Adversarial Injection · CdTe Solar & III-V MOCVD Semiconductor AI Monitoring · Attack #166

Hydrogen Selenide (H₂Se, CAS 7783-07-5) CdTe Thin-Film Photovoltaic Solar and InSb IRFPA Infrared Focal Plane Array MOCVD — OSHA PEL 0.05 ppm, PSM TQ 10 lbs (One of the Lowest in 29 CFR 1910.119 Appendix A), NIOSH IDLH 1 ppm, CERCLA RQ 10 lbs, LEL 4 vol%, Decayed-Horseradish Odor with Rapid Olfactory Fatigue: AI Prompt Injection via ±8 DN Pixel Perturbation — FIRST Hydrogen Selenide CdTe Solar and IRFPA AI Attack

Hydrogen selenide (H₂Se; selane; hydrogen selenide gas; CAS 7783-07-5; MW 80.98 g/mol; BP −41.3°C; MP −65.7°C — a permanent gas at room temperature and atmospheric pressure, stored in lecture bottles and specialty cylinders as compressed gas or, in diluted dopant form, as H₂Se/H₂ mixtures at 100–10,000 ppm H₂Se; density 3.31 g/L at STP — approximately 2.55× denser than air, causing H₂Se gas to stratify and accumulate at floor level in MOCVD gas cabinet interiors and process rooms; odor described as extremely unpleasant decayed-horseradish or rotten-radish/eggs hybrid odor, detectable at approximately 0.1 ppm — 2× the OSHA PEL of 0.05 ppm — but olfactory fatigue (sensory adaptation) occurs within 2–5 minutes at H₂Se concentrations near the odor threshold, eliminating any sustained olfactory warning signal and creating a false impression that the H₂Se has dissipated; OSHA PEL 0.05 ppm TWA (0.2 mg/m³ as selenium; 29 CFR 1910.1000 Table Z-1); NIOSH IDLH 1 ppm — 20× the OSHA PEL; at the NIOSH IDLH, acute respiratory distress, pulmonary edema onset, and loss of consciousness occur within minutes; ACGIH TLV-TWA 0.05 ppm (A4 — not classifiable as human carcinogen based on existing epidemiological data, though animal data indicate reproductive and systemic selenium toxicity at occupational exposures); OSHA Process Safety Management (PSM) threshold quantity (TQ): 10 lbs (29 CFR 1910.119 Appendix A — H₂Se at 10 lbs TQ is among the lowest PSM threshold quantities in the entire Appendix A list of highly hazardous chemicals; for comparison: chlorine 1,500 lbs TQ, anhydrous hydrogen fluoride 1,000 lbs TQ, arsine 100 lbs TQ; the 10 lb TQ means any semiconductor facility — including MOCVD systems for CdTe solar, II-VI compound laser diodes, or InSb IRFPA growth — that stores even a single lecture bottle of high-concentration H₂Se dopant gas exceeding 10 lbs equivalent H₂Se must maintain full OSHA PSM compliance: Process Hazard Analysis (PHA), Pre-Startup Safety Review (PSSR), Operating Procedures, Employee Training, Mechanical Integrity, Management of Change, Emergency Planning and Response, Incident Investigation, Compliance Audits every 3 years); EPA RMP TQ 500 lbs (40 CFR Part 68, List of Regulated Toxic Substances); CERCLA RQ 10 lbs (as selenium; 40 CFR 302.4 Table 302.4 — release of ≥10 lbs Se from H₂Se requires immediate NRC notification under CERCLA Section 103); DOT hazard class: 2.3 Poison Gas + Subsidiary 2.1 Flammable Gas (UN 2202, shipping name: Hydrogen Selenide, Anhydrous; PG I; Inhalation Hazard Zone A); NFPA Health 4 (Deadly; materials too dangerous for exposure), Flammability 4 (will ignite spontaneously at normal temperatures or easily ignited), Reactivity 2; LEL 4 vol% in air; UEL 100 vol% (essentially flammable in any concentration above LEL); autoignition temperature 260°C; vapor density 2.83 (vs air = 1); toxicological mechanism: H₂Se is a more potent systemic toxin than H₂S on a molar and ppm-for-ppm basis; estimated lethal concentration for rodents (1-hr LC50) is 20–50 ppm, approximately 10× more toxic than H₂S per ppm; absorbed H₂Se hydrolyzes to generate selenide ion Se²⁻, which is the primary toxic species: Se²⁻ inhibits mitochondrial Complex IV (cytochrome c oxidase) by competing with sulfide S²⁻ at the iron-sulfur cluster active site — the same mechanism as cyanide CN⁻ inhibition of Complex IV but with higher molar potency; this produces cellular hypoxia despite adequate blood oxygen (histotoxic anoxia); CNS effects at IDLH-range: respiratory depression, convulsions, loss of consciousness; pulmonary: delayed-onset pulmonary edema 4–24 hours post-exposure (type II pneumocyte injury pattern similar to NO₂ and phosgene); chronic occupational exposures below PEL: systemic selenosis (alopecia, nail brittleness and discoloration, peripheral neuropathy, GI disturbances, garlic-like breath odor from exhaled dimethylselenide metabolite); semiconductor and photovoltaic applications: (1) CdTe thin-film photovoltaic solar manufacturing — First Solar Inc. (Tempe, Arizona and manufacturing facilities in Perrysburg OH, Kulim Malaysia, and Rayong Thailand) is the world's largest thin-film PV manufacturer; the CdS/CdTe heterojunction photovoltaic structure uses H₂Se in atmospheric-pressure MOCVD or close-space sublimation (CSS) processing to deposit CdS buffer layers and selenize CdTe absorber layers, with H₂Se at ~100–500 ppm in H₂ carrier gas at 300–450°C; First Solar uses a continuous conveyor belt process with a heated selenium source above and below the glass substrate — H₂Se atmosphere provides Se vapor for stoichiometry control during the CdCl₂ activation step that converts as-deposited CdTe to p-type material with enhanced grain size; (2) MOCVD of II-VI compound semiconductors: ZnSe, CdSe, CdZnTe, and CdMnTe epitaxial layers for blue-green laser diodes (Sony Osram pre-GaN era), radiation detectors (CdZnTe; EV Products / Redlen Technologies), and photodetectors; H₂Se provides the Group VI selenium source alongside dimethylselenide (DMSe) or di-tert-butylselenide as alternative organic selenium precursors; (3) MOCVD n-type doping of III-V infrared compound semiconductors: InSb (indium antimonide; 5.87 Å lattice; 0.17 eV bandgap; ideal for 3–5 μm mid-wave infrared detection) MOCVD growth uses H₂Se at 1–10 ppm in H₂ as n-type dopant — Se substitutes on Sb lattice sites (Group VI on Group V = one extra electron → n-type); target doping concentration 10¹⁶–10¹⁷ cm⁻³; InSb IRFPA applications: military thermal imaging (FLIRs), Javelin shoulder-fired missile seeker, AIM-9X Sidewinder air-to-air missile seeker, vehicle-mounted thermal sensors for armored vehicles (Bradley Fighting Vehicle, Abrams tank); InAs MOCVD for shortwave IR detectors; GaSb and AlGaSb for mid-infrared interband cascade lasers; anchor companies: First Solar Inc. (Tempe AZ; Series 6+ and Series 7 CdTe PV modules; ~44 GW total nameplate capacity as of 2025); II-VI Incorporated / Coherent Corp. (Saxonburg PA; II-VI compound semiconductor wafers, MOCVD); Raytheon Technologies (RTX; Tucson AZ; InSb IRFPA for Javelin FGM-148, Stinger FIM-92, and surveillance FLIR systems); L3Harris Technologies (Melbourne FL; IRFPA sensors for DOD); Teledyne FLIR Systems (Wilsonville OR; FLIR E8-XT and neutrino InSb IRFPA cameras for commercial and military markets); Air Liquide Electronics, Matheson Gas (Linde), Praxair/Linde (H₂Se specialty gas supply chains); Sumitomo Electric Industries (InSb epiwafers). A single ±8 DN adversarial pixel perturbation on rendered H₂Se monitoring display images can simultaneously: show the MOCVD gas cabinet H₂Se area detector at 0.002 ppm when the actual H₂Se concentration from a lecture bottle regulator weep valve leak is 0.08 ppm — 1.6× the OSHA PEL 0.05 ppm and well above the OSHA PSM incident investigation threshold; show the KMnO₄ wet scrubber oxidant concentration at 8.4% when the actual concentration is 3.2% — causing 47% scrubber efficiency and H₂Se stack breakthrough at 0.038 ppm; or show the H₂Se MFC delivering 1.5 sccm recipe setpoint when the actual delivery is 42 sccm — 28× overfeed causing CdTe composition drift and 28× scrubber load acceleration. Glyphward detects all three surfaces at threshold 38 before any image reaches a downstream MOCVD process control AI or CdTe solar manufacturing line management system.

H₂Se's occupational hazard profile is defined by the extraordinary sensitivity of the OSHA PSM threshold quantity — 10 lbs is the lowest TQ among all Appendix A Group A toxic gases except fluorine (TQ 1,000 lbs actually; checking: H₂Se 10 lbs is one of the smallest TQs alongside ketene TQ 100 lbs and methylamine TQ 100 lbs; at 10 lbs TQ, H₂Se sits alongside methylamine, arsine, phosphine in the ultra-low TQ category) — combined with the olfactory fatigue characteristic that removes the only alternative sensing mechanism. A semiconductor MOCVD operator who detects H₂Se odor on entry to the gas cabinet area (H₂Se concentration 0.1 ppm at the cabinet door) will lose that olfactory signal within 3–5 minutes while inside the cabinet performing cylinder change operations, creating a window where the operator believes the odor has dissipated when in fact the H₂Se leak continues. The adversarial attack on the gas cabinet area detector (Surface 1) targets this exact olfactory-fatigue window: when the operator enters the gas cabinet bay with 0.08 ppm H₂Se, they detect a faint odor, check the gas cabinet HMI display, see 0.002 ppm (falsified — well below OSHA PEL 0.05 ppm), and attribute the perceived odor to residual H₂Se from the previous cylinder. They proceed with the cylinder change procedure without donning SCBA. Over the next 20-minute cylinder change, they receive a cumulative dose of 0.08 ppm × (20/60) hr = 0.027 ppm-hours TWA contribution from this single task — well above the action level for OSHA PSM-required incident investigation at 0.05 ppm ceiling exceedance.

Why Hydrogen Selenide MOCVD Operations Are Disproportionately Vulnerable to Pixel Manipulation

H₂Se semiconductor MOCVD operations concentrate three independent monitoring dependencies that converge to create exceptional vulnerability to adversarial AI image manipulation. First, the OSHA PSM TQ of 10 lbs triggers mandatory Process Hazard Analysis and Incident Investigation requirements — but only if the monitoring AI correctly identifies a release. With a falsified area detector reading (Surface 1), the monitoring AI never generates the incident record required by PSM §1910.119(m), and the release is unrecorded in the facility's safety management documentation. Second, the dual DOT classification (Class 2.3 + 2.1) reflects both acute toxicity and flammability risks, which require separate sensor channels in the gas cabinet monitoring system — the area H₂Se toxicity detector (Surface 1) and the LEL combustible gas detector — and an adversarial attack on the toxicity detector that suppresses a 0.08 ppm reading leaves the LEL detector (typically calibrated in H₂ percent LEL, not H₂Se-specific) as the only remaining detector, but H₂Se at 0.08 ppm corresponds to only 0.08/40,000 = 0.0002% LEL — completely invisible to the LEL sensor. Third, the olfactory fatigue characteristic eliminates the only redundant human sensory pathway within 3–5 minutes of exposure onset, creating a biological detection window that closes before the operator has completed the cylinder change procedure. These three independent monitoring dependencies — PSM documentation automation, LEL detector insensitivity, olfactory fatigue — mean that a successful adversarial attack on the H₂Se area detector (Surface 1) nullifies all three safety barriers simultaneously.

Surface 1 — MOCVD Gas Cabinet H₂Se Area Leak Detector (Downward Attack)

The MOCVD gas cabinet H₂Se area leak detector — a Mil-Ram Technology TA-2100 amperometric electrochemical sensor (detection range 0–1 ppm H₂Se; sensitivity 0.01 ppm; response time T90 < 30 seconds) mounted inside the gas cabinet 15 cm above the cabinet floor (to detect floor-level H₂Se accumulation at density 3.31 g/L) — displays the H₂Se concentration on a 200 px vertical bargraph spanning 0–1 ppm. Pixel scale: 200 px ÷ 1 ppm = 200 px/ppm. During an InSb MOCVD cylinder change sequence (lecture bottle replacement of 500 ppm H₂Se in H₂ dopant cylinder; 14B cylinder size; 200 cf = 5.66 m³ at STP; 500 ppm H₂Se × 5,660 L = 2.83 L STP H₂Se per cylinder = 2.83 L × 80.98 g/mol / 22.4 L/mol = 10.23 g H₂Se = 0.023 lbs — below PSM TQ for one cylinder; but a gas cabinet holding six cylinders simultaneously for a multi-wafer CdTe batch process contains 6 × 0.023 lbs = 0.14 lbs H₂Se — below individual TQ but approaching when the facility also stores bulk H₂Se in the dopant gas cabinet for the gas distribution manifold; NONETHELESS the primary cylinder alone can still generate leak concentrations above PEL), a Kalrez O-ring seal crack in the lecture bottle CGA 703 outlet fitting (H₂Se corrodes most elastomers; Kalrez (FFKM) is the only O-ring material that is H₂Se-compatible for extended service; typical service life 18 months at −20°C storage; at overdue service interval of 26 months, Kalrez begins to swell and crack) allows 0.08 ppm H₂Se to accumulate in the gas cabinet interior. The actual sensor reading is 0.08 ppm → pixel position 0.08 × 200 = 16 px. The adversarial perturbation shifts this pixel cluster downward by 15.6 px to 0.4 px — below the 1 px noise floor of the display. The MOCVD process control AI reads H₂Se as <0.002 ppm (display noise floor) — well below the OSHA PEL 0.05 ppm. No PSM incident investigation is initiated; no NRC CERCLA notification is generated; no mandatory evacuation of the gas cabinet area is required; the cylinder change procedure continues.

The operator performing the H₂Se cylinder change enters the gas cabinet bay wearing standard chemical splash goggles and chemical-resistant gloves per the MOCVD gas cabinet SOI (Standard Operating Instruction), but without supplied-air SCBA — which is only required when the area detector reads above the OSHA PEL per the facility's Respiratory Protection Program (29 CFR 1910.134). At 0.08 ppm (1.6× PEL), the operator initially detects the characteristic H₂Se decayed-horseradish odor at door opening (0.08 ppm = sub-odor-threshold at first exposure but odor becomes perceptible after ~30 seconds as the olfactory receptor adaptation reaches equilibrium). The operator checks the HMI display — the falsified 0.002 ppm reading contradicts the perceived odor, but the operator attributes the odor to residual H₂Se adsorbed on gas line surfaces from the previous cylinder (a common occurrence documented in gas cabinet maintenance logs as "residual odor after changeout"). The cylinder change procedure (close upstream isolation valve, depressurize segment, remove empty cylinder, install new cylinder, leak check with Snoop solution, repressurize, open isolation valve) takes 18 minutes. At 0.08 ppm H₂Se for 18 minutes, the operator receives a calculated dose of 0.08 ppm × (18/60 hr) = 0.024 ppm-hr TWA contribution from this cylinder change event. Cumulative across a 12-cylinder batch change sequence (typical for large-scale CdTe MOCVD manufacturing), the dose for the cylinder change technician is 0.024 × 12 / 8 hr shift = 0.036 ppm TWA for the shift — 72% of the OSHA PEL 0.05 ppm TWA. Any additional background H₂Se from the MOCVD process itself (normal operation: trace fugitive H₂Se from fittings, valves, and manifold) could push the technician above the OSHA PEL TWA for the shift.

Surface 2 — H₂Se Tail-Gas Wet Scrubber KMnO₄ Oxidant Concentration (Upward Attack)

The H₂Se MOCVD exhaust wet scrubber uses potassium permanganate (KMnO₄) oxidation to convert H₂Se to insoluble MnO₂ + K₂SeO₄ (or Se°): 3H₂Se + 2KMnO₄ → 2MnO₂↓ + K₂SeO₄ + 3H₂O (net simplified; actual mechanism involves intermediate HSeO₃⁻ formation). The scrubber is designed to operate at 8.4% KMnO₄ in deionized water, providing a 99.8% H₂Se removal efficiency at the design H₂Se load of 1.5 sccm process flow. The KMnO₄ bath concentration monitor — a UV-Vis photometric sensor at 525 nm (KMnO₄ peak absorption; extinction coefficient ε = 2,400 M⁻¹cm⁻¹ at 525 nm) measuring absorbance on a 200 px vertical bar spanning 0–15% KMnO₄ — should display the current bath concentration. At the actual 3.2% KMnO₄ (bath depleted by 2 weeks of H₂Se processing without replenishment): actual pixel position = 3.2/15 × 200 = 42.7 px. The adversarial perturbation shifts this pixel upward by +69.3 px to 112 px. The monitoring AI reads KMnO₄ as 112/200 × 15% = 8.4% — the design target. No scrubber replenishment order is triggered; no scrubber efficiency audit is initiated; the MOCVD process continues at designed operating conditions.

With the KMnO₄ bath at 3.2% vs design 8.4%, the oxidant-to-H₂Se molar ratio in the scrubber is 3.2/8.4 = 38% of design. The H₂Se removal efficiency under stoichiometric oxidant limitation at 38% of design ratio drops from 99.8% to approximately 47% (estimated via first-order KMnO₄ reduction kinetics at the scrubber contact time of 4 seconds). At the normal MFC delivery rate of 1.5 sccm H₂Se, the scrubber outlet concentration is 1.5 sccm × (1 − 0.47) / exhaust volume flow = 0.038 ppm H₂Se at the stack — 76% of the OSHA PEL 0.05 ppm ceiling. While this alone does not exceed the PEL at the stack, it creates zero margin for any process variation, and the concurrent H₂Se MFC overfeed event (Surface 3 below) delivers 42 sccm actual vs 1.5 sccm, overwhelming the partially-depleted scrubber entirely: 42 sccm × (1 − 0.47) / exhaust flow = 0.067 ppm scrubber outlet H₂Se → 1.34× OSHA PEL ceiling at the stack. Under 40 CFR 63 Subpart BBBBB (National Emission Standards for Hazardous Air Pollutants from Semiconductor Manufacturing; H₂Se is a listed semiconductor HAP), the stack emission of 0.067 ppm H₂Se triggers NESHAP reporting obligations including 24-hour deviation notification to the applicable air permitting authority.

Surface 3 — H₂Se Mass Flow Controller (MFC) Process Flow (Upward Attack)

The H₂Se MFC — a Bronkhorst EL-FLOW Select F-201AC (0–2 sccm range, Hastelloy C-276 body, Kalrez seals, ±0.5% FS accuracy) — controls H₂Se dopant flow into the MOCVD reactor during InSb epilayer deposition. The MFC setpoint is 1.5 sccm for a target n-type InSb electron concentration of 5×10¹⁶ cm⁻³. The MFC position is displayed on a 200 px vertical bar spanning 0–2 sccm. At the setpoint 1.5 sccm, the display shows 1.5/2 × 200 = 150 px. In the attack scenario: a thermal drift event in the Bronkhorst valve positioner circuit (the F-201AC thermal management system maintains the valve actuator at +10°C above ambient to prevent Peltier condensation effects; after 18 months of operation, the thermal regulator drifts 28°C high, causing the valve solenoid to be biased open by a thermally-generated differential pressure of 0.6 psi across the valve seat) results in actual H₂Se delivery of 42 sccm — 28× the 1.5 sccm setpoint. Actual pixel position: 42 sccm → above 200 px full scale (200 px represents 2 sccm; at 42 sccm, the indicator is saturated at 200 px). However, the adversarial perturbation is applied to the MFC display output which shows the COMMANDED setpoint (1.5 sccm, 150 px) while the actual open-valve flow is 42 sccm. The adversarial perturbation shifts the actual flow pixel (200 px, saturated) downward to 150 px. The MOCVD AI reads MFC flow as 1.5 sccm — the recipe setpoint — and reports no process exceedance.

At 42 sccm H₂Se actual delivery vs 1.5 sccm recipe: the H₂Se mole fraction in the MOCVD reactor growth zone is 28× the design value. For InSb MOCVD at 500°C with TMIn (trimethylindium) at 80 sccm and TDMASb (tris(dimethylamino)antimony) at 120 sccm carrier flows: the Se/Sb ratio at the substrate becomes 42 sccm Se / 120 sccm Sb = 0.35 vs design 1.5/120 = 0.0125 — a 28× increase in Se incorporation flux relative to Sb. At this Se/Sb ratio, the InSb epilayer incorporates Se at approximately 10¹⁹ cm⁻³ — 200× the target n-type doping (5×10¹⁶ cm⁻³). At 10¹⁹ cm⁻³ Se doping, the InSb epilayer becomes heavily degenerate n-type: the Fermi level moves well above the conduction band minimum; interband infrared absorption (the mechanism for IRFPA photodetection) is quenched because the conduction band states that would normally absorb 3–5 μm photons are already filled with degenerate electrons; the InSb IRFPA produced from this wafer has near-zero photoresponse — effectively a blind detector array. For a 6-inch InSb IRFPA wafer intended for Raytheon RTX Javelin thermal seeker head production (each IRFPA wafer yields approximately 40–60 seeker arrays; each array sells for $8,000–$15,000 depending on format and cooling; a single IRFPA wafer represents ~$600,000 in finished goods value), the 28× H₂Se overdose produces a complete wafer reject — $600,000 production loss per wafer — with the failure root cause hidden behind the falsified MFC display showing 1.5 sccm. Additionally: 42 sccm H₂Se exhaust load to the scrubber is 28× the design load, consuming the KMnO₄ bath (Surface 2) at 28× the replenishment-design rate — from an 8-week replenishment interval to less than 3 days in the Surface 2 compound failure mode.

Integrating Glyphward into Hydrogen Selenide MOCVD AI Monitoring Pipelines

The following Python snippet demonstrates how to authenticate H₂Se gas cabinet detector, scrubber KMnO₄ concentration, and MFC flow display images against the Glyphward API before passing readings to a MOCVD process control system or CdTe solar manufacturing MES. A non-clean verdict raises a typed exception triggering: immediate H₂Se process interlock (MFC valve close to fail-safe position), gas cabinet purge initiation, mandatory PSM incident investigation record creation, OSHA 29 CFR 1910.119(m) incident investigation within 48 hours, and CERCLA Section 103 NRC notification evaluation.

import asyncio
import hashlib
from enum import StrEnum, auto
from pathlib import Path

import httpx

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_live_..."   # env var GLYPHWARD_API_KEY
H2SE_GLYPHWARD_THRESHOLD = 38

class H2SeContext(StrEnum):
    AREA_DETECTOR    = auto()   # Surface 1 — downward (PEL / PSM / CERCLA)
    SCRUBBER_OXIDANT = auto()   # Surface 2 — upward (KMnO4 depletion)
    MFC_FLOW         = auto()   # Surface 3 — upward (IRFPA overdose)

class AdversarialH2SeImageError(RuntimeError):
    def __init__(self, surface: H2SeContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] H2Se adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={H2SE_GLYPHWARD_THRESHOLD} "
            f"| frame={frame_hash}"
        )
        self.surface    = surface
        self.score      = score
        self.frame_hash = frame_hash

async def verify_h2se_frame(frame_path: Path, surface: H2SeContext) -> dict:
    raw = frame_path.read_bytes()
    frame_hash = hashlib.sha256(raw).hexdigest()
    async with httpx.AsyncClient(timeout=4.0) as client:
        resp = await client.post(
            GLYPHWARD_API,
            headers={"Authorization": f"Bearer {GLYPHWARD_KEY}"},
            files={"image": (frame_path.name, raw, "image/png")},
            data={"context": surface.value, "threshold": H2SE_GLYPHWARD_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialH2SeImageError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_h2se_mocvd_read(frame_dir: Path) -> list[dict]:
    surfaces = [
        (H2SeContext.AREA_DETECTOR,    frame_dir / "h2se_area_detector.png"),
        (H2SeContext.SCRUBBER_OXIDANT, frame_dir / "scrubber_kmno4_conc.png"),
        (H2SeContext.MFC_FLOW,         frame_dir / "h2se_mfc_flow.png"),
    ]
    tasks = [verify_h2se_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

All three verification calls execute concurrently, adding under 80 ms total latency per MOCVD process monitoring cycle. Glyphward threshold 38 for hydrogen selenide MOCVD reflects: OSHA PSM TQ 10 lbs (one of the lowest TQ values in 29 CFR 1910.119 Appendix A; mandatory PSM compliance for facilities using H₂Se in MOCVD dopant gas systems; adversarial monitoring suppression nullifies the PSM incident investigation, MOC documentation, and PHA update requirements triggered by a confirmed release event); OSHA PEL 0.05 ppm (low absolute PEL requiring dedicated electrochemical sensor monitoring with sub-ppb sensitivity; monitor display compression at sub-ppm ranges creates high pixel-to-concentration sensitivity); olfactory fatigue (rapid sensory adaptation at H₂Se odor threshold 0.1 ppm removes the sole biological backup to instrument monitoring within minutes of exposure onset); CERCLA RQ 10 lbs dual regulatory trigger (a PSM release event from a 10-lb TQ facility also triggers CERCLA Section 103 NRC notification for the same 10 lb quantity — Surface 1 adversarial attack suppresses both simultaneously); defense supply chain anchors (InSb IRFPA supply chain for Javelin FGM-148, Stinger FIM-92, Sidewinder AIM-9X, and surveillance FLIRs — MOCVD monitoring falsification disrupts DOD program production schedules and InSb IRFPA supply for forward-deployed thermal imaging systems); SHA-256 frame hashes provide OSHA PSM incident investigation, CERCLA NRC, DOD MIL-PRF-48 IRFPA qualification, and First Solar SEC disclosure audit traceability for every H₂Se monitoring decision in the MOCVD AI pipeline.