SAVI LOOPIQ
Evidence before claims

Validation evidence and limitations

LoopIQ distinguishes implementation consistency, independently sourced equation evidence, independent implementation cross-checks, and true third-party or field validation. Claims advance only when reproducible artifacts and review evidence exist.

AreaEvidence retainedWhat it provesRemaining gate
Published-equation conformanceRetained SIMC, IMC, Cohen–Coon and Ziegler–Nichols cases are transcribed from published sources and checked against the production TypeScript rules.Equation conformance for the retained cases; not certification of the product or plant performance.Expand the independently sourced case set as new model/rule families become release-critical.
Standalone reference oracleA pure-Python standard-library implementation, with no LoopIQ imports, recomputes eight published tuning cases and first-principles frequency-domain robustness metrics.Independent implementation agreement for the retained FOPDT cases.Obtain named third-party tool comparisons and independent controls-engineer review.
Cross-language tuning parityAutomated corpus compares Python, Dart and TypeScript gains at strict tolerances.Consistency among implementations; the generated parity corpus remains implementation-derived.Keep parity separate from independently sourced reference evidence.
Numerical robustnessMs, Mt, gain margin and phase margin are calculated from a numerical frequency sweep and cross-checked against the standalone oracle on retained cases.Numerical agreement for the supplied models, controllers and retained reference cases.Retain licensed external-tool comparisons before claiming third-party validation.
Process identificationPython/Dart share the 63.2%, Sundaresan and LM-NLS contract. Browser identification enforces the same 20:1 data-quality refusal threshold but retains a different estimator.Regression behavior and data-quality refusal boundaries.Add a statistically justified browser-vs-reference equivalence study over noisy/adversarial captures or port the common estimator.
Field outcomesNo retained independent field-validation package is claimed today.Nothing about guaranteed plant performance.Collect consented, anonymized before/predicted/after cases and independent review.

Equations and sources

The retained reference package cites Skogestad SIMC, Rivera–Morari–Skogestad IMC, Cohen–Coon and Ziegler–Nichols. The standalone oracle independently implements those equations plus the FOPDT/PID frequency response used to calculate sensitivity peaks and stability margins. It does not import LoopIQ code.

Inspect the corpus, equations and citations →

Safety boundary

LoopIQ is decision support. Independent implementation agreement is not third-party certification, a field trial, or a guarantee of safe or profitable plant performance. LoopIQ does not write gains to a live controller. A qualified engineer must verify model quality, controller form, units, constraints and commissioning procedure.

Try the bounded demo