# Driver for #25715 (scratch, synthetic data only)

Base: origin/main dbfbe3993f. Apply race-subcommand.patch (adds a scratch `t3 thread race`
subcommand: t3_thread_send, sleep --delay ms, t3_thread_interrupt over the server's MCP endpoint).

    T3CODE_DEV_SEED=0 vp run dev            # fresh .t3 in the worktree, Replay (dev) provider
    cd apps/server; B=$PWD/../../.t3; W=<a git repo>
    node src/bin.ts project add $W --title demo --base-dir $B
    node src/bin.ts thread start $W Start --provider devReplay --model showcase --wait --json --base-dir $B
    node src/bin.ts thread race <threadId> "go" --delay 0 --base-dir $B
    node src/bin.ts thread send <threadId> "followup" --wait --json --base-dir $B

Observed (3 of 3 tries at delay 0): the stopped run is marked interrupted with
"Run interrupted before provider start", its provider turn then starts and runs ~5 s to
completion (reasoning, tool calls, file change, command, final message), and the follow-up
fails: "The provider could not start this turn: the previous turn is still running."
Persisted afterwards: stopped run interrupted, its provider turn completed, follow-up run failed.