Canonical Dedup Strategy Reference

Metadata

Freeze one authoritative canonical dedup contract for every current source family.
This reference explains the machine-readable registry at contracts/canonical-dedup-strategy-v1.json and how it relates to live write behavior.

Canonical dedup is authoritative only at the canonical write boundary.
The dedup lookup scope is current-truth lineage:
- canonical_event_log_active_v1
- plus logical reset-boundary lineage
Raw append history remains forensic truth and does not override logical reset-and-rewrite intent.

natural_key: stable source-native identity or stable composite.
deterministic_synthetic_key: deterministic key over stable logical-observation fields and never over value fields.
hybrid_key: durable source-native entity identity plus stable snapshot or revision boundary.

Binance / OKX / Bybit: trade identity remains exchange-native.
ETF: canonical identity is record_source_id:observed_date:metric_name:sha256(sorted_dimensions_json) and excludes value fields.
FRED: canonical identity is record_source_id:observed_date:metric_name:realtime_start:realtime_end and excludes value fields while keeping revision-window identity explicit.
Bitcoin block transactions: canonical identity is block_height:transaction_index:txid.
Bitcoin mempool: canonical identity is snapshot_at_unix_ms:txid.
Bitcoin blockchain-derived metrics: canonical identity is block-height keyed per stream.

Those behaviors may remain visible in proof outputs, but they are hygiene only and must not redefine canonical correctness.

Exchange fast or staged write paths may still operate as empty-partition initial-load exceptions where declared in the registry.
They are fenced exceptions, not a second idempotency regime.