Context Graphs Are Judgement Graphs

Part of the Judgement Layer series on decision infrastructure.

For twenty-five years I’ve worked at the intersection of enterprise systems and supply chain operations for companies like Syngenta, Pfizer, AstraZeneca, and BASF. The pattern I kept encountering was the same structural absence: organizations capture what happened, but never systematically capture why it was allowed to happen. The reasoning behind decisions: the tradeoffs weighed, the constraints bent, the bets placed. It evaporates when the meeting ends or the person leaves.

When I started building ChainAlign in 2025, the architecture was shaped by that accumulated experience. And now the broader industry conversation is arriving at similar territory.

Jaya Gupta and Ashu Garg at Foundation Capital recently identified something important: the next trillion-dollar enterprise platforms won’t be built by adding AI to existing systems of record. They’ll be built by capturing what enterprises have never systematically stored, the reasoning behind decisions. They call this layer the context graph.

The context graph framing clarifies an important shift: decisions are no longer isolated events. They live inside structured memory.

For some time, I’ve been exploring a closely related idea under a different label: the judgement graph. The emphasis is slightly different. If the context graph captures the relationships around a decision (who decided, what data was used, what reasoning was applied) the judgement graph focuses on what happens next.

When those decisions are evaluated against outcomes over time, something powerful emerges. Not just a trace of reasoning, but a calibration signal. Which assumptions held? Which reasoning patterns consistently produced strong results? Which decision-makers were directionally right even when outcomes were noisy?

In that sense, judgement isn’t separate from context. It’s what compounds when context is revisited against reality. The economic value doesn’t come merely from remembering why something was decided, but from learning which forms of reasoning improve with feedback.

That compounding layer is what I mean by a judgement graph. The key shift is the move from stored context to calibrated judgement.

What “Context” Actually Means (And Why It’s Three Things)

When people talk about context in enterprise AI, they’re usually conflating three distinct things.

The first is memory. What happened before. The decision traces Foundation Capital describes: who approved a discount exception, which vendor was chosen last quarter, why a product launch was delayed. This is institutional history, and most of it lives in people’s heads or buried in email threads.

The second is situation. What’s happening right now. Current inventory levels, open customer commitments, capacity constraints, market conditions. This is the live data that any decision needs to reference. Most enterprises have this in their systems of record, though accessing it across systems remains hard.

The third is working state. What’s active in this specific decision process. The scenarios being compared, the tradeoffs on the table, the assumptions being tested. This is the most ephemeral and the least captured. It evaporates when the meeting ends.

Building a context graph from memory alone, which is what passive decision trace capture gives you, gives you a record of what happened. It misses the reasoning that connected the situation to the choice, and the working state that shaped how options were evaluated.

A judgement graph captures all three. It connects institutional memory to the live situation to the active reasoning process. And because it captures the reasoning, not just the outcome, it creates something context alone cannot: a system that learns which reasoning patterns lead to good outcomes and which don’t.

Context Is Scalar. Judgement Is a Vector.

The difference is structural.

Context is scalar. It describes a state. Inventory is at 12,000 units. The customer’s churn risk score is 74. Last quarter’s revenue missed forecast by 8%. These are points in space. They just are.

Judgement is a vector. It has magnitude and direction. It describes a force applied to move an organization from one state to another. “We’re going to increase safety stock by 15% because the supplier concentration risk in Southeast Asia exceeds our tolerance, even though it raises working capital by €2M.” That’s a decision with direction, magnitude, reasoning, and an implicit bet about the future.

A context graph stores points. A judgement graph stores the forces between them. And forces compound in ways that points cannot.

How Judgement Compounds

The most overlooked property of captured judgement is that it earns compound interest. Three kinds.

Calibration compound. When a judgement graph captures not just the decision but the bet behind it, the system can score accuracy over time. Not in the abstract, but per person, per domain, per type of call. After two years of captured judgements, the system knows that the VP of Sales is consistently 20% over-optimistic on Q3 demand projections. It knows that the operations lead in Frankfurt underestimates logistics lead times by about a week. This is calibration data. It makes the next forecast structurally better because the system can weight inputs by demonstrated accuracy rather than by seniority or confidence.

Structural reusability. When you capture context, you capture data that’s specific to a moment. When you capture judgement, you capture reasoning that can travel. A decision about inventory buffers in the Swiss division (the constraints evaluated, the tradeoffs accepted, the risk thresholds applied) can be cloned and adapted for the Brazilian division. Not the numbers. The logic. The reasoning template. Because the judgement graph stored why the decision was made, not just what was decided, it becomes a transferable asset across geographies, business units, and time.

The anti-atrophy effect. Every enterprise has experienced the quiet catastrophe: a senior architect leaves, a veteran supply chain planner retires, a PM who held the institutional memory of why a system was designed a certain way moves on. And with them goes judgement that took a decade to build. The organization keeps its data. What it loses is the reasoning that made the data meaningful. Worse, there’s a silent tax on every subsequent decision in that domain. The replacement doesn’t know which constraints are sacred and which are negotiable. They don’t know which supplier relationships have history behind them, or why a particular safety stock formula was chosen over the textbook alternative. So they guess. And guessing at constraints that someone else understood deeply is how organizations make locally rational decisions that are globally incoherent. A judgement graph is organizational memory that doesn’t walk out the door. Not because it replaces the expert, but because it captured the structure of their thinking while they were still making decisions.

Context decays. Last quarter’s inventory levels are already stale. Judgement appreciates. The reasoning pattern behind a good inventory decision is more valuable a year from now, when the same type of decision comes around again with different numbers.

The Trillion-Dollar Shift

Enterprises have never systematically stored the reasoning behind their decisions. That reasoning is their least captured asset.

The opportunity is in building infrastructure that compounds human judgement across people, divisions, and time.

The companies that capture this layer will have something no amount of data can replicate: an organizational record of not just what was decided, but why it was decided, how confident the decision-makers were, whether that confidence was warranted, and what reasoning patterns consistently led to good outcomes.

Context tells you what the world looks like. Judgement tells you which direction to push, how hard, and why. One is a photograph. The other is a navigation system that improves with every trip.

The next trillion-dollar platforms won’t just capture context. They’ll compound judgement.