (10¢ per call); the bare lane runs on our provider keys. No account? Get a Gavel key.

Step 2 - Model-l stack mode (the proposer stack)

nohands - the adaptive cascade: propose, escalate, verify inside the declared bounds.
answer - the receipted answer lane: one question in, one receipted answer out.

Step 3 - Bare racer (the model without the stack)

Bare model - full live list from the provider API

Step 4 - The question (fuzzy, up to 2MB) + the right answer

...or drop / choose a file (2MB max)

Expected answer (optional - leave empty to grade with your own eyes)

Trap library - ? push-button problems in ? groups, incl. GPQA-Diamond stoppers (held out, blind to the stack)

Bare model-

The bare model's answer lands here, with its clock and grade.

Model-l stack-

The stack's answer lands here, with its clock, grade, and receipt.

The input may be fuzzyHard tests, research problems, trash - up to 2MB. Model-l's job: fuzzy in, structured state out.

The output should not beThe stack lane returns structured state with receipts; the bare lane returns whatever it returns.

Same key, both surfacesOne Gavel key opens Compare and Model-l; the wallet is yours, the receipts are checkable without us.