Events

PING distinguishes layers of truth and refuses to let them silently merge. Source truth is what happened: an event, an observation, an imported artifact. Derived truth is what the system calculated: embeddings, projections, summaries, classifications. Presented truth is what PING chooses to show. Evidence is what lets someone else verify the claim.

A projection is not the source event. A summary is not the underlying evidence. A hash is not the object it identifies. Every transformation must preserve identity, or the system is manufacturing certainty.

An event moves through a typed pipeline:

• Capture: the raw occurrence is recorded with identity, timestamp, source, and provenance.
• Canonicalization: the event is normalized into canonical form so identical causes produce identical records.
• Distillation: observations are extracted with source spans and confidence; contradictions are preserved.
• Emission: verified, high-quality intelligence is emitted to the canonical knowledge layer.

Each stage has typed states, including failure states. A stage that fails must say so explicitly; it must never emit an empty success.

PING never confuses “nothing exists” with “the system failed to find it.” Service failures, configuration failures, and schema failures each have deterministic semantics: what failed, which boundary failed, whether retry is safe, whether state changed, and whether the result can be trusted. An empty answer is never a hiding place for an infrastructure failure.

The event pipeline is in active development. The capture, canonicalization, and distillation stages exist with typed state tracking; emission policies (what qualifies as canonical knowledge vs. operational data) are being refined. This page will be updated as stages stabilize.