# What else looked different between BEFORE (8d3b061121) and AFTER (b9d6319d60)

Nothing outside the pending-answer submit path differed on screen.

- Rows C (Enter and Submit during a running turn), D1 (no provider, normal message), D2 (follow-up during a running turn goes to the "Queued" tray with Steer) and E (follow-up after the answer is a normal turn) looked and behaved identically.
- The settings Providers page, the provider toggles, the question card layout, the mobile 390px layout and dark mode looked identical.
- Option descriptions on the question card differ between runs ("Pick red" vs "Choose red"). That is model output, not a code change; it varies between threads on the same side.

Observations that apply to both sides (not caused by the PR):
- With both providers disabled, the pending Submit pill is DOM-disabled on BOTH sides (row B). The PR's fix reaches the Enter path only. The click path for "no provider" is unchanged, as matrix row B predicted.
- Disabling the Claude provider instance in Settings did NOT end the live Claude session or remove the question card on either side. On AFTER the answer went through and Claude replied "You picked blue." while both providers were still disabled (row A).
- Disabling only Claude is not enough to reach noProviderAvailable: the composer falls back to Codex (GPT-6-Astra). Both Claude and Codex had to be turned off.
- Clicking an option on a single-question card submits immediately on both sides, bypassing both Enter and Submit (row-C-option-after shows this).
- With no provider, the disabled send button in a new draft thread has the aria-label "Environment disconnected" on both sides, which is misleading: the environment is connected and only providers are off.
- The "hello" text from row D1 persists as a draft thread at the top of the sidebar on both sides.