What Is the Composition Impossibility Theorem?
The Composition Impossibility Theorem states that no composition of independently advancing data systems can guarantee decision coherence: a decision that reads its context from multiple systems — each ingesting, deriving, and serving state on its own timeline — cannot be guaranteed one consistent view of reality, regardless of how fast or well-tuned each system is. Coherence lost at the seams between systems cannot be recovered by improving the systems themselves.
Updated
What is the Composition Impossibility Theorem?
The theorem is formalized in the Context Lake paper (arXiv:2601.17019) and is part of the vocabulary Tacnode uses to explain why the context gap is structural rather than operational. The informal statement: you cannot compose separate infrastructure into a coherent layer — the seams between the pieces become failure points, and a decision assembled across those seams can observe combinations of state that never coexisted.
The key primitive in the argument is atomic visibility: guaranteeing that a decision sees a consistent view across systems would require each system to hold its effects invisible pending external coordination — exposing a hook by which a cross-system snapshot could be constructed. Independent systems, by definition, do not expose this capability; each makes its own state visible on its own schedule. Distributed commit protocols provide a limited form of it for writes among participating databases, but the composed decision stack — caches, indexes, stream processors, serving stores — neither participates in such protocols nor could without giving up the availability and latency properties it was chosen for.
What the theorem implies
Three practical consequences follow:
- Tuning has a ceiling. Faster pipelines and shorter TTLs shrink each system's lag but never synchronize the systems; the incoherence between them remains available to any sufficiently concurrent or adversarial workload.
- Integration is not coherence. Connecting every system — more pipelines, more sync jobs — adds seams rather than removing them. As the compressed version puts it: correctness guarantees don't compose; you can connect every system and still get the wrong answer.
- Coherence must be native. If a decision requires one consistent view, the state it reads must live under one snapshot authority — the design conclusion embodied in the Context Lake, where facts, derived signals, and semantic evidence are served from one substrate under one internally coherent snapshot.
Why the Composition Impossibility Theorem matters
The theorem converts a recurring engineering argument into a settled one. Teams facing incoherent decisions typically respond with more tuning of the composed stack, and the theorem's contribution is showing that this path has a hard limit that no budget escapes — the failure is a property of composition itself. It thereby draws the honest boundary for both sides: architectures composed of independent systems can be excellent at everything except cross-system coherence at decision time, and systems that need that coherence need it by construction, not by optimization. The empirical face of the theorem is the incident pattern described in context under concurrency: every component healthy, the composite wrong.
FAQ
Related terms
Decision coherence is the property that concurrent automated decisions act on the same version of shared state, instead of contradicting each other.
A context gap is the difference between the state of the world an automated decision acts on and the actual state at the moment the decision commits.
A validity window is the interval within which a decision's context remains an accurate basis for action — typically 10ms to 1s for automated decisions.
The divergence problem: services each maintain their own pipeline over the same events, advancing at different paces, so their views of shared state disagree.
A Context Lake is real-time, multi-modal data infrastructure that gives AI agents and decision systems fresh, internally coherent context at decision time.