Module ledger 

Content-hash idempotency ledger for zeph knowledge ingest.

The ledger records every (source_uri, content_hash) pair that has been successfully ingested, so unchanged inputs are skipped on subsequent runs (spec-067 FR-012, INV-5).

Scope and limitations

This is a re-read / cost guard only — it prevents redundant embedding and LLM extraction for inputs whose content has not changed since the last run. It does not:

• Reconcile LLM extraction drift across model versions. If the model changes, re-ingest must be forced by clearing ledger rows for the affected sources.
• Delete stale Qdrant points when a file’s content changes. When content_hash changes for the same path, the old vectors remain in Qdrant (orphaned). Stale-point cleanup is a Phase-2 concern and is documented as a known MVP limitation in the testing playbook.

Database

The ledger shares the agent’s existing SQLite pool (SemanticMemory::sqlite().pool()). It is operator-triggered, one row per ingested file, and is not on the hot path — no dedicated file is needed (mirrors the skill_trace_sessions pattern).

Examples

use zeph_memory::graph::ingest::IngestLedger;

// In practice, pool comes from SemanticMemory::sqlite().pool().clone().
let ledger = IngestLedger::new(pool);
let hash = IngestLedger::content_hash(b"hello world");
let uri = "specs/README.md@abc123";
let batch = "01J0000000000000000000000";

ledger.mark_ingested(uri, &hash, batch, 0, 0).await?;
assert!(ledger.is_ingested(uri, &hash).await?);

Structs

IngestLedger

Content-hash idempotency ledger for zeph knowledge ingest.

LedgerEntry

A single row from the knowledge_ingest_ledger table.

Enums

BatchIdResolution

Outcome of resolving a (possibly abbreviated) batch id prefix against the ledger.