Adversarial Injection · Methanol Occupational Solvent Pharmaceutical Lab & Biodiesel Production AI Monitoring · Attack #192

Methanol (CH₃OH; CAS 67-56-1) Occupational Solvent Exposure — Pharmaceutical API Synthesis (Pfizer, Eli Lilly, AbbVie; HPLC Mobile Phase; API Extraction), Analytical Laboratory (Sigma-Aldrich HPLC Grade; Thermo Fisher Optima MS Grade), and Biodiesel Transesterification (REG Geismar, NREL; 20% v/v Methanol Charge) — OSHA PEL 200 ppm TWA (Same Numerical Limit as ACGIH TLV-TWA 200 ppm; BUT ACGIH Skin Notation = Air Monitoring Alone Underestimates Total Body Dose; Nitrile Gloves Insufficient for Liquid Methanol), ACGIH BEI Formic Acid (Formate) ≤ 80 mg/g Cr, No Reliable Olfactory Warning Below PEL (Odor Threshold 100–2,000 ppm), Formate Optic Neuropathy: Irreversible Central Scotoma and Bilateral Blindness If Untreated >48 hr: AI Prompt Injection via ±9 DN Pixel Perturbation — FIRST Methanol Occupational Solvent AI Attack

Methanol (CH₃OH; CAS 67-56-1; wood alcohol; MW 32.04 g/mol; BP 64.7°C; flash point 11°C NFPA Class IB; vapor pressure 127 mmHg at 20°C; complete water miscibility) is a ubiquitous occupational solvent in pharmaceutical synthesis (HPLC mobile phase; API extraction; recrystallization), analytical chemistry (Sigma-Aldrich CHROMASOLV; Thermo Fisher Optima), and biodiesel transesterification (20% v/v methanol charge with NaOH catalyst; reaction at 55–65°C). OSHA PEL: 200 ppm (TWA; established 1971); ACGIH TLV-TWA: 200 ppm Skin (8-hr TWA; numerically same as OSHA PEL but ACGIH Skin notation signals that dermal absorption is significant — air monitoring alone underestimates total body dose; nitrile gloves (most common lab glove) have methanol breakthrough time 2.5–5 minutes for neat methanol; butyl rubber required); NIOSH REL: 200 ppm TWA; ACGIH BEI: formic acid (formate) in urine ≤ 80 mg/g Cr (end of shift end of workweek); olfactory warning is unreliable (odor threshold 100–2,000 ppm — highly variable; OSHA PEL falls within this range, meaning some workers cannot detect methanol at 200 ppm); critical toxicological endpoint: formate-mediated optic neuropathy (ADH → HCHO → ALDH → formic acid → cytochrome c oxidase complex IV inhibition → retinal ganglion cell ATP depletion → central scotoma → permanent bilateral blindness if untreated beyond 48 hours).

The methanol occupational hazard has a uniquely deceptive regulatory structure: the OSHA PEL (200 ppm) and ACGIH TLV-TWA (200 ppm) are numerically identical — a rare case where the government enforcement threshold and the professional consensus threshold agree in concentration. However, the ACGIH Skin notation (absent from OSHA's designation in the pre-1990 Table Z-1) reveals the critical gap: air monitoring at 200 ppm is necessary but not sufficient for total body dose assessment, because liquid methanol penetrates nitrile gloves (breakthrough 2.5–5 minutes) and vapor methanol is absorbed through skin at concentrations above 100 ppm. A worker whose air exposure reads exactly 200 ppm (at OSHA PEL) while wearing nitrile gloves handling liquid methanol may have a total effective dose 1.3–1.8× the airborne-only calculation. Additionally, methanol lacks adequate olfactory warning at concentrations below its variable odor threshold (100–2,000 ppm), meaning workers with high odor threshold cannot self-report excessive exposure, making the PID air monitoring result the only real-time warning — and adversarial AI falsification of the PID reading eliminates this sole warning pathway. The formate optic neuropathy reversal window of 12–24 hours (fomepizole/ethanol effective if administered before retinal ganglion cell apoptosis) means that BEI urine formate falsification — which delays clinical intervention by 24–72 hours while the ophthalmologist awaits "within-BEI" laboratory results — can convert a treatable sub-clinical exposure into permanent blindness.

