Canonical Event Names

For OpenClaw, event names should follow:

domain.subject.measure[.suffix]

Event names are raw event identities.

• use event names for stable events
• use rollups and policies for hourly or daily logic
• do not encode time windows into event names
• treat meta as inspection-only context in v1

Core agent-monitoring events:

• tool.errors.count
• tool.calls.count
• task.duration_ms
• tokens.used
• heartbeat.missed.count

Market and sentiment events:

• market.bitcoin.price_usd
• market.bitcoin.volume_usd
• market.bitcoin.sentiment_score
• sentiment.bitcoin.reddit.score
• sentiment.bitcoin.reddit.positive.count
• sentiment.bitcoin.reddit.negative.count
• sentiment.bitcoin.reddit.post_volume.count

Do not encode time windows into event names. Bad examples:

• tool.errors.last_hour
• reddit_sentiment_today
• market.bitcoin.price_change_pct_1h

Use meta for inspection context like tool_name, task_name, workflow, source, or query, not for alert semantics in v1.

Recommended aggregates by event shape

These are operator conventions for policies and charts, not event-definition metadata stored in Ingres.

• *.errors.count, *.count, *.used -> usually sum; sometimes count when you need sample volume
• *.duration_ms -> usually average, p95_est, p99_est, or max
• *.score, *.price_usd -> usually average, min, or max
• *.gauge -> usually average, min, or max
• unknown suffix -> start with sum only if the event is intentionally additive

Analytics windows

When you need explicit comparisons, use the analytics window API rather than inventing derived event names:

• period=1h&previous=previous_window
• period=1h&previous=previous_7d
• period=1h&previous=previous_1m
• period=1d&previous=previous_1m
• period=7d&previous=previous_1m
• period=1m&previous=previous_window