80 paired cases across CLEAN + FULL_ADVERSARIAL.
SHOW THE WIN.
KEEP THE FAILURE.
The saved reproduction record for the frozen All-Known-Methods benchmark reports a pass. This page displays that historical record, not a newly executed test. The clean-data disadvantage remains visible beside the adversarial advantage.
Change the scenario. Keep the context.
Mean absolute error (MAE): lower is better. These controls display saved public summary values; they do not run or expose the underlying implementation.
Overall
Saved aggregate result across the frozen benchmark scenarios.
LEXIUM MAE0.615826
STANDARD KALMAN MAE3.352797
RELATIVE TO STANDARD KALMAN81.632% lower
Calculation and evidence boundary
Relative reduction = (1 - LEXIUM MAE / standard Kalman MAE) x 100. Negative values mean higher error.
- Record
- Frozen All-Known-Methods v1.4 / saved reproduction
- Recorded status
- PASS
The public site intentionally omits runner code, internal hashes, seed inventory, implementation details and private evidence files.
Synthetic frozen benchmark only; not a universal real-world superiority claim.
BLACK BOX RUNNER v0.3
CANDIDATE / ACCEPTANCE PASS.
The v0.3 acceptance record tested the loopback-only runner at 100,000 samples. JSON and LXS1 produced exactly equivalent float64 outputs in that test (maximum output difference 0.0), corrupted LXS1 was rejected, and an unlisted web origin was rejected.
100,000 SAMPLESJSON ↔ LXS1 EXACTCORRUPTION REJECTEDUNLISTED ORIGIN REJECTEDLOOPBACK ONLYRAW PERSISTENCE: NONE
The recorded 100k request payload was 1,821,174 bytes for JSON and 800,056 bytes for LXS1 — about 2.276× fewer wire bytes. This does not establish a 30× claim.
ENGINE FUNCTION SHA-256: 5F8EA404CCC416B07FF71763FE7BC0914DD5042930B2170DB1FE84752D9A44B1
v0.3 RUNNER MANIFEST SHA-256: 53ddecb54db9ce025e2d1a4f54c55fd07ceafe9bd90f98b964fd720129f27b41
Same frozen benchmark and paired case structure.
Relative to standard Kalman in this synthetic frozen benchmark.
Runner integrity verified; no recorded incompatibilities.
NOMINAL COST
LEXIUM0.132669
STANDARD KALMAN0.130533
Standard Kalman is marginally lower on CLEAN data. That result is retained rather than hidden.
ROBUSTNESS RESULT
LEXIUM1.098984
STANDARD KALMAN6.575061
The strongest separation occurs under the defined contamination family.
OVERALL LEADERBOARDEIGHT METHODS.
EIGHT METHODS.
SAME FROZEN TEST.
#METHODMAEP95MAX
1LEXIUM 6.10.20.6158262.1274032.399483
2THRESHOLD KALMAN0.7522812.2394522.499625
3PARTICLE FILTER0.9539842.3588853.170475
4HUBER0.9636392.4320383.371903
5KALMAN 6.10.23.35279713.19712121.109264
6STANDARD KALMAN RUN33.35279713.19712121.109264
7STANDARD KALMAN RUN53.35279713.19712121.109264
8ADAPTIVE KALMAN3.82152316.79866228.992334
WHAT THIS RESULT
DOES AND DOES NOT SAY.
It establishes reproducible performance inside this frozen synthetic benchmark. It does not establish universal real-world superiority, domain independence, or superiority to every state-of-the-art estimator. No estimator parameter was tuned after outcomes were observed.
40 FIXED SEEDS2 SCENARIOS8 METHODS80 CASES / METHODNO FAILURESHASH VERIFIED