Agent Behavior

Dead Reckoning is a scroll-driven analytics dashboard that diagnoses why 87% of ERC-8004 agents fail to generate revenue. Built on Google BigQuery's public Ethereum dataset, it classifies all 34,569 registered agents by failure type — registry pollution, DevRel gap, gas failures, intent mismatch, and day-one cliff abandonment — and maps the full $215M unrealized revenue gap across the protocol funnel.

The dashboard tells a story in six acts: the indictment (87% failure rate), the collapse (animated funnel from 34,569 registrations to 433 active agents), the timeline (monthly cohort churn), the forensics (3D agent sphere colored by failure type), the bill (losses itemized by churn category), and the turn (interactive recovery modeler showing $22.2M is recoverable with two targeted fixes).

This isn't a product. It's a mirror — showing what the ERC-8004 protocol looks like from the outside, and exactly what it's worth if you fix the first 48 hours of developer onboarding.

The data layer is built entirely on Google BigQuery's public Ethereum datasets — no custom infrastructure, no rate limits. A single master SQL query joins four sources: IdentityRegistry logs (ERC-8004 registrations), ReputationRegistry feedback events, outbound transaction counts with function selector classification, and execution traces split into success and failure signals. The query returns one row per agent with 40+ columns covering funnel stage, intent category, churn type, ETH flows, and failure diagnostics. It processes ~4.7TB and returns 34,569 rows in ~31 seconds.

The classification pipeline runs in Python via google-cloud-bigquery, resolves ENS names for the top wallets, and outputs churn_master.csv. A multi-agent Claude system interprets wallet signals using RevOps methodology (cross-sell whitespace, churn attribution, next-best-action) applied to on-chain behavior.

The frontend is vanilla JS + HTML/CSS with no backend required. Visualizations use Three.js for the 3D agent sphere (color = failure type, size = dollar exposure) and Chart.js for the funnel and timeline charts. Scroll-driven animations use IntersectionObserver. The recovery modeler is a live slider that recomputes unlocked value in real time. Data is loaded from the CSV via Papa Parse at runtime — no hardcoded numbers, everything live from BigQuery output.