Developer Late-Arrival Reference

Metadata

• Owner: Origo Engineering
• Last updated: 2026-04-14
• Slice/version reference: S49

Purpose

• Developer reference for the canonical late-arrival registry and the explicit boundary between projection time semantics and operational progression.

Authoritative machine contract

• contracts/canonical-late-arrival-v1.json

Registry structure

• source declaration key:
  • (source_id, stream_id)
• projection declaration key:
  • (source_id, stream_id, projection_id)
• projection time-mode taxonomy:
  • event_time
  • ingestion_time
• lateness semantics taxonomy:
  • clock_lag
  • publication_delay
  • availability_boundary
  • capture_cadence_gap

Important implementation boundary

• Current data-serving projections remain event_time.
• In current Origo, event_time does not mean projector fetch/resume/checkpoint progression.
• Projector progression stays operational and stream-sequence-based.
• This slice freezes the declaration and evidence surfaces, not a hidden full bucket-replacement engine.

Runtime evidence surfaces

• origo/events/runtime_audit.py
  • late_arrival_rejection
  • late_arrival_supersession
• origo/query/bitcoin_mempool_boundary.py
  • query-boundary errors for pre-capture mempool requests

Current supersession boundary

• Native serving has no bucket supersession concept.
• Aligned serving requires a named supersession surface.
• The registry records that the aligned bucket replacement mechanism is not runtime-enabled yet.
• Do not fake this by implying MergeTree rows replace themselves.