# Narration (timestamps relative to recording start, from the agent's action log) row-C-option-after 0:00 question card visible (thread "Q1 enter", running turn); 0:04 "Red" option clicked: single-question card submits immediately, card replaced by "Do you prefer red or blue? Red" + Thinking (screenshot at 0:06); ~0:30 server reports assistant reply "You picked Red." (via CLI thread show); 0:33 stop. Note: clicking an option bypasses Enter/Submit entirely. row-C-enter-after 0:00 question card visible (thread "Q2 click", running turn); 0:03 custom answer "blue" typed in composer; 0:05 Enter pressed; 0:05-0:07 card gone, timeline shows "Do you prefer red or blue? blue" + Thinking (screenshot 0:07); 0:20 server reports reply "You picked blue."; 0:23 stop. row-E-after 0:00 same thread after the answer, reply "You picked blue." visible, composer in normal mode ("Ask anything..."); 0:03 "follow-up" typed; 0:05 Enter pressed; 0:10 "follow-up" shown as a normal user message, new turn "Working for 4s"; no answer card; 0:15 stop. row-C-submit-after 0:00 question card visible (thread "Q4 submit", running turn); 0:03 "blue" typed; 0:05 Submit pill clicked; 0:06 card gone, "Do you prefer red or blue? blue" + Thinking (screenshot 0:06); 0:22 server reports "You picked blue."; 0:24 stop. row-C-enter-before 0:00 question card visible (thread "Q1 enter", running turn); 0:03 "blue" typed; 0:05 Enter pressed; 0:05-0:07 card gone, "Do you prefer red or blue? blue" + Thinking (screenshot 0:07); 0:20 server reports "You picked blue."; 0:23 stop. Same as AFTER. row-E-before 0:00 after the answer, "You picked blue." visible, composer in normal mode; 0:04 "follow-up" typed; 0:06 Enter pressed; 0:11 "follow-up" shown as a normal user message and the agent answers it as a new turn ("Worked for 5.1s"); 0:17 stop. Same as AFTER. row-C-submit-before 0:00 question card visible (thread "Q4 submit", running turn); 0:04 "blue" typed; 0:06 Submit clicked; 0:09 card gone and reply "You picked blue." already rendered (screenshot 0:09); 0:16 stop. Same as AFTER. row-A-after 0:00 question card visible (thread "Q3 noprovider", Claude, running turn); 0:04 Settings opened; 0:11 Providers; 0:25 Claude toggled off; 0:39 Back to thread: card still there, but composer fell back to GPT-6-Astra (Codex), so a provider is still available; 0:52 Settings again; 1:01 Providers, Codex toggled off (both now "Disabled"); 1:10 Back to thread: card still visible, composer footer shows "Open provider settings" (noProviderAvailable), Submit pill is DOM-disabled (checked ~1:25); 1:31 "blue" typed; 1:33 Enter pressed; 1:35 card gone, timeline "Do you prefer red or blue? blue" + Thinking, placeholder "Enable a provider in Settings to send a message"; ~1:51 server reports "You picked blue." and thread completed (the Claude session survived disabling the provider instance); 1:55 stop. row-D1-after 0:00 new demo-repo draft thread, Claude and Codex both disabled, footer "Open provider settings"; 0:05 "hello" typed; 0:07 Enter pressed: nothing sent, text stays (screenshot 0:09); ~0:20 send arrow clicked: DOM-disabled (aria-label "Environment disconnected"), nothing sent (screenshot 0:22); 0:30 stop. row-A-before 0:00 question card visible (thread "Q3 noprovider", Claude, running turn); 0:05 Settings opened; 0:07 Providers; 0:18 Claude toggled off; 0:21 Codex toggled off (both "Disabled"); 0:31 Back to thread: card still visible, footer "Open provider settings", Submit pill DOM-disabled; 0:44 "blue" typed; 0:46 Enter pressed: nothing happens, "blue" stays in the composer, card stays (screenshot 0:49); 1:10 server still reports "running, Waiting for 1 answer(s)" (CLI); 1:11 stop. row-B-before 0:00 same state as end of row-A-before (card visible, "blue" in composer, no provider); 0:05 hover Submit pill; 0:07 Submit clicked: nothing happens, button disabled (screenshot 0:11, card and text unchanged); 0:20 stop. row-D1-before 0:00 new demo-repo draft thread, both providers disabled, placeholder "Enable a provider in Settings to send a message"; 0:04 "hello" typed; 0:04 Enter pressed: nothing sent, text stays; 0:06 send arrow DOM-disabled (aria-label "Environment disconnected"); 0:07 send arrow clicked: nothing sent (screenshot 0:10); 0:16 stop. Same as AFTER. row-B-after 0:00 question card visible (thread "Q5 noprovider click", Claude on, Codex already off); 0:06 Settings; 0:08 Providers; 0:11 Claude toggled off (both "Disabled", screenshot 0:11); 0:20 Back to thread: card still visible, footer "Open provider settings"; 0:24 "blue" typed; 0:25 Submit pill checked: DOM-disabled; 0:34 hover Submit; 0:35 Submit clicked: nothing happens, card and "blue" unchanged (screenshot 0:39); 0:52 stop. Same as BEFORE: the click path with no provider is still blocked at head. row-D2-after 0:00 thread "D2 slow", Claude running the 40 x sleep 1 bash loop ("Running command"); 0:05 "what is 2+2?" typed; 0:06 Enter pressed; 0:08 message sits in a "Queued 1" tray above the composer with Steer/edit/remove; the turn keeps running, no parallel turn (screenshot 0:08); ~1:11 turn finished ("Worked for 48s"), queued message then sent as the next turn and answered "4" (screenshot 1:11); 1:23 stop. row-D2-before (recording started 1.6 s after AFTER's, same actions in lockstep) 0:00 same state; 0:03 typed; 0:05 Enter; 0:07 "Queued 1" tray with Steer, turn keeps running (screenshot 0:07); ~1:10 turn finished ("Worked for 45s"), queued message runs next, answer "2+2 is 4."; 1:22 stop. Same as AFTER. Notes: - Recordings are .mp4 (the extension the Browser panel returned), not .webm. - Stills: still-rowA-{before,after}-1440-light.png were taken at a 1440x900 CSS viewport but the screenshot tool downscales them to 1280x800 PNGs. The 390-wide dark stills are 780x1688 (2x). Both pairs show the identical state: question card open, "blue" typed, both providers disabled, before Enter is pressed. The extra still-rowA-after-390-dark-enter-submitted.png shows AFTER right after Enter (answer accepted, Thinking); the BEFORE counterpart is the row-A-before recording at 0:49 (nothing happens). Audit note (pass 2): row-C-enter-after (9.8s), row-C-enter-before (12.3s) and row-D1-before (11.4s) are variable-frame-rate recordings that dropped idle stretches. Their narrated times after the video's end (0:20, 0:23, 0:16) are wall-clock times from the capture agent's action log, not video positions. The narrated outcomes are visible in each video's last frame.