Tangle¶
Shopify's 3rd-generation reproducible ML/data workflow system, per mikhail-parakhin — the lineage he sketches runs Ether (Microsoft-era) → intermediate systems → Tangle. Design goals: clone an existing experiment cheaply, change a small piece, run it, then ship the exact same artifact to production in one click with no porting step.
Tangle is the substrate on which tangent (the auto-research loop) runs.
Why the generational framing matters¶
Parakhin's historical argument: reproducibility and the clone-and-ship primitive matter more than any specific orchestrator choice. Without them auto-research is impossible because you can't trust that a winning experiment will behave identically in production. Compare durable-observable-debuggable-agents and infrastructure-as-context — same underlying thesis for agents.