Building the Event Clock — Kirk Marple

Summary

Kirk Marple (Graphlit) responds to Foundation Capital’s “Context Graphs: AI’s Trillion-Dollar Opportunity” with a sharp distinction: every system has a state clock (what’s true now) and an event clock (what happened and the reasoning behind it). We’ve built trillion-dollar infrastructure for state clocks — databases, warehouses, CRMs — but barely anything for event clocks.

The mental model is clean:

1. State Clock vs. Event Clock. State tells you the current value. Events tell you how you got there and why. A CRM says “closed lost” but not why the deal died. A treatment plan shows a drug switch but not the clinical reasoning. A config file shows the current value but not the decision that set it. The reasoning layer is almost always lost.

2. The Missing Infrastructure. We have mature, battle-tested infra for state (RDBMS, warehouses, object stores). We have almost nothing for events — especially unstructured events carrying reasoning, context, and temporal relationships. This is the gap.

3. Unstructured Data Across Time and Space. Graphlit’s bet since 2021: index unstructured data with temporal and spatial awareness. Not just what exists, but when it happened, where it happened, and why the state changed.

The data community reaction — “they are going to reinvent data warehousing” — misses the point. The warehouse captures state snapshots. The event clock captures the narrative between snapshots.

Connections

• 01-projects/phdata/index — phData clients live in this gap constantly. They have warehouses full of state but lose the reasoning trail. Event clocks could be a consulting wedge: “your warehouse tells you what, we’ll help you capture why.”
• 01-projects/data-marketplace/index — Data products that capture event context (the why behind state changes) are dramatically more valuable than raw state exports. This is a differentiation axis.
• 06-reference/2026-04-03-the-data-warehouse-toolkit — Kimball’s dimensional model is state clock infrastructure. Slowly changing dimensions are a crude attempt at event clocks — they track that something changed but not why. The event clock is what SCDs were trying to be.
• 06-reference/2026-04-04-ontology-taxonomy-knowledge-graphs — Context graphs and event clocks are complementary. The ontology defines the structure; the event clock adds temporal reasoning to the graph edges.
• 06-reference/concepts/compounding-knowledge — Event clocks compound. State clocks don’t — they overwrite. A system that preserves reasoning gets smarter over time because it can pattern-match across historical decisions.

Open Questions

• What does event clock infrastructure actually look like in practice? Is it append-only logs with semantic indexing? Something closer to a knowledge graph with temporal edges?
• Could our vault function as an event clock for Ray Data Co decisions? We’re already capturing reasoning — are we missing the temporal indexing?
• Is there a product opportunity in “event clock as a service” for specific verticals (healthcare reasoning trails, deal loss forensics, config decision audit)?