{
  "claim": "Three open System One releases in one week, three different answers to the same architectural question: where does the probability come from? Only one of them builds the scorer from scratch; the other two put a decision head on top of a general model. Every figure below is quoted from the release it belongs to.",
  "method": "Read from each project's own model card, dataset card, README and config on 2026-09-18, and re-checked on 2026-09-19 against the published artifacts themselves rather than the prose about them. Parameter counts and checkpoint sizes are the ones each project states; where a size and a count can be cross-checked against each other, the note says so. The calibration column was wrong in the first version: it read \"none\" for every row.",
  "captured": "2026-09-19",
  "note": "cua-s1-forms is the only one of the three whose parameter count and checkpoint size independently agree: 706,048 parameters at fp32 is 2,824,192 bytes, which is the 2.8 MB the card claims for state_dict plus config plus training history. That is a weak check, but it is the kind that catches a card written from memory rather than from the artifact — and reading that same sidecar is what corrected the calibration column. cua-s1-forms does publish an expected calibration error; it lives in metadata.best_validation, on a split where top-1 is 0.99941, which is where a near-zero ECE comes from. Bespoke's public suite, added the day after this article, is the first calibration measurement of Jev by anyone, and a competitor ran it.",
  "columns": [
    {
      "key": "project",
      "label": "release"
    },
    {
      "key": "base",
      "label": "base"
    },
    {
      "key": "newparams",
      "label": "new params",
      "align": "right"
    },
    {
      "key": "outspace",
      "label": "what the probability spreads over"
    },
    {
      "key": "objective",
      "label": "objective"
    },
    {
      "key": "calib",
      "label": "calibration reported"
    }
  ],
  "rows": [
    {
      "project": "cua-ai/cua-s1-forms",
      "base": "none — trained from scratch",
      "newparams": "706,048",
      "outspace": "one scalar per option; softmax over the live option count",
      "objective": "cross-entropy over the live option count",
      "calib": "ECE 0.000148 (22,054 val rows) — in cua-s1-forms.json, not the card"
    },
    {
      "project": "bespokelabs/Bespoke-Nimble-9B",
      "base": "Qwen3.5-9B (LoRA r=16)",
      "newparams": "adapter only, ~193 MB on disk",
      "outspace": "allowed answer tokens, scored directly",
      "objective": "hard reference labels from rule application; no RL, no distillation",
      "calib": "ECE + Brier on 13 public datasets, 3,880 rows, added 2026-09-19"
    },
    {
      "project": "browser-use/jev-ultrafast",
      "base": "calls the hosted Jev API",
      "newparams": "no model of its own",
      "outspace": "an index into the runtime's validated action list",
      "objective": "not a training project",
      "calib": "none — not a training project"
    },
    {
      "project": "TypeSafe Jev 1.13",
      "base": "not disclosed",
      "newparams": "not disclosed",
      "outspace": "up to 255 Choice options / 2-10 Score levels / 1 Noul",
      "objective": "not disclosed",
      "calib": "none first-party; third-party ECE + Brier since 2026-09-19"
    }
  ]
}