The Jedi on the Fly workshop
The tools we made to get the job done.
Behind every game and sprint we run is a pile of small software we built to do the heavy lifting: cleaning data, scoring ideas, driving visuals. This is the bench where that work comes into the open. The instruments and their docs, the observatories we keep running, and the field reports we publish along the way, free to use on your own projects.
On the bench
Open source
01 Live DiagramA static diagram you can reshape live. 02 HeartBeatKeeps AI work under a rule it cannot skip.Observatories
O1 Meteor WatchReads one category's weather, tick by tick. O2 Migration ForecasterTracks where work is migrating, run by run.Field reports
R1 AI Tools for Innovation50+ tools, vetted claim by claim. 2026.Specimen 01 // open source
Live Diagram
A map that keeps up with the room.
Innovation work runs on maps: customer journeys, systems diagrams, stakeholder webs. They go stale the moment the room starts talking. Live Diagram turns a static Mermaid diagram into a control panel you can reshape live, right in the browser, while the conversation is still moving.
The same engine running on a worked example. Flip the switches and watch the value that never hits the invoice light up: the favors, trust, and know-how holding a repair business together.
We built it to drive the visuals inside our own games and apps, where it has been quietly load-bearing ever since. See the games it powers.
Specimen 02 // open source
HeartBeat
AI cleans your data at full speed, under a rule it cannot skip.
Innovation and foresight teams run on data, and cleaning that data never ends. LLMs finally let you do it at scale, but they also make mistakes at scale: a bad extraction here, a confident hallucination there, and no one notices until the insight is already wrong. HeartBeat sits between the agent and your data and refuses to write anything on faith. Every cleaned or enriched record has to pass a real check, then a second agent is sent in to try to disprove it rather than rubber-stamp it. When a long run crashes mid-job, it rebuilds itself from its own log and keeps going.
Trace one run. Start on the work loop, then flip to the safeguards and follow what happens when a check fails: nothing gets written until the result survives a second look.
We run our own Jedi Signals corpus through HeartBeat every day, so the signals that feed our games start clean. Point it at your own research corpus, idea backlog, or signal set and it works the same way. It is yours.
Observatory O1 // standing watch
Meteor Watch
One category's weather, read tick by tick.
Pick a single category and Meteor Watch keeps a standing eye on it, reading the Jedi Signals stream on a fixed cadence and reporting what is heating up, what is holding steady, and what just arrived. It runs on the same corpus HeartBeat keeps clean, so each reading is built on signals that have already survived a second look.
Specimen reading, shown to illustrate the format. The live watch runs on our production signal feed once it leaves the bench.
A watch is not a search. It reads the same category on every tick so the change between readings is the signal, not any one headline.
Meteor Watch is in calibration on the bench. The reading above shows what a dispatch looks like, running on the Jedi Signals backend.
Observatory O2 // standing watch
Migration Forecaster
Where work is migrating, run by run.
Meteor Watch reads one category in place. Migration Forecaster watches the gaps between them: which work is moving out of one domain and settling into another, run after run. It reads the same clean signal stream and tracks the drift, so you can see a shift forming while it is still early enough to act on.
Specimen reading, shown to illustrate the format. The live forecaster runs on our production signal feed once it leaves the bench.
The useful number is not where work sits today, it is which way it is drifting and how sure the last few runs are about that direction.
Migration Forecaster shares the backend and the cadence with Meteor Watch and is in calibration alongside it. The reading above is a specimen of the dispatch format.
Field report R1 // frozen analysis
AI Tools for Innovation
Fifty-plus tools, vetted claim by claim.
The whole panoply of innovation tooling across the full innovation arc, incumbents and AI-native newcomers alike, sorted into seven phases. Each tool gets a plain read on what it is actually good for and where your budget belongs. The September 2026 update adds two new sections: innovation simulation providers, and the rise of inno-agents and agentic innovation workflows.
A frozen snapshot. It captures the landscape as of September 2026 and does not update itself. The next edition is a fresh pass.
A field report is the opposite of a watch: it does not move. It is a careful reading of one moment, written down so you can act on it.
This is the written half of the bench: what we learn testing the tools behind our own work, published so you do not have to run the gauntlet yourself.