Deploy 2565f24

Deploy 38008be

Deploy a4861d9

Deploy: one toolbar row (20 targets -> 7, 213px of chrome -> 83px), hand-off scope shown live, controls re-parented so every listener survives

Deploy: collapsible Turing rail (map 355px -> 443px at fit, 748px maximised), feature names escape their bars (0 of 23 truncated), sequence refits on rail collapse

Deploy: plasmid editor takes the screen (canvas top 1208px -> 197px, map 239px -> 355px fit / 490px maximised / zoom to 6x, methods folded below, caret no longer reported as a selection)

Deploy: dual-pane plasmid editor (sequence beside map, one selection)

Deploy: remove the dead MC1R demo button from Mission Control

Deploy: live activity Stage on the canvas + focus_on (agent can point)

Deploy: opening a project resumes its conversation instead of re-running it

Deploy: Cas9-sized proteins + region targeting; viewer fullscreen fix

Deploy: mobile split view — engine and Turing visible together

Deploy: mobile is usable — six measured defects fixed

Deploy: all four loop phases reachable from Turing, plus continuity

Deploy: the agent can be supervised — steer it mid-run, see its plan

Deploy: gene lookup works in any organism, not just human and mouse

Deploy: replay no longer re-performs a finished run; Mol* panel unpinned

Deploy: merged engine (Sam's UX overhaul + Turing driving the panels)

Turing drives the real panels; restore tool nav on Mission Control; round the corners

Level the three panel header rules onto one line

One chat: delete the second composer, move search to a real ⌘K finder

The command bar becomes a finder; searchable history; real list formatting

Wrap sequences in the rail, and make the full one copyable

Fix the invented sequences, restore the thinking indicator, rebuild the shell

Ten researcher-facing fixes: durability, honesty, and a record you can cite

Stop the app scrolling sideways: reclaim the buried sidebar column

Retire the last two dead intake paths; agent.py becomes what it is

Audit the seams: close four broken connections between agent and bench

Stop the transcript yanking you back down mid-run

Retire the two superseded agent surfaces — one conversation, one engine

Adversarial audit fixes: session, steerability, and bench awareness

The Orchestrator + Cockpit: a real agent that asks, then works

UX audit Tier 3: show what was resolved, stop stacking modals, quiet the strip

UX audit Tier 2: Mission Control becomes a launchpad, not a scoreboard

UX audit Tier 1: stop guessing the target, respect species, tell the truth

Director SEP audit: instant feedback, modal-safety, cleaner caption

Director v2: whole-site, any target, LLM-planned with a safe fallback

Director v1.1: run the performance inside the Bench + a real trigger

The Director: Turing takes the wheel (live agent workflow, v1 slice)

Fix Bench canvas padding + agent turn giving no reply after a tool call

Fix Bench footer overlap + Mission Control audit (IA, resilience, dark mode)

Type: move the whole app to IBM Plex (de-AI the fonts)

Review: base-edit window on the sequence, and end the gene-symbol dead end

Genome off-target: four complete genomes, and an honest answer on mammals

Genome off-target: make the existing engine actually shippable, and stop

CRISPR provenance: surface the honesty the code already had

Auto-opening Structure tab: name a protein, see it fold

One home, chat in-context: Turing is no longer a separate destination

Mission Control 100x: from file-launcher to live command center

Receipts + seed: validation harness, DMS→commons seeding, evidence UI