Repository navigation
test: add property tests for bitemporal interval and retraction semantics #211
Description
Source
Extracted from the reliability audit snippet: overlapping valid/tx time, historic range, straddling boundaries, gaps, conflicting facts, retractions, point/interval query classes.
Programmatic test scope
Add property-style tests for bitemporal semantics across generated transaction sequences and valid-time intervals.
Cases to cover
- Overlapping valid intervals for the same EAV triple.
- Gaps and boundary points at
valid_from/valid_to. - Retraction followed by re-assertion.
:as-oftransaction-count snapshots before, at, and after retractions.- Point-in-time, time interval, time-point lookup, and time-interval lookup queries.
Acceptance criteria
- Tests compare query output against a small independent reference model.
- Generated cases include edge boundaries and conflicting facts.
- Failures print a minimal reproducible transaction/query sequence.
Activity
adityamukho commentedon May 2, 2026
Audit finding: The deep audit identified a concrete bug that this issue's test cases would catch:
net_asserted_facts()(storage.rs:463-484) keys only on(entity, attribute, encode_value)— ignoringvalid_from/valid_to. This means if the same EAV triple is asserted with two different valid-time intervals (e.g., salary=100k valid 2020-2022, then salary=100k again valid 2024-2026), only the latest-transacted one survives. A:valid-at "2021-06-01"query incorrectly returns empty.The
pending_key(storage.rs:17) correctly includesvalid_from, valid_toas part of the dedup key, so storage treats them as distinct facts. But the query pipeline collapses them before the valid-time filter runs.A property test covering "overlapping valid intervals for the same EAV triple" (listed in this issue's cases) would expose this immediately. The fix will land separately, but this issue's test scope is confirmed as high-value.