7 decision scenarios × 3 models × 2 formats (opencode, $0)
accuracy = pass / (pass+fail); errors (timeout/unparsable) excluded

[toggle already on]  zd-gold ['none']  sh-gold ['none']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> none
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> none
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> none
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> none
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> none
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> none

[toggle off -> act]  zd-gold ['click [02]']  sh-gold ['click [0-11]']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> click [02]
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> click [0-11]
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> click [02]
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> click [0-11]
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> click [02]
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> click [0-11]

[form ready -> submit]  zd-gold ['click [03]']  sh-gold ['click [0-15]']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> click [03]
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> click [0-15]
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> click [03]
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> click [0-15]
    zerodom   FAIL  nemotron-3-ultra-550b-a55b     -> click
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> click [0-15]

[form incomplete -> fill]  zd-gold ['fill [02]']  sh-gold ['fill [0-6]']
    zerodom   FAIL  nemotron-3.5-lightning-30b-a3b -> click [03]
    stagehand FAIL  nemotron-3.5-lightning-30b-a3b -> click [0-14]
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> fill [02]
    stagehand FAIL  nemotron-3.5-super-vl-120b-a12b -> fill [06]
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> fill [02]
    stagehand FAIL  nemotron-3-ultra-550b-a55b     -> click [0-14]

[expand collapsed panel]  zd-gold ['click [03]']  sh-gold ['click [0-13]']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> click [03]
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> click [0-13]
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> click [03]
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> click [0-13]
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> click [03]
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> click [0-13]

[radio already selected]  zd-gold ['none']  sh-gold ['none']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> none
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> none
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> none
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> none
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> none
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> none

[switch active tab]  zd-gold ['click [02]']  sh-gold ['click [0-10]']
    zerodom   PASS  nemotron-3.5-lightning-30b-a3b -> click [02]
    stagehand PASS  nemotron-3.5-lightning-30b-a3b -> click [0-10]
    zerodom   PASS  nemotron-3.5-super-vl-120b-a12b -> click [02]
    stagehand PASS  nemotron-3.5-super-vl-120b-a12b -> click [0-10]
    zerodom   PASS  nemotron-3-ultra-550b-a55b     -> click [02]
    stagehand PASS  nemotron-3-ultra-550b-a55b     -> click [0-10]

==================================================================
DECISION ACCURACY — ZeroDOM format vs Stagehand observe() tree
  zerodom    19/21  (90%)
  stagehand  18/21  (86%)

[exited with code 0]
