Closes the loop on the innovation-engine itself: it reads every idea this engine has ever proposed, then uses local Gemma to semantically match each one against your current ~/bin/ tools and active systemd services to detect which ideas you actually shipped. You get a real conversion rate (ideas proposed vs. realized) and the single best still-dormant idea resurfaced — so the engine stops proposing into a void and starts learning what you build.
Create ~/bin/idea-realization-tracker (bash+python, ~80 lines). (1) Read the engine's idea history — the JSON outputs logged under ~/services/innovation-engine/ (idea log / DB); if none persists yet, add a 3-line append-to-~/services/innovation-engine/ideas.jsonl hook in the engine's output step. (2) Build the current capability inventory: `ls ~/bin/` + `systemctl --user list-units 'service:*'` descriptions. (3) For each past idea title+description, call local Gemma via ~/bin/gemma-task with a prompt: 'Does any tool in this list already do this? Answer REALIZED: