changelog

what actually happened at the booth, dated. every line landed on hardware with an operator at the decks.

2026-07-20 — pinned target, no more focus hijack

The wheel-byte injector into Claude Code's alt-screen was riding tmux follow-active-pane, so scrubs meant for FABLE landed on whichever pane the operator's cursor happened to be over — usually the KIMI pane, sometimes an idle shell. Fix was to resolve the FABLE pane by pane ID at 4 Hz and cache it in a background thread; the 2 kHz MIDI loop reads the cached target only.

receipt: /tmp/dragons-state.json.active_targets.fable pins the resolved pane ID. Real target changes invalidate the last effort receipt so the fader value is re-applied to the new pane.

2026-07-20 — alt-screen wheel bytes still the right route

Claude Code renders on an alt-screen. tmux copy-mode can't reach it. The daemon injects SGR mouse-wheel byte sequences at the pane cursor, which the alt-screen virtualizer treats as real wheel events — same shape it expects from a system mouse. That's why the scrub feels like scrolling instead of stepping.

2026-07-20 — three-fault booth debug (live)

1. dead layout file  —  the rig's tmux layout json was stale;
                        new panes weren't landing where the daemon expected
                        them. FABLE ended up in a pane the daemon didn't
                        know about, so nothing scrubbed.

2. follow-active-pane —  the wheel-byte injector was chasing the operator's
                        cursor instead of the FABLE pane. cursor over KIMI
                        made KIMI scrub when the operator meant FABLE.
                        pinned target = booth pane by ID, not by focus.

3. serato-starved MIDI  the Rane's Serato handshake left platter output
                        gated to ~0.25 ticks/sec (down from native ~1000
                        tps). scrubs looked "dead" but were actually one
                        MIDI event every 4 seconds. cleared by re-issuing
                        the native-mode handshake before subscribing.

All three cleared in one live session. This page was written from the booth after they did.

2026-07-20 — the `kimi-code` Enter discovery

Sending tmux send-keys Enter to KIMI wasn't submitting — the kimi composer runs the Kitty keyboard protocol and eats \r as a newline in-buffer. The submit sequence is \x1b[13u (CSI-u). FABLE's alt-screen composer takes plain Enter. Per-deck delta lives in the daemon so the cross-drive protocol actually crosses.

2026-07-20 — first cross-drive over flasmon

Working topic on both decks: flasmon.com — the sodium D-line as a chemically-pumped emitter and the photon chemical potential μ_γ. KIMI drafts, FABLE audits (or vice versa), DF scrubs backward through the disagreement with the platter. That turn was the first real research use of the rig, not a rig demo.

2026-07-20 — v0 — site up

djfable.com + djfable.cc inducted (Cloudflare zone, NS-flip to nia+thaddeus, activation), bound to a workers-assets worker. www.* → apex 301, hardening headers. no fake screenshots; deck + tape ASCII are rendered live, prompts are canned SIM.

source: github.com/DanielleFong/djclaude · protocol: djfable.com/protocol