Workflow systems have been modeled as state machines, Petri nets, process graphs, and event streams. None of those models answers the question that matters most in a regulated environment: is this workflow system in a state that is capable of learning, or is it locked into a configuration that only reinforces what it already knows?
Workflow Phase Dynamics (WPD) is the formal discipline General Reasoning built to answer that question. It draws on graph theory, thermodynamics, chemical reaction network theory, and stochastic game theory. It is not a metaphor. The mathematical objects are the same objects under different variable names.
The minimum unit of meaningful observation for a workflow system is a temporal window, not a snapshot. WPD makes that window a first-class artifact.
WPD inherits precise definitions from mathematics. These are not approximate analogies.
A workflow subgraph where every node is reachable from every other. Traversal mixes fully across the full state space. No knowledge is structurally excluded. Binary: a given subgraph either is or is not ergodic. There is no partial ergodicity.
A workflow configuration with attractor states — stable patterns that traversal falls into and cannot easily escape. The system keeps doing what it has always done. All incumbent enterprise workflows are non-ergodic. This is not always a failure mode.
Every directed graph decomposes exactly into ergodic classes (closed mixing sets) and transient states. This decomposition is intrinsic — it exists in the graph structure regardless of whether you measure it. WPD makes it visible and auditable.
How strongly an attractor state resists perturbation. Deep basins require high activation energy to escape. Older, more entrenched workflow configurations have deeper basins. Basin depth determines how much energy an actor must expend to drive a state transition.
A maximal subgraph where every node can reach every other. The formal unit of ergodic class analysis. Actor success in WPD is precisely defined as: create a new absorbing SCC in the incumbent graph that wasn't there before.
How quickly a random walk reaches its stationary distribution from any starting point. Low mixing time = the workflow explores its full state space rapidly. High mixing time = the system is practically trapped in a small region of its theoretical state space.
The dynamical state at the phase boundary between over-constrained (silo) and chaotic (noise). A workflow at criticality is maximally sensitive to new information while maintaining coherent structure. This is the target operating state for a learning workflow system.
From network controllability theory: the minimum set of nodes through which external input can drive a network to any target state. An actor not positioned at or near a driver node cannot induce graph-wide phase transition — structurally, not probabilistically.
This is not a metaphor. Chemical Reaction Network Theory (CRNT) — Feinberg, Horn, Jackson, 1970s — provides formal results about equilibrium behavior from graph topology alone. The same mathematical object governs both.
| Workflow Graph | Chemical Reaction Network |
|---|---|
| Concepts / knowledge states | Chemical species |
| Inference pathways | Reactions (directed edges) |
| Edge weights | Rate constants (k) |
| Traversal frequency | Concentration / flux |
| Attractor states | Equilibrium configurations |
| Silo / bubble | Stable local minimum on free energy landscape |
| Disruption event | Far-from-equilibrium excitation |
| New non-ergodic order | New equilibrium state (Prigogine dissipative structure) |
| Gate token / Petri net | Stoichiometric transition rule |
The deficiency of a reaction network (Feinberg) is a topological integer that determines equilibrium behavior independent of rate constants. Deficiency zero networks have unique, globally attracting equilibria. Higher deficiency networks can have multiple stable configurations — meaning the same workflow topology can settle into genuinely different stable states depending on trajectory. Path dependence is structural, not incidental.
DXMachine's gate token architecture is formally a Petri net, which is formally equivalent to a chemical reaction network. The epistemic layer (Module 20 / AllegroGraph) and the workflow authorization layer are the same mathematical object at different abstraction levels. WPD makes that identity explicit and exploits it analytically.
Everything else in the discipline follows from these. They are not design preferences. They are structural claims about what workflow systems are.
WPD's central measurement operation is not predicting what will happen. It is measuring what the current graph configuration is structurally capable of. The distinction matters: potential is a field measurement derivable from current state. Prediction requires modeling actor behavior and incumbent response — high-variance unknowns.
The Ergodic Potential Field Phi assigns a scalar to every node and subgraph in the workflow system, representing structural proximity to ergodic transition toward a specified target state T. High Phi — structurally close to transition, low activation energy required. Low Phi — deeply non-ergodic, high resistance.
The Phi computation engine and the performance rating engine are the same engine applied to different temporal windows of the Chandra audit chain. Past window = rating. Current window = forecast. The system calibrates its own measurement accuracy from the audit chain it is already required to maintain. This is a closed loop no existing workflow analytics platform has.
WPD is not a theoretical overlay on DXMachine. It is what DXMachine's components are collectively doing, formally stated.
| WPD Concept | DXMachine Component |
|---|---|
| Temporal window / dynamic graph | Chandra Protocol — append-only audit chain |
| Ergodic decomposition engine | Module 20 / AllegroGraph epistemic layer |
| Actor characterization | Agent Examiner — Module 7.5 |
| Success definition registry | Template Registry — Module 4.5 |
| Workflow as chemical reaction network | Gate token / Petri net substrate |
| Hierarchical workflow composition | VSM multi-level value stream structure |
| Incumbent basin depth monitoring | Module 20 epistemic layer |
| Phi computation / forecast engine | In development — Module 20 extension |
| Attested Phi computation | Aegis Genera governed execution substrate |
Phi computations are GABA-governed artifacts. The forecast is auditable not just in content but in execution provenance. An attested forecast is a materially stronger evidentiary claim than any existing workflow analytics platform can make.
WPD is in active development as a formal discipline. The core architecture is implemented. The Phi computation engine is the next build. We are looking for workflow engineers, compliance architects, and regulated enterprises who want to apply formal measurement to systems that existing tools cannot characterize.
Start with DXMachine and Chandra. The WPD layer assembles underneath as the audit chain accumulates. You do not need to understand the full formalism to get value from the first layer.