When a Sensor Loses Its Oracle: Three-Arm Differential Observability and Zero-Oracle Fail-Closed Abstention for Membrane Fouling Diagnostics

Soft sensors for reverse osmosis (RO) and forward osmosis (FO) desalination must separate reversible concentration polarization (CP) from irreversible mineral scaling and biofouling while remaining resilient to sensor drift, temperature shocks, and reference-channel contamination. We identify two structural vulnerabilities in differential membrane diagnostics: (1) a 2-arm differential cartridge cannot distinguish reference-arm contamination from test-arm cleanliness, producing a 71.17% estimation error without warning under 10% partial reference fouling (Scenario NC5), whereas a 3-arm differential manifold with a witness-ratio guard (τref = |rref(t) − rref,0|/rref,0 > 0.03) restores Rank-2 Fisher observability (det I > 0) and cuts error to 7.72%; and (2) autonomous diagnostic codebases suffer from oracle leakage when simulator ground-truth fields (ground_truth_scale_um) or synthetic nominal reference values remain accessible inside production classes or when invariants rely on strip-prone assert statements under python -O. After hardening CommercialPhysicalWitnessB (removing all simulator fields, enforcing explicit runtime ValueError guards across 19/19 standard and -O unit tests), the zero-oracle estimator achieves 5/5 (100%) clean operating-sweep coverage and reduces false-confident commits on 9 adversarial fault challenges from 7/9 (77.8%, SystemA) and 8/9 (88.9%, SystemC/R0) to 1/9 (11.1%, residual C07 80 nm sub-resolution CaSO4 film), while correctly abstaining (21/21 ABSTAIN_MISSING_REFERENCE_EVIDENCE) on 21 published literature rows that lack a physical reference channel. Evidence classification: simulation only (SIMULATION_VERIFIED_WET_LAB_UNTESTED). No wet-lab membrane experiment has been executed; zero wet-lab membrane hours. Published as a methods protocol / registered-protocol preprint, not as experimental results. The \$551.32 COTS benchtop flow-loop protocol in Section 3 is pre-registered and specified, not run. All metrics are simulation replays and literature-row audits, deterministically reproduced by reproduce_paper2_zero_oracle_pareto.py, included in the reproducibility bundle. Reproducibility code in the bundle is Apache License 2.0; paper text is CC BY 4.0.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24324954
Primary Topic
Membrane Separation Technologies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

When a Sensor Loses Its Oracle: Three-Arm Differential Observability and Zero-Oracle Fail-Closed Abstention for Membrane Fouling Diagnostics

Terrynce A. White
2 links
Zenodo (CERN European Organization for Nuclear Research)
Membrane Separation Technologies
preprint

When a Sensor Loses Its Oracle: Three-Arm Differential Observability and Zero-Oracle Fail-Closed Abstention for Membrane Fouling Diagnostics

Terrynce A. White
preprint en

Abstract

Soft sensors for reverse osmosis (RO) and forward osmosis (FO) desalination must separate reversible concentration polarization (CP) from irreversible mineral scaling and biofouling while remaining resilient to sensor drift, temperature shocks, and reference-channel contamination. We identify two structural vulnerabilities in differential membrane diagnostics: (1) a 2-arm differential cartridge cannot distinguish reference-arm contamination from test-arm cleanliness, producing a 71.17% estimation error without warning under 10% partial reference fouling (Scenario NC5), whereas a 3-arm differential manifold with a witness-ratio guard (τref = |rref(t) − rref,0|/rref,0 > 0.03) restores Rank-2 Fisher observability (det I > 0) and cuts error to 7.72%; and (2) autonomous diagnostic codebases suffer from oracle leakage when simulator ground-truth fields (ground_truth_scale_um) or synthetic nominal reference values remain accessible inside production classes or when invariants rely on strip-prone assert statements under python -O. After hardening CommercialPhysicalWitnessB (removing all simulator fields, enforcing explicit runtime ValueError guards across 19/19 standard and -O unit tests), the zero-oracle estimator achieves 5/5 (100%) clean operating-sweep coverage and reduces false-confident commits on 9 adversarial fault challenges from 7/9 (77.8%, SystemA) and 8/9 (88.9%, SystemC/R0) to 1/9 (11.1%, residual C07 80 nm sub-resolution CaSO4 film), while correctly abstaining (21/21 ABSTAIN_MISSING_REFERENCE_EVIDENCE) on 21 published literature rows that lack a physical reference channel. Evidence classification: simulation only (SIMULATION_VERIFIED_WET_LAB_UNTESTED). No wet-lab membrane experiment has been executed; zero wet-lab membrane hours. Published as a methods protocol / registered-protocol preprint, not as experimental results. The \$551.32 COTS benchtop flow-loop protocol in Section 3 is pre-registered and specified, not run. All metrics are simulation replays and literature-row audits, deterministically reproduced by reproduce_paper2_zero_oracle_pareto.py, included in the reproducibility bundle. Reproducibility code in the bundle is Apache License 2.0; paper text is CC BY 4.0.

Zenodo (CERN European Organization for Nuclear Research)
Membrane Separation Technologies
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon