PR 27397 runtime verification, 2026-11-01
baseline github/main: 7195b699547b538e49f758e314dde5c8c5d907cf
verified head: 9550a8ffc189610129a9b661fe5e133dae923214
Linux; repository Playwright; /usr/bin/google-chrome; vp i in both worktrees.
One background dev server at a time; isolated home; T3CODE_DEV_SEED=0; T3CODE_BUNDLED_DEV=1.
Main: Pi gpt-6.1-sol visible button text 6.1 Sol; no selected thinking level or Default badge.
Fixed: same model visible button text 6.1 Sol Thinking; still no selected level or Default badge.
High/Low change labels; High survives completed-thread reload; active Pi Luna keeps Medium Default.

Real Pi session reached through the web composer/server adapter:
{"type":"thinking_level_change","thinkingLevel":"medium"}
{"type":"thinking_level_change","thinkingLevel":"low"}
{"model":"gpt-6.1-sol","thinkingLevel":"low","stopReason":"stop","text":"FINAL_PI_LOW_VERIFIED"}
{"type":"thinking_level_change","thinkingLevel":"high"}
{"model":"gpt-6.1-sol","thinkingLevel":"high","stopReason":"stop","text":"FINAL_PI_HIGH_VERIFIED."}

Real Codex session reached through the same web composer/server:
{"type":"turn_context","model":"gpt-6.1-sol","effort":"high"}
{"type":"task_complete","last_agent_message":"FINAL_CODEX_HIGH_VERIFIED"}

blacksmith-check run: vp run check:changed github/main
PASS: format, lint, both knip passes, seven affected typechecks, five modelTraits cases, existing ComposerModelEffort browser test, compiler coverage.
Two independent code reviews: no findings at 9550a8ff.

Environment limitations: Pi automatic extensions produced empty proxy responses initially; --no-extensions configured through client settings enabled real turns. Stock external Pi services, Electron/native mobile, remote/relay/tunnel unverified. Cold Pi version probe timed out; refreshed provider status recovered. VM browser/export slowdown exceeded ten minutes before freeing unused browser context; final turns completed within seconds.