Improvement Roadmap TODO¶

[x] Complete initial codebase analysis
[x] Clarify the CLI contract for replay
[x] Decide whether replay should remain mutating or become fully read-only
[x] Update documentation home and CLI help text to match the final replay behavior
[x] Add tests that lock the intended CLI contract in place
[x] Align InMemoryEventStore idempotency behavior with production backends
[x] Reuse shared idempotency validation logic in test fakes where practical
[x] Add parity tests covering idempotent retries when trace_seq changes
[x] Decide whether dg_precedent_index is a real query surface or an internal artifact
[x] Either wire find_precedents() to dg_precedent_index or remove/simplify the unused projection path
[x] Add tests that document the intended precedent-query data flow
[x] Refactor projection sync/replay to process events in bounded batches instead of loading the full log at once
[x] Add store-level iteration or paging helpers for large-log replay
[x] Add performance tests for replay/sync on larger datasets
[x] Normalize projection staleness behavior across query APIs
[x] Decide whether get_trace_summary() should enforce staleness checks like graph and precedent queries
[x] Add tests for the chosen get_trace_summary() staleness behavior
[x] Add typed helper APIs for common event writes on top of the generic dict-based append_event() surface
[x] Keep the current generic API stable while introducing typed convenience helpers
[x] Add API contract tests for any new helper methods
[x] Add projection health tooling for cursor lag, digest inspection, and recovery workflows
[x] Expose projection health details in the CLI or a small admin/debug surface
[x] Document operational guidance for projection recovery and multi-writer scenarios