OpenIE · The Stack · An Overview
Three names.
One substrate.
A teaching map of three foundational theories — information theory, game theory, mechanism design — and the eight working stacks that compose across them.
⟨ I · NAMING · II · ARRANGEMENT · III · OPERATION · IV · INSTRUMENTS ⟩
The map of the family · four sections, one substrate
I · The naming
Three names, side by side.
Information theory. Claude Shannon’s 1948 paper A Mathematical Theory of Communication showed that any signal you can distinguish from noise has a minimum bit-rate at which it can be transmitted reliably. Thirteen years later, Rolf Landauer at IBM proved that erasing one bit dissipates a minimum quantity of heat — kBT · ln 2. The thermodynamic floor was confirmed experimentally in 2012.
Game theory. John von Neumann and Oskar Morgenstern’s Theory of Games and Economic Behavior (1944), sharpened six years later by John Nash’s equilibrium concept. Game theory describes the stable configurations rational actors fall into when they face each other.
Mechanism design. Leonid Hurwicz, Eric Maskin, and Roger Myerson received the Nobel for it in 2007. Mechanism design is the inverse of game theory: instead of predicting what strategic actors will do, it asks what rules will make their rational play produce the outcome you wanted. Auction theory, matching markets, voting systems.
Listed this way the three appear as peers. They are not. The next figure shows how they actually sit relative to each other.
II · The arrangement
Three layers, one substrate.
The three theories sit at different altitudes.
Information theory is the substrate. Every bit stored, every signal transmitted, every erasure performed pays a cost in joules bounded by Shannon and Landauer. Anything formalizable in computing lives inside those bounds.
Game theory is the structure on the substrate. Multiple agents with preferences, information, and available actions fall toward equilibrium. Pushed far enough, equilibria become free-energy minima — the math of game theory begins to share a language with statistical mechanics.
Mechanism design is the routing. Given the substrate and its structure, mechanism design specifies the trajectory by which an input is directed across the slope so that the agent’s rational fall arrives at the intended outcome.
Position in the well is a measurable quantity. Where on the slope a configuration sits tells you what it pays in joules per useful bit — deeper costs less.
III · The operationalization
The substrate has terrain.
Make the gravity well concrete and it develops topography. Different kinds of work occupy different regions. Depth equals joules per decision. Deeper costs less.
Wells at the bottom — closed-form work: arithmetic, parsing, structured lookup, deterministic control. Output is exactly derivable from input. Verification is a comparison.
Basins in the middle — pattern-bound work: retrieval against structure, classification, small-model inference. Output is constrained but not fully determined. Verification carries a citation.
The plateau on top — open-ended synthesis: generation without named grounding. Cost scales with distance to a grounding point.
From E(x) = Σ θ(p) · μ(p, H): each cell has a calculable per-input energy — sum the Landauer floors of its primitives, scale by hardware impedance. And the routing itself is a lookup, which is itself a well. The router is the table.
IV · The working instruments
Eight stacks. One substrate.
The OpenIE family is eight periodic stacks. Each describes one kind of system or object using a finite set of orthogonal axes.
Knowledge is the object that travels. Compute is the substrate floor. Agent is the actor. Between them sit five working instruments — data (how records lie), state (how things run), memory (how it retains), communication (how it carries), synthesis (how it is routed).
The eight are periodic: a primitive carries the same coordinates wherever it appears. The thermodynamic class T inherits from compute. The verifiability V travels with answers across synthesis, data, memory, and communication. The encoding regime R crosses synthesis, data, memory. The temporal axis τ crosses data, state, memory, communication. Coordinates compose across the family.
What follows is the directory. Each card opens a sibling site.
knowledge
a claim with seven axes
Strip an axis, it stops being knowledge. The object every other stack handles.
synthesis
verb, not noun
The cost surface and the function that places each input on it. Zones Z₁, Z₂, Z₃.
data
how records lie
Storage and retrieval as a structured design space. What answers come for free; what costs Landauer.
state
how things run
The shape of orchestration. Plans, machines, transitions. Structure is the controller.
memory
how it retains
Retention as a coordinate. Register, hippocampus, DNA, songline — same map.
communication
how it carries
Eleven quantities each bounded by a named theorem. TCP and a deep-space DTN bundle differ by coordinate, not by kind.
verify
joules per verified output
Seven axes for a verified artifact. The cheapest sufficient solver wins. E(x) ≥ θ(D) · μ(S, V).
agent
who occupies the seat
A computer permits a finite set of acts. The agent is whoever selects from that set. The seat is older than the occupant.
compute
the floor · joules = bits
The permitted-act set. The thermodynamic floor every other stack inherits from. 258 primitives, 33 families.
Closing
Standing on prior work.
Each axis on this map has its own discipline behind it. Shannon 1948, Landauer 1961, Petri nets 1962, the process calculi of Milner and Hoare in the late 1970s, bitemporal databases (Snodgrass and Ahn, 1985), statecharts (Harel, 1987), three decades of provenance-aware knowledge representation.
What this site adds is the reading: the axes on one map, the coordinates shared across stacks.
A teaching artifact. Take it where it’s useful.