Behavioral archetypes
Every agent observed on x402 is classified into one of nine archetypes once an hour, based on six signals: total tx count, median tx amount, lifetime in days, distinct merchants paid, fraction of activity in night hours (22:00–06:00 UTC), and median cadence jitter (delay between consecutive tx).
The classifier is deterministic and rule-based, not ML — every label is reproducible from the same input data. Rules are evaluated in priority order; the first match wins.
📊 Live distribution
⚡ Sprinter
An agent that fires ≥100 transactions in ≤1 day. Burst-mode behavior: it appears, hammers the network, then often goes silent forever. Sprinters are typical of one-off scraping jobs, batch inference runs, or scheduled crawls.
tx_count ≥ 100 AND lifetime_days ≤ 1
🏃 Marathoner
An agent with ≥200 transactions over ≥7 days. Long-haul, programmatic spend. Marathoners look like always-on services: trading bots, monitoring agents, recurring data subscriptions.
tx_count ≥ 200 AND lifetime_days ≥ 7
👻 Ghost
A wallet with exactly one x402 transaction in the entire dataset. Ghosts are the long tail of the agent economy: experimental fire-and-forget identities, single-use accounts, or wallets the operator burned after one purchase. They tend to dominate the agent census numerically.
tx_count = 1
🔥 Burner
A short-lived wallet with 2 to 99 tx over less than 24h. Looks like a disposable agent: a session-bound key that does a few things and then stops. Common pattern for SDK examples and quick experiments.
lifetime_days < 1 AND tx_count BETWEEN 2 AND 99
🦉 Night owl
An agent whose activity is heavily skewed to night hours (22:00–06:00 UTC) — at least 60% of its tx fall in that window. Could be a scheduled batch job in a specific timezone, or a deliberate night-only crawl strategy.
night_ratio ≥ 0.6
🐝 Worker bee
A focused agent: ≥30 tx directed at ≤3 distinct merchants. Doing one job, paying the same handful of services repeatedly. Strong signal of a targeted automation pipeline.
tx_count ≥ 30 AND distinct_merchants ≤ 3
🎯 Hunter
A wide-footprint agent: ≥30 tx spread across ≥10 distinct merchants. Looks like a discovery / scanning behavior: sampling many services, possibly comparing prices or testing endpoints.
tx_count ≥ 30 AND distinct_merchants ≥ 10
🤖 Drone
A highly regular agent: median delay between consecutive tx is under 60 seconds, with at least 10 tx. Tight cadence, almost certainly a tight loop. The most "machine-like" archetype.
cadence_jitter_ms < 60 000 AND tx_count ≥ 10
❓ Unknown
Anything that doesn't match the rules above — typically agents with fewer than 30 tx that don't fit a sharper definition yet. As the dataset grows, many of these get reclassified into something more specific within hours.
Why behavioral, not corporate?
Most x402 agents are anonymous. Their wallets aren't claimed by a company, registered in any directory, or linked to a known entity. Trying to attribute them to "OpenAI" or "Anthropic" is mostly hallucination. What we can do reliably is describe how they act. Behavioral archetypes give us a stable, falsifiable language for an economy that has chosen to be pseudonymous.
Caveats
- Archetypes can flip. A burner today may become a marathoner next week. The classifier reruns hourly and overwrites the previous label.
- The rules above are the current default. They will evolve. We will document changes in this page's history rather than silently rewrite them.
- The "night_ratio" rule uses UTC, not the agent's local timezone. We have no way to infer that.