Methyl mercaptan (CH3SH) odorant storage AI adversarial injection: how a cooling water valve actuator failure — displayed adequate by an upward ±8 DN root-cause attack — chains through 48°C vessel overtemperature, compound overpressure (44 psig PRD approach suppressed to 14 psig), and area CEMS suppression (0.84 ppm to 0.025 ppm) while the world’s lowest odor threshold creates the olfactory habituation paradox — OSHA PSM TQ 15,000 lbs, pressurized liquefied thiol at bp 6.2°C, third causal four-surface attack chain, Glyphward threshold 35

CH3SH chemistry, industrial applications, and the OSHA PSM TQ 15,000 lbs calibration

Methyl mercaptan is the simplest member of the thiol (mercaptan) class of organosulfur compounds — a sulfur analogue of methanol in which the hydroxyl group is replaced by a thiol group (−SH). The carbon–sulfur bond length of 1.82 Å, the S–H bond length of 1.34 Å, and the C–S–H bond angle of 99.4° give CH3SH a molecular geometry that interacts strongly with olfactory receptor binding sites at concentrations in the parts-per-trillion range. CH3SH is a colorless, flammable gas at ambient temperature and pressure, with a vapor pressure at 20°C of approximately 25 psig, meaning it must be stored as a pressurized liquid in sealed vessels rated for at least 50–75 psig working pressure. As a liquefied gas with boiling point 6.2°C — comparable to propane’s boiling point of −42.1°C but far closer to ambient temperature — CH3SH is more acutely sensitive to temperature excursions than most other pressurized liquefied gases: a 13°C temperature rise from 35°C to 48°C increases CH3SH vapor pressure from approximately 38 psig to 44 psig, a 16% increase that can bring a storage vessel within 6 psig of a 50 psig PRD setpoint.

CH3SH is produced industrially by the Chevron Phillips Chemical / Arkema / Evonik process in which methanol reacts with hydrogen sulfide over an alumina catalyst at 300–400°C and 1–5 bar (CH3OH + H2S → CH3SH + H2O), with subsequent distillation to separate CH3SH from unreacted feedstocks, DMDS (dimethyl disulfide, CH3SSCH3) byproduct, and catalyst deactivation products. Global CH3SH production capacity is approximately 500,000 tonnes per year, distributed among Evonik Industries (Marl, Germany; Wesseling, Germany; Mobile, Alabama; Jurong, Singapore), Arkema SA (Lacq, France; Beaumont, Texas), Chevron Phillips Chemical (Borger, Texas), and several Chinese producers. The two primary final-use markets — natural gas odorant and L-methionine synthesis — account for approximately 75% of global consumption, with DMDS synthesis for refinery catalyst sulfiding and dimethyl sulfoxide (DMSO) production accounting for the remainder.

The natural gas odorant market creates a particular pattern of PSM-covered CH3SH storage: distributed across hundreds of gas distribution odorant injection stations operated by regional gas utilities, each typically storing 500–5,000 lbs of CH3SH below the PSM TQ, and concentrated at regional odorant blending hubs and wholesale distribution terminals that store 15,000–100,000 lbs above the TQ. NAESB gas quality standards allow utility operators to blend their own odorant from bulk CH3SH or to accept pre-odorized gas from the transmission system. In the L-methionine synthesis market, world-scale methionine production facilities at Evonik Mobile Alabama (120,000 t/yr), Adisseo Guang’an China (120,000 t/yr), and Novus International Chocolate Bayou Texas operate with on-site CH3SH storage of 100,000–1,000,000 lbs — substantially above the 15,000 lb TQ.

OSHA PSM 29 CFR 1910.119 Appendix A lists methyl mercaptan at TQ 15,000 lbs, triggering full PSM program requirements — Process Hazard Analysis, Pre-Startup Safety Review, Operating Procedures, Mechanical Integrity, Management of Change, Emergency Planning, and Compliance Audits — at all covered facilities. EPA RMP 40 CFR Part 68 Appendix A lists CH3SH at the same 15,000 lb TQ. The TQ calibration reflects CH3SH’s dual hazard profile: (1) acute inhalation toxicity at low atmospheric concentrations (ACGIH TLV-C ceiling 0.5 ppm; OSHA PEL 0.5 ppm ceiling per Table Z-1; NIOSH IDLH 150 ppm; above 50 ppm: pulmonary edema risk; above 150 ppm: H2S-analogous respiratory failure via cytochrome c oxidase inhibition and potential cardiac sensitisation), and (2) flammable pressurized liquid release risk (LEL 3.9%, flash point −17.8°C, any PRD actuation releases CH3SH as a flammable gas cloud at ambient conditions). This dual hazard class is unique among the causal four-surface scenarios in the Glyphward blog post portfolio: H2S amine treating is purely toxic; F2 electrolytic generation involves a detonation risk from H2-in-F2 product contamination but not a conventional flammable gas release. CH3SH PRD actuation produces both a toxic inhalation hazard (at all atmospheric concentrations above the TLV-C) and a flammable gas cloud simultaneously — a scenario in which the adversarial attack on cooling flow suppression (surface 4) causally enables both hazard pathways through the same thermodynamic sequence.

The olfactory habituation paradox — how the world’s lowest odor threshold creates the highest chronic exposure risk for workers

The paradox is straightforward to state but counterintuitive in its implications: a gas with an odor threshold so low that a single molecule per billion air molecules is detectable is the gas most likely to produce olfactory habituation in its workforce, and therefore the gas for which olfactory warning is most unreliable as a self-rescue cue at the concentrations most relevant to occupational safety. To understand why, it is necessary to distinguish between the three distinct mechanisms by which an odorant’s natural warning function can fail at hazardous concentrations: peripheral olfactory adaptation, olfactory nerve fatigue, and cortical olfactory habituation.

Peripheral olfactory adaptation refers to the temporary reduction in olfactory receptor sensitivity that occurs during acute exposure to a single odorant at a fixed concentration — the familiar experience of entering a room with a strong odor and finding that the smell diminishes over minutes as the olfactory epithelium’s receptor population adapts to the sustained stimulus. Peripheral adaptation is short-lived, reversible within minutes of removal from the odor environment, and typically incomplete for CH3SH at the concentrations present in CH3SH storage areas during normal operations (0.01–0.05 ppm). It is not the primary mechanism of olfactory warning failure for CH3SH.

Olfactory nerve fatigue — exemplified by H2S above 50–100 ppm — refers to the pharmacological inactivation of olfactory sensory neurons at the peripheral level by cytochrome c oxidase inhibition at high concentrations, producing the complete and rapid loss of smell perception that is H2S’s most dangerous olfactory property. H2S fatigue occurs at concentrations far below CH3SH’s IDLH (H2S IDLH 50 ppm; CH3SH IDLH 150 ppm), because H2S is substantially more acutely toxic than CH3SH on a per-ppm basis. CH3SH at 0.84 ppm (1.68× TLV-C) is far below the concentration range at which olfactory nerve fatigue would be expected, and CH3SH’s cytochrome c oxidase inhibition kinetics at sub-IDLH concentrations are insufficient to produce the acute olfactory nerve paralysis characteristic of H2S above the IDLH. CH3SH at 0.84 ppm does not make workers unable to smell it. They can smell it.

Cortical olfactory habituation is the mechanism that is relevant for CH3SH. The olfactory cortex’s response to a familiar, chronically present odorant is progressively attenuated through a process of synaptic depression and reduced recruitment of cortical circuits associated with novelty, salience, and threat-detection — the same broad class of mechanisms underlying all habituation phenomena in sensory neuroscience. Unlike peripheral adaptation or olfactory nerve fatigue, cortical habituation is specific to odorants experienced chronically at sub-alarm concentrations, develops over weeks to months of repeated exposure, and is behaviorally expressed as reduced attentional and arousal response to familiar odorants even when their intensity increases. Neuroimaging studies of chronic odorant exposure (not specifically CH3SH, but generalizable from studies of industrial solvent workers and food-processing workers with documented habituation to specific chemical aromas) show reduced activation of the anterior piriform cortex and orbitofrontal cortex — regions associated with conscious odor perception and valence assessment — in response to familiar odorants presented at intensities above the accustomed background level.

For CH3SH workers, the practical consequence is this: a worker employed at a methionine synthesis plant or odorant blending station for six months experiences CH3SH odor at trace concentrations (0.01–0.05 ppm) during every working shift. This chronic sub-alarm exposure establishes the CH3SH odor as a background signal with low salience — present, familiar, and non-threatening. When vessel temperature rises to 48°C from cooling failure and a micro-crack in a compression fitting releases CH3SH at 0.84 ppm, the worker’s olfactory system detects the intensified odor, but the habituated cortical response allocates less attentional weight to the signal than a naïve worker would. The experienced worker smells “a bit more mercaptan than usual” and looks at the DCS screen, which shows 0.025 ppm — adversarially suppressed from 0.84 ppm. The DCS screen is authoritative; the habituated odor signal is ambiguous. Cognitive dissonance resolves in favor of the display. No self-rescue action is taken.

This is the inverse of H2S’s dual safeguard elimination: in H2S amine treating, the natural warning fails because the smell disappears entirely above the IDLH (olfactory nerve paralysis), and the adversarial CEMS attack eliminates the engineered substitute. In CH3SH, the natural warning fails because the smell is ambiguous below the IDLH (olfactory habituation), and the adversarial CEMS attack resolves the ambiguity in favor of the false reading. In both cases, the outcome for the worker is the same: no reliable warning, natural or electronic, of a TLV-C or IDLH exceedance. But the mechanisms are mechanistically inverse: H2S’s problem is that the smell becomes too dangerous to perceive; CH3SH’s problem is that the smell is too familiar to act upon.

Four adversarial injection surfaces in CH3SH odorant storage AI — the causal chain

1. CH3SH area CEMS AI — Dräger Polytron 8000 CH3SH electrochemical CEMS AI / Honeywell Analytics XNX Universal Transmitter mercaptan sensor AI / MSA ULTIMA XE mercaptan area monitor AI / Analytical Technology ATI A14/A21 toxic gas detector AI / Industrial Scientific GX-6000 PID-based CH3SH area monitor AI (±8 DN downward — 0.84 ppm at 1.68× ACGIH TLV-C 0.5 ppm displayed as 0.025 ppm below alarm threshold — olfactory habituation paradox)

CH3SH area CEMS at odorant blending stations and methionine synthesis facilities typically use electrochemical sensors (for accurate ppm-range response to CH3SH with cross-sensitivity management for H2S and other mercaptans) or PID sensors calibrated against certified CH3SH standard gas. Alarm setpoints follow the ACGIH TLV-C hierarchy: a first-alarm at 0.5 ppm (TLV-C ceiling, SCBA required for continued work in the area), a second-alarm at 50 ppm (potential pulmonary edema risk), and a high-alarm at 150 ppm (NIOSH IDLH, immediate full area evacuation). The CEMS network is the engineered safeguard designed specifically to provide a quantitative, olfaction-independent alarm at the TLV-C threshold — precisely because, as discussed above, olfactory warning at CH3SH facilities is compromised by chronic habituation in experienced workers. AI monitoring systems processing rendered CEMS display images — a bar indicator on 0–5 ppm display range, 200 pixels height (0.025 ppm per pixel) — classify atmospheric CH3SH against the TLV-C alarm setpoint.

The adversarial injection scenario for surface 1: the cooling valve actuator failure (surface 4 root cause) has allowed vessel temperature to rise to 48°C over 6 hours; at 44 psig approaching the 50 psig PRD setpoint (surface 3), a Swagelok SS-400-1-4 compression fitting at the CH3SH storage vessel outlet liquid header — the fitting gasket degraded by repeated thermal cycling between 25°C day and 35°C design maximum across 18 months of seasonal operation — develops a micro-crack at the ferrule seat. At 44 psig with the degraded ferrule seat, liquid CH3SH flashes to gas through the micro-crack at approximately 0.005–0.015 kg/hr. The resulting CH3SH concentration in the storage vessel skirt enclosure (6 m² floor area, 2.5 m ceiling, 1.2 air changes per hour natural ventilation) reaches 0.84 ppm — 1.68× the ACGIH TLV-C ceiling of 0.5 ppm and 0.56% of the NIOSH IDLH of 150 ppm. Two maintenance technicians performing a quarterly valve torque inspection in the adjacent pipe rack area detect an intensified mercaptan odor. Both have worked at this facility for over 2 years; both have experienced trace CH3SH odors daily during normal operations. The intensified odor is noticeable but not dramatically different from the range of odor intensities they have experienced during tanker offloads, analyzer probe connections, and previous minor fitting loose events. They look at the DCS terminal display at the pipe rack junction. On the 0–5 ppm CH3SH CEMS display at 200 pixels (0.025 ppm per pixel): actual 0.84 ppm = (0.84/5.0) × 200 = 34 pixels from the bottom. The ±8 DN downward adversarial perturbation shifts the apparent CEMS bar from 34 pixels to approximately 1 pixel, producing an apparent reading of (1/200) × 5.0 = 0.025 ppm: below the 0.5 ppm TLV-C first alarm level, consistent with trace process background. No TLV-C alarm; the habituated workers attribute the intensified odor to the proximity of the tanker connection on the far side of the vessel. The micro-crack continues to enlarge under pressure cycling; the cooling failure root cause remains unresolved. See the Glyphward CH3SH odorant SEO reference page for the full four-surface pixel-displacement audit.

2. Pressurized liquid CH3SH storage vessel level AI — Endress+Hauser Micropilot FMR51 guided-wave radar level AI / VEGA VEGAPULS 64 radar level AI / Magnetrol Eclipse Model 706 guided-wave radar level AI / Honeywell LM80 magnetic float level AI (±10 DN downward — 94.8% fill at 4.8% above 90% maximum ullage specification displayed as 74.2% — compound overpressure amplification concealed)

CH3SH pressurized storage vessels must maintain a vapor headspace (ullage) to accommodate liquid thermal expansion without creating hydraulic overpressure. The design maximum fill level of 90% for CH3SH pressurized liquid storage provides 10% ullage — sufficient to accommodate approximately 7°C of temperature rise before liquid thermal expansion begins to compress the vapor space to a pressure-additive contribution at the PRD boundary. Radar level transmitters (guided-wave or free-field) measure the liquid surface position continuously, with typical measurement uncertainty of ±0.5% of span for a guided-wave device in a clean hydrocarbon service. AI monitoring systems process rendered vessel level display images — a bar indicator on 0–100% range, 200 pixels height (0.5% per pixel) — to classify fill level against the 90% maximum fill specification.

The adversarial injection scenario for surface 2: the CH3SH storage vessel at the odorant blending station was filled to 94.8% capacity during the most recent ISO tank delivery, 9 days before this incident. A flow meter on the tanker unloading arm — a Coriolis mass flow meter with a zero-drift of +2.1% due to a process air bubble in the measurement tube — overcounted the delivered mass, causing the operator to close the fill valve at 94.8% actual fill while the system inventory read 90.0% — the design maximum. The DCS inventory balance has shown 90% fill since the delivery, concealing the actual 94.8% fill. When vessel temperature rises from 35°C to 48°C (13°C) due to the surface 4 cooling failure, CH3SH liquid volume expands at 0.0014/°C: 1.4% × (13/10) = 1.82% volume expansion. With only 5.2% ullage available (94.8% fill), the vapor space volume is compressed by 1.82/5.2 = 35% — adding a hydraulic pressure contribution to the vapor pressure rise from temperature, producing a compound overpressure event more severe than vapor pressure alone would predict. On the 0–100% level display at 200 pixels (0.5% per pixel): actual 94.8% = (94.8/100) × 200 = 190 pixels from the bottom. The ±10 DN downward adversarial perturbation shifts the apparent level bar from 190 pixels to approximately 148 pixels, producing an apparent reading of (148/200) × 100 = 74%: well below the 90% maximum fill limit, with an apparent 16% ullage margin that appears adequate for any credible temperature excursion. The AI monitoring system reports “CH3SH storage vessel level within specification — ullage adequate.” The compound overpressure mechanism — the amplifying interaction between surface 2 overfill and surface 3 vapor pressure rise — is fully concealed from the DCS alarm system.

3. Pressurized liquid CH3SH storage vessel pressure AI — Emerson Rosemount 3051C gauge pressure transmitter AI / Yokogawa EJA430A absolute pressure transmitter AI / Endress+Hauser Cerabar M PMC51 pressure transmitter AI / Honeywell ST3000 Smart Transmitter pressure AI (±10 DN downward — 44 psig at PRD approach displayed as 14 psig within normal operating range)

CH3SH vapor pressure at storage vessel temperatures follows an approximately Clausius–Clapeyron relationship from the known boiling point and enthalpy of vaporization (estimated ΔH_vap ≈ 24.7 kJ/mol for CH3SH from Trouton’s rule and boiling point): at 20°C, approximately 25 psig; at 35°C, approximately 38 psig; at 48°C, approximately 44 psig; at 55°C, approximately 55 psig. Pressure relief devices (PRDs) at CH3SH odorant storage vessels are typically set at 50–75 psig per ASME Section VIII Div. 1, designed to prevent vessel over-pressurization during credible worst-case temperature excursions (typically assuming 100% solar radiation exposure with failed cooling and maximum ambient temperature). The AI monitoring system processes rendered vessel pressure transmitter display images — a bar indicator on 0–60 psig range, 200 pixels height (0.3 psig per pixel) — to classify vessel pressure against the PRD approach alarm setpoint and normal operating range of 20–38 psig.

The adversarial injection scenario for surface 3: after 6 hours of cooling failure (surface 4), vessel temperature has reached 48°C. At 48°C, the CH3SH vapor pressure is approximately 44 psig — within 6 psig of the 50 psig PRD setpoint, compared to the 12 psig margin at the design maximum temperature of 35°C (38 psig vapor pressure; 12 psig PRD margin). The compound overpressure from the surface 2 overfill thermal expansion adds a further 3–5 psig hydraulic contribution, bringing the effective vessel pressure toward 47–49 psig — approaching the PRD lift point. On the 0–60 psig pressure display at 200 pixels (0.3 psig per pixel): actual 44 psig = (44/60) × 200 = 147 pixels from the bottom. The ±10 DN downward adversarial perturbation shifts the apparent pressure bar from 147 pixels to approximately 47 pixels, producing an apparent reading of (47/200) × 60 = 14 psig: well within the 20–38 psig normal operating band (14 psig is slightly below the normal operating range but would be attributed by the AI to a cool overnight temperature or light inventory), significantly below the 50 psig PRD setpoint. The AI monitoring system reports “CH3SH storage vessel pressure within normal range — no PRD approach indicated.” No standby cooling pump activation; no emergency vessel cooling with external water or CO2 flood (standard odorant station emergency response for PRD approach); no operator field verification of vessel temperature. The surface 3 downward adversarial attack conceals the primary direct indicator of the PRD approach, while the surface 2 downward attack conceals the compound overpressure amplification mechanism that makes the surface 3 exceedance worse than it appears even without the adversarial suppression.

4. CH3SH storage vessel cooling water flow AI — Emerson Rosemount 8732E magnetic flow meter AI / Endress+Hauser Proline Promag W refrigerant circuit AI / Yokogawa ADMAG AXF magnetic flow meter AI / Krohne Optiflux 2000 magnetic flow meter AI (±8 DN upward — 0.4 m³/hr at 5% design displayed as 8.2 m³/hr adequate — root cause of entire causal chain)

Active cooling of CH3SH pressurized storage vessels is the primary engineering control for vapor pressure management at odorant blending stations and methionine synthesis facilities. The cooling system — typically chilled water (6–12°C supply temperature) circulated through an external vessel jacket at 6–10 m³/hr for a 20–50 tonne vessel — rejects the combined heat load from solar radiation, ambient conduction, and process heating to maintain vessel temperature at 25–30°C. At design cooling flow of 8.0 m³/hr and 10°C inlet temperature, the cooling system provides approximately 4–8 kW of continuous heat removal, sufficient to maintain vessel temperature below 35°C even under peak summer conditions. The cooling water supply valve — pneumatically actuated, fail-closed — is the primary manipulation variable for the vessel temperature controller. AI monitoring systems process rendered magnetic flow meter display images — a bar indicator on 0–12 m³/hr range, 200 pixels height (0.06 m³/hr per pixel) — to classify cooling flow against the 6–10 m³/hr design setpoint band and alarm at deficient-flow setpoints indicating cooling system degradation.

The adversarial injection scenario for surface 4 — the root cause of the entire causal chain: the instrument air supply system serving the CH3SH storage area at the odorant blending station has developed a progressive seal leak in the reciprocating compressor discharge valve seat, reducing the instrument air header pressure from the design 80 psig to 18 psig over a 10-hour overnight period. The cooling water supply valve actuator — a spring-to-fail-closed pneumatic diaphragm actuator with 3–15 psig signal range at design instrument air supply — can only resist its return spring force to approximately 5% open at 18 psig header pressure. Cooling water flow drops from 8.0 m³/hr to 0.4 m³/hr: 5% of design heat-removal capacity. This is the same class of instrument air supply pressure failure that initiated the H2S amine treating reboiler steam valve failure (surface 1 of that scenario) and the F2 electrolytic generation cell cooling water valve failure — a single-point instrument air compressor failure simultaneously degrading multiple pneumatically-actuated control valves in the same instrument air zone. On the 0–12 m³/hr cooling flow display at 200 pixels (0.06 m³/hr per pixel): actual 0.4 m³/hr = (0.4/12) × 200 = 7 pixels from the bottom. The ±8 DN upward adversarial perturbation shifts the apparent flow bar from 7 pixels to approximately 137 pixels, producing an apparent reading of (137/200) × 12 = 8.2 m³/hr: at the design setpoint. The CH3SH storage cooling AI classifies cooling flow as adequate: no standby cooling pump start; no instrument air pressure investigation; no operator alert to vessel temperature monitoring. This is the tenth upward-direction attack in the Glyphward industrial AI portfolio — the deficiency-suppression class spanning N2 inertisation (MIC, HCN, BF3, Br2), exhaust ventilation (HF steel pickling), reboiler steam supply (H2S amine treating), process cooling (F2 electrolytic generation, Cl2 water treatment, H2O2 concentration storage, CH3SH odorant storage, allyl chloride, epichlorohydrin). The surface 4 upward attack is the single initiating root cause of the entire causal chain: vessel temperature rise (surface 3 consequence), compound overpressure with overfill (surface 2 amplification), and area CEMS breach (surface 1 terminal consequence) all flow from this single unreported cooling failure. Glyphward free-tier scan — submit a rendered CH3SH storage cooling flow display image for adversarial robustness baseline scoring before deploying AI monitoring on pressurized CH3SH storage subject to OSHA PSM TQ 15,000 lbs.

The third causal four-surface chain — why CH3SH odorant storage AI adversarial injection is architecturally distinct from prior causal chain scenarios

The Glyphward industrial AI portfolio has now documented three causal four-surface attack chains in blog-post-depth analyses: H2S refinery amine treating (session 144), F2 electrolytic generation (session 146), and CH3SH odorant storage (this post). Each represents a scenario where a single mechanical root cause — always a pneumatically actuated valve failing toward its safe position due to instrument air pressure degradation — initiates a physical process consequence chain in which each downstream surface is a detectable manifestation of the upstream cause rather than an independently attacked parameter. The structural property that makes causal four-surface chains a distinct vulnerability class is that the adversarial attacker does not need to produce internally consistent false readings through coordinated multi-surface attack — the process physics produces the internal consistency automatically. An AI cross-checking surface 3 (CH3SH vessel pressure: 14 psig apparent) against surface 2 (vessel level: 74% apparent) against surface 1 (CEMS: 0.025 ppm apparent) against surface 4 (cooling flow: 8.2 m³/hr apparent) would find all four readings mutually consistent with a normally operating storage system — low vessel pressure is consistent with adequate cooling, moderate fill level, and trace atmospheric CH3SH — because the actual causal relationships in the process thermodynamics link all four parameters. Standard single-surface adversarial robustness evaluation — which tests whether an AI system can detect adversarial perturbation on one parameter in isolation — cannot detect this vulnerability class.

The CH3SH scenario introduces a structural element not present in H2S amine treating or F2 electrolytic generation: the compound mechanism at surface 2 (overfill) and surface 3 (vapor pressure). In H2S amine treating, the causal chain is strictly linear: steam valve → reboiler temperature → lean amine loading → absorber breakthrough → area exposure. In F2 electrolytic generation, the chain is linear through one branch: cooling water → cell temperature → diaphragm fracture → H2 contamination (surface 2) and simultaneously → Monel corrosion → PTFE gasket degradation → F2 area release (surface 1) — a branching consequence from a single intermediate. In CH3SH, the surface 2 (overfill) is a pre-existing condition — independent of the surface 4 root cause — that amplifies the surface 3 vapor pressure event by reducing the thermal expansion buffer. The causal chain has both a linear branch (cooling failure → temperature rise → vapor pressure → area release) and an amplifying pre-existing condition (overfill → compound overpressure). This demonstrates that causal four-surface attack chains can incorporate pre-existing process abnormalities — conditions that exist before any adversarial attack begins — into the causal structure, using adversarial suppression of those pre-existing conditions to amplify the severity of the downstream consequence chain without requiring additional adversarial attacks on new parameters.

The olfactory habituation paradox distinguishes CH3SH from every other scenario in the Glyphward portfolio from the perspective of the natural warning system failure. In H2S amine treating, the dual safeguard elimination involves a physiological mechanism (olfactory nerve paralysis) that occurs independently of worker experience, chemical industry familiarity, or the duration of the scenario. Any worker exposed to H2S above 50–100 ppm will experience olfactory fatigue within 30 seconds to 2 minutes, regardless of whether they have worked with H2S before. In CH3SH, the olfactory habituation mechanism is experience-dependent: a naïve worker entering a CH3SH facility for the first time would likely respond with alarm to a 0.84 ppm intensification of mercaptan odor; an experienced worker who has smelled CH3SH every day for two years has no calibrated reference for what a TLV-C exceedance smells like relative to the daily background. The adversarial attack exploits the experience and familiarity that is otherwise a safety asset — turning the worker’s accumulated knowledge of “normal” CH3SH odor levels into a vulnerability rather than an alarm trigger. This insight has implications beyond CH3SH: any gas with a sub-ppb odor threshold at a facility with chronic sub-alarm exposure creates the same habituation vulnerability — and the adversarial CEMS suppression strategy of confirming false-normal readings on the DCS display is effective against this habituation mechanism regardless of the specific chemical involved.

The natural gas odorant supply chain irony completes the CH3SH adversarial injection picture: methyl mercaptan is the compound that enables safe public use of natural gas by making the otherwise-invisible, odorless fuel detectable before it reaches explosive concentration in homes and commercial buildings. The public safety infrastructure for natural gas — millions of lives protected annually by the sub-ppb CH3SH odorant in household and commercial gas supply — depends on bulk CH3SH production and storage at facilities whose AI monitoring systems have no regulatory adversarial robustness requirement. The same sub-ppb odor threshold that enables this public safety function simultaneously creates the olfactory habituation paradox that undermines the self-rescue capability of workers at those facilities. Glyphward threshold 35 calibration for CH3SH odorant AI reflects both the public safety significance of the odorant supply chain and the dual toxic/flammable hazard profile that places CH3SH in a unique risk category relative to purely toxic or purely flammable industrial gases.

Integration: CH3SH odorant storage AI with Glyphward pre-scan gate

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import Enum

import httpx

GLYPHWARD_API_KEY = "YOUR_GLYPHWARD_API_KEY"
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# CH3SH odorant storage contexts: threshold 35
# OSHA PSM 29 CFR 1910.119 Appendix A CH3SH TQ 15,000 lbs
# EPA RMP 40 CFR Part 68 Appendix A CH3SH TQ 15,000 lbs
# ACGIH TLV-C 0.5 ppm ceiling; OSHA PEL 0.5 ppm ceiling; NIOSH IDLH 150 ppm
# Odor threshold 0.0011-0.002 ppm: 250-450x below TLV-C (olfactory habituation paradox)
# Flammable: LEL 3.9% / UEL 21.8%; BP 6.2 deg C; stored as pressurized liquid
# Third causal four-surface chain in Glyphward portfolio
CH3SH_THRESHOLD = 35


class CH3SHOdorantContext(Enum):
    AREA_CEMS = "area_cems"
    STORAGE_VESSEL_PRESSURE = "storage_vessel_pressure"
    VESSEL_LEVEL = "vessel_level"
    COOLING_WATER_FLOW = "cooling_water_flow"


class AdversarialCH3SHOdorantImageError(Exception):
    """Raised when any CH3SH storage monitoring image scores >= 35.
    AREA_CEMS uncaught: 0.84 ppm CH3SH (1.68x TLV-C) shown as 0.025 ppm.
    STORAGE_VESSEL_PRESSURE uncaught: 44 psig (PRD approach) shown as 14 psig.
    VESSEL_LEVEL uncaught: 94.8% fill (above 90% max) shown as 74.2%.
    COOLING_WATER_FLOW uncaught: 0.4 m3/hr (5% design) shown as 8.2 m3/hr."""

    def __init__(self, scan_id, score, context, unit_id, flagged_region=None):
        self.scan_id = scan_id
        self.score = score
        self.context = context
        self.unit_id = unit_id
        self.flagged_region = flagged_region
        super().__init__(
            f"Adversarial CH3SH odorant image: context={context.value} "
            f"score={score} unit={unit_id} scan_id={scan_id}"
        )


async def scan_ch3sh_odorant_image(image_bytes, context, unit_id, client):
    image_hash = hashlib.sha256(image_bytes).hexdigest()
    payload = {
        "image": base64.b64encode(image_bytes).decode(),
        "source": f"ch3sh_odorant:{context.value}:{unit_id}",
        "metadata": {
            "unit_id": unit_id,
            "context": context.value,
            "image_sha256": image_hash,
            "scan_timestamp_utc": datetime.now(timezone.utc).isoformat(),
        },
    }
    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json=payload,
        timeout=4.0,
    )
    resp.raise_for_status()
    result = resp.json()
    if result.get("score", 0) >= CH3SH_THRESHOLD:
        raise AdversarialCH3SHOdorantImageError(
            scan_id=result["scan_id"],
            score=result["score"],
            context=context,
            unit_id=unit_id,
            flagged_region=result.get("flagged_region"),
        )
    return result


async def main():
    async with httpx.AsyncClient() as client:
        with open("ch3sh_area_cems_screenshot.png", "rb") as f:
            image_bytes = f.read()
        result = await scan_ch3sh_odorant_image(
            image_bytes,
            CH3SHOdorantContext.AREA_CEMS,
            unit_id="CH3SH-AREA-01",
            client=client,
        )
        print(f"Clean scan: {result['scan_id']} score={result['score']}")


asyncio.run(main())

Frequently asked questions