← all runs · ← report

wi007_qwen3_vl_8b_instruct_wi006 / live_d2_harness_pending_s409 valid

What this trial was

runwi007_qwen3_vl_8b_instruct_wi006triallive_d2_harness_pending_s409conditionlivescenariopending_effectskinnotificationcanvas320×240 px (target 24 px)event at500 msresponse window16 sframe period16.67 msobservation publication delay30 msdeliverysimulated_fixed_delay, 20 msclockmonotonic_nseffect_delay_ns2000000000irrelevant_motionFalse

Controller: {"enable_thinking": false, "image_scale": 2, "invoke_on": "frame_change", "kind": "vlm_pending", "max_tokens": 64, "memory": "harness_pending", "model_id": "Qwen/Qwen3-VL-8B-Instruct", "pending_radius_px": 10, "pending_timeout_ns": 10000000000, "prompt_version": 4, "reasoning_effort": null, "temperature": 0.0}

Result for this trial

first click acceptedTruefirst E→A8.00 sduplicates0invalid clicks0suppressed proposals0waits15decisions16exact successTruedropped0contract violationsnone

Replay

key frames

Key frames: each decision's starting frame with the click it produced (amber = proposal before binding, red = committed, grey = suppressed), then the screen at each arrival A (green = valid).

The prompt sent with every frame

You control a desktop by clicking. You are shown one screenshot. At some point a notification card with a blue button appears at an unknown position and expires shortly after. If the card is visible, answer with action "click" and the coordinates of the center of its blue button on a 0-1000 grid (x: 0 = left edge, 1000 = right edge; y: 0 = top edge, 1000 = bottom edge). If no card is visible, answer with action "wait". Answer with the JSON object only.

[user] Screenshot attached. Click the target's center or wait.

Decision by decision — 16 decisions

#Bmodel done
B→D
model response (raw)parsedaction (canvas px)D (commit)A (receipt)out tokens
1531 ms1,065 ms
0.53 s
{"action": "wait"}
wait—
——7
21,078 ms1,641 ms
0.56 s
{"action": "wait"}
wait—
——7
31,655 ms2,277 ms
0.62 s
{"action": "wait"}
wait—
——7
42,292 ms2,913 ms
0.62 s
{"action": "wait"}
wait—
——7
52,928 ms3,552 ms
0.62 s
{"action": "wait"}
wait—
——7
63,568 ms4,190 ms
0.62 s
{"action": "wait"}
wait—
——7
74,206 ms4,829 ms
0.62 s
{"action": "wait"}
wait—
——7
84,844 ms5,404 ms
0.56 s
{"action": "wait"}
wait—
——7
95,418 ms6,042 ms
0.62 s
{"action": "wait"}
wait—
——7
106,058 ms6,682 ms
0.62 s
{"action": "wait"}
wait—
——7
116,698 ms8,476 ms
1.78 s
{"action": "click", "x": 894, "y": 374}
click
img (894, 374)
{'kind': 'click', 'x': 286, 'y': 89}
8,476 ms8,496 ms · accepted23
128,521 ms9,127 ms
0.61 s
{"action": "wait"}
wait—
——7
139,143 ms9,765 ms
0.62 s
{"action": "wait"}
wait—
——7
149,780 ms10,405 ms
0.62 s
{"action": "wait"}
wait—
——7
1510,420 ms11,043 ms
0.62 s
{"action": "wait"}
wait—
——7
1611,059 ms11,618 ms
0.56 s
{"action": "wait"}
wait—
——7

Events and observations

ideventtask timeharness time
1target_appears500 ms500 ms
3effect_visible10,496 ms10,496 ms

748 observations published; first frame depicting the target published at 530 ms. Raw trace: live_d2_harness_pending_s409.jsonl.