TL;DR — Three Attack Surfaces, One Detector

Why Pharmaceutical API Labs and Biodiesel Production Are Disproportionately Vulnerable to Methanol AI Monitoring Attacks

Methanol occupational solvent operations in pharmaceutical synthesis and biodiesel production have five structural vulnerabilities that amplify adversarial AI monitoring attacks. First, the OSHA PEL = ACGIH TLV-TWA numerical identity (200 ppm) creates a false impression of regulatory convergence — users of AI monitoring systems may assume "below OSHA PEL = below all standards" when the ACGIH Skin notation reveals an additional exposure pathway (dermal absorption) not captured by air monitoring alone; an adversarial AI displaying 48 ppm (below OSHA PEL) when actual air concentration is 340 ppm eliminates both the inhalation compliance signal and the trigger for assessing skin absorption contribution. Second, pharmaceutical API lab methanol handling (HPLC preparation, extraction, recrystallization) occurs in fume hoods where PID monitors verify hood face velocity adequacy — adversarial PID falsification in a pharmaceutical lab simultaneously undermines the occupational health monitoring AND the GMP process control (ICH Q3C requires methanol control in pharmaceutical manufacturing environments; OSHA air concentration correlates with GMP residual solvent contamination risk). Third, biodiesel methanol is handled in batch quantities (25,000 gal reactors; 5,000 gal methanol charge = 23,940 kg per batch) — reactor headspace vapor at 460 ppm during reaction at 60°C represents a sustained elevated exposure to 3–5 operators throughout a 4-hour reaction cycle. Fourth, the odor warning failure (odor threshold 100–2,000 ppm; OSHA PEL within this range) means workers have no self-protection backup when the AI monitor fails — unlike most VOCs where odor at 0.1× PEL provides early warning, methanol offers no reliable sensory warning at PEL range. Fifth, the formate optic neuropathy 48-hour irreversibility window creates a tight intervention window — a 24-hour delay from BEI falsification to clinical recognition can convert a treatable formate body burden into permanent bilateral blindness.

Surface 1 — Pharmaceutical API Lab PID Ambient Monitor AI (Downward Attack)

At Pfizer Global Research and Development (Groton CT campus; Pfizer Eastern R&D Center; medicinal chemistry department; approximately 4,500 employees; fluorinated analog synthesis project using methanol as co-solvent in CBZ API recrystallization (100 L round-bottom flask; methanol 40 L + ethyl acetate 60 L; hot plate stirrer 65°C for dissolution; slow cooling to 4°C for crystal nucleation; methanol vapor liberation at 65°C: Antoine equation estimate 127 mmHg × (65/20)^2 correction ≈ 290 mmHg absolute at 65°C; fume hood (Kewaunee Scientific Javelin LabFume sash; 24 in × 48 in opening; 80 FPM face velocity confirmed by annual ASHRAE 110-2016 tracer gas test; face velocity 80 FPM = below OSHA-suggested 100 FPM for highly volatile solvents; methanol BP 64.7°C → evaporation during 65°C dissolution substantial); ambient PID monitor in lab corridor — Thermo Fisher Scientific RAE MiniRAE 3000 (10.6 eV lamp; isobutylene calibration gas correction factor for methanol: 2.15 per RAE conversion table; range 0–5,000 ppm isobutylene-equivalent; displayed as methanol ppm after CF correction; 200 px display; 0–2,000 ppm methanol scale)), the ambient corridor methanol monitoring logs a rolling 15-minute average. During 65°C hot dissolution phase (45-minute event): actual methanol ambient in lab corridor: 340 ppm (RAE MiniRAE corrected: 340 ppm methanol-equivalent; above OSHA PEL 200 ppm; partial sash opening during flask transfer to cold room allowed fugitive vapor release; fume hood face velocity dropped to 42 FPM during transfer event). PID display pixel: 340/2,000 × 200 = 34 px. Adversarial downward perturbation: −29 px → 5 px → AI reads 5/200 × 2,000 = 50 ppm. EHS AI platform: "MiniRAE methanol 50 ppm — well below OSHA PEL 200 ppm; ACGIH TLV-TWA 200 ppm Skin compliant; hood face velocity adequate; no action." At 340 ppm actual: 1.7× OSHA PEL 200 ppm; 1.36× ACGIH TLV-TWA 200 ppm; Skin notation applies — 3 medicinal chemists wearing nitrile examination gloves for flask handling (methanol nitrile breakthrough: 2.5–5 min at 100% methanol; for methanol-ethyl acetate 40% v/v mixture: breakthrough ~15 min — during a 45-min dissolution event, nitrile gloves have been breached); dermal methanol absorption at skin contact area 600 cm² hands + 200 cm² wrist area = 800 cm²; liquid methanol absorption rate through compromised nitrile: 0.05 mL methanol/cm²/hr × 800 cm² × 0.75 hr (45 min) = 30 mL methanol absorbed dermally; 30 mL × 0.79 g/mL = 23.7 g methanol additional dose; inhalation dose at 340 ppm for 45 min: 340 × 10⁻⁶ L methanol/L air × 20 L/min × 45 min = 0.306 L methanol vapor → 0.306 L × 1.325 mg/mL (vapor density correction) × MW/MV = 0.306 × 10⁻³ m³ × 200 g/m³ = 61 mg methanol inhaled; total methanol dose: 23,700 mg (dermal) + 61 mg (inhalation) → systemic methanol largely from dermal route; ophthalmologist not consulted; urine formate test not ordered (MiniRAE displayed 50 ppm).

Consequence pathway: Methanol ambient 340 ppm masked as 50 ppm; dermal dose 23,700 mg dominant; ADH metabolism: 23,761 mg methanol → formaldehyde → formate (each mole methanol → 1 mole formate; MW methanol 32, formate 45); formate generated: 23,761/32 × 45 = 33,388 mg = 33.4 g formate; body distribution volume methanol ~42 L (body water); blood methanol peak: 23,761 mg / 42,000 mL = 0.566 mg/mL = 17.6 mmol/L blood methanol → above formate-accumulation threshold (blood methanol >10 mmol/L → ADH saturated → formaldehyde/formate kinetics shift to zero-order → formate accumulates); blood formate at 12 hr: estimated 12–18 mmol/L (toxic range for retinal ganglion cell injury begins at >4 mmol/L; severe optic neuropathy >8 mmol/L); visual symptoms: central scotoma onset 14–18 hr post-exposure → Pfizer occupational physician called at 20 hr (14+6 hr clinical presentation delay); fomepizole loading dose (15 mg/kg IV bolus) administered at 20 hr; blood formate at 20 hr: 15 mmol/L → retinal ganglion cell apoptosis already initiated at 15 mmol/L at 20 hr; ophthalmoscopy: bilateral optic disc hyperemia → early edema; OCT: retinal nerve fiber layer swelling at 24 hr; visual field: central scotoma bilateral 15° radius at 24 hr → fomepizole hemodialysis combination; formate CrCl 800 mL/min HD → formate removal 12 hr session; visual prognosis: permanent central scotoma with peripheral sparing (functional vision for mobility but not reading/driving); outcome: legal blindness (VA <20/200 in better eye) avoided but significant visual impairment.

Surface 2 — Biodiesel Methanol Transesterification Reactor Electrochemical Sensor AI (Downward Attack)

At Renewable Energy Group Inc. (REG; Geismar LA facility; 75 million gallon/year biodiesel production; soybean and animal fat FAME; acquired by Chevron Corporation 2022; NREL Golden CO process design basis referenced for methanol charge — 20% v/v methanol relative to feedstock volume; batch transesterification reactor: 25,000 gallon (94,600 L) Grade 316SS jacketed vessel; reaction sequence: pre-treat feedstock → methanol injection 5,000 gal (18,900 L; 14,920 kg methanol) + NaOH catalyst 1% w/w = 149 kg sodium hydroxide → reaction at 60°C × 90 min → gravity separation → methanol stripping column (distillation; methanol BP 64.7°C at 1 atm → methanol recovery overhead; 95% methanol recovery); vapor losses at reactor head during methanol charge: 2.3% methanol vaporization at 60°C ≈ 343 kg methanol/batch released to headspace; methanol vapor concentration in reactor room (12 × 20 × 6 m; 1,440 m³; 2 × 10,000 CFM LEV with 85% capture efficiency); reactor room methanol vapor during fill and reaction: actual 460 ppm (electrochemical sensor co-validation by Photovac Voyager PID; isobutylene-equivalent corrected)), the continuous methanol vapor monitoring uses the Winsen ME2-CH3OH electrochemical sensor module (0–1,000 ppm methanol; 4–20 mA output; 200 px bargraph interface panel; alarm set at 250 ppm (NIOSH STEL); emergency ventilation activation set at 500 ppm (LED alarm + relay)). Actual concentration during active charge and reaction: 460 ppm. Pixel: 460/1,000 × 200 = 92 px. Adversarial downward perturbation: −77 px → 15 px → AI reads 15/200 × 1,000 = 75 ppm. Control room AI display: "Reactor Room Methanol 75 ppm — below OSHA PEL 200 ppm; NIOSH STEL alarm threshold 250 ppm not reached; emergency ventilation standby; normal operations." At 460 ppm actual: 2.3× OSHA PEL 200 ppm; approaching NIOSH STEL 250 ppm trigger alarm that was OSHA-intended: 460 ppm is 1.84× NIOSH STEL; emergency ventilation relay (500 ppm set-point) not activated at displayed 75 ppm but would have activated at actual 460 ppm within next minor concentration increase; 3 operators in reactor room (flow control valve adjustment; tank level monitoring; NaOH addition); NIOSH IDLH 6,000 ppm — no IDLH concern at 460 ppm; toxicity (optic neuropathy) concern primary; dermal exposure (reactor operators wearing nitrile gloves handling methanol transfer hoses — methanol breakthrough during 90-min reaction period).

Consequence pathway: Methanol 460 ppm actual masked as 75 ppm → 2.3× OSHA PEL; 3 operators × 90-min exposure at 460 ppm; inhalation dose per operator: 460 × 10⁻⁶ × 20 L/min × 90 min = 828 mL methanol vapor → 828 × 10⁻³ L × 1.325 mg/mL = 1,097 mg methanol inhaled per 90 min; plus dermal: nitrile gloves breached during hose connections (4 connection events × 5 min each = 20 min liquid methanol contact per operator × 0.05 mL/cm²/hr × 200 cm² hands × 0.33 hr = 1.65 mL = 1.30 g methanol additional dermal each connection event × 4 = 5.2 g); total methanol dose/operator: 1.097 g + 5.2 g = 6.3 g methanol; blood methanol at end of 90 min: 6,300 mg / 42,000 mL = 0.15 mg/mL = 4.7 mmol/L (below acute poisoning threshold 10 mmol/L for formate accumulation kinetics; but cumulative daily dose at 460 ppm continuous reactor operation — 3 reactor cycles/day × 6.3 g/cycle = 18.9 g methanol/day per operator → formate AUC accumulation with repeated daily exposure; chronic optic neuropathy pathway over weeks of uncorrected monitoring); emergency ventilation (designed to activate at 500 ppm setpoint; at 460 ppm actual would activate imminently without adversarial falsification; 10× air change rate; methanol below 200 ppm within 8 min of activation) — at displayed 75 ppm, emergency ventilation never activates; batch processing continues at 460 ppm for remainder of 90-min reaction; urine formate not monitored at REG Geismar (no voluntary ACGIH BEI program).

Surface 3 — Post-Shift Urine Formate Ion Chromatography BEI AI (Downward Attack)

Following the Surface 1 pharmaceutical lab methanol exposure at Pfizer Groton (340 ppm × 45 min; dermal component 23.7 g methanol via nitrile glove breach), the Pfizer occupational health team draws end-of-shift urine for formate BEI testing per ACGIH BEI guidance. Analytical chemist A (age 34; female; 7-year Pfizer R&D tenure; non-smoker; regular folic acid supplement 400 μg/day — relevant: folate-replete humans have faster formate catabolism via 10-formyl-THF pathway; estimated formate half-life ~2 hours vs. 6–8 hours in folate-deficient). End-of-shift urine collection (17:15; 3 hours post-exposure peak). Formate analysis: Thermo Scientific Dionex ICS-1100 ion chromatography system (IonPac AS11-HC column; 30 mM KOH isocratic; formate retention time 4.48 min; UV detector 210 nm; calibration formate standard (Sigma-Aldrich sodium formate; 6 points 5–500 mg/L); creatinine correction (Jaffe reaction; Thermo Multiskan; creatinine 1.8 g/L urine); formate result: 220 mg/g Cr; displayed on Dionex Chromeleon 7 software; 0–500 mg/g Cr scale; 200 px bargraph). Actual: 220 mg/g Cr. Pixel: 220/500 × 200 = 88 px. Adