Skip to content
mempill

As-Of / Bi-Temporal Correctness Benchmark

mempill ships a reproducible, assertion-based benchmark of its as-of query correctness on both independent bi-temporal axes — valid-time and transaction-time — including the valid-time succession scenario described in Valid-Time Succession.

Reproducing the results

Section titled “Reproducing the results”
Terminal window
cargo run --release --example asof_correctness_benchmark -p mempill
  • Artifact: mempill-facade/examples/asof_correctness_benchmark.rs
  • Data: none external — every claim ingested by the benchmark is constructed inline with fixed, deterministic RFC 3339 timestamps. No network access, no proprietary fixtures, no environment-specific configuration.
  • Backend: in-memory SQLite (mempill::open_default_in_memory()) — the same public API surface a first-time integrator uses.

A third party can clone the public mempill repository and run the command above to reproduce identical PASS/FAIL results.

Results

Section titled “Results”

Run against mempill 0.4.0, in-memory SQLite, release build:

mempill as-of / bi-temporal correctness benchmark
===================================================


[PASS] valid_time::point_in_time_recall
[PASS] valid_time::gap_returns_no_belief
[PASS] tx_time::independent_of_valid_time
[PASS] combined::valid_time_and_tx_time
[PASS] succession::three_way_chain
[PASS] contested::genuine_conflict_not_silently_resolved
[PASS] point_claim::documented_no_belief_at_instant


===================================================
Result: 7/7 scenarios passed
Scenario Axis What it demonstrates
valid_time::point_in_time_recall valid-time Point-in-time recall correctly selects the right value across a 3-claim history at three distinct query instants
valid_time::gap_returns_no_belief valid-time A query instant in the gap between two non-overlapping windows returns no belief, not a false contested or resolved result
tx_time::independent_of_valid_time tx-time “What did we believe as of tx-time T?” correctly excludes a claim learned after T, independent of that claim’s valid-time window
combined::valid_time_and_tx_time both The two axes compose correctly when both are set on the same query — the valid-time axis governs claim selection without being overridden by tx-time visibility
succession::three_way_chain valid-time (succession) A 3-window non-overlapping succession chain selects the correct claim at 5 distinct query instants, including exact boundary instants (start-inclusive/end-exclusive)
contested::genuine_conflict_not_silently_resolved valid-time (conflict) Two trusted, overlapping windows surface as Contested with both candidates exposed — the honest counter-case to succession
point_claim::documented_no_belief_at_instant valid-time (edge case) A zero-duration point claim queried at its own instant returns NoBelief, matching mempill’s documented half-open [start, end) interval semantics

What these results affirmatively demonstrate

Section titled “What these results affirmatively demonstrate”
  • mempill’s as-of queries correctly separate two independent time axes: what was true in the world (valid-time) and what the system had learned by a given moment (transaction-time). Setting one axis does not corrupt or override the other.
  • A chronological chain of non-overlapping facts (a succession) is resolved to a single correct value at any queried instant, including exact boundary instants, without producing a false conflict.
  • A genuine conflict — two trusted claims that actually overlap in time — is surfaced explicitly as Contested, with both candidates visible, rather than silently guessed at or averaged away.
  • Every one of the above is reproducible from a single public command against the published crate, with no external data or environment-specific setup.
Section titled “Related concepts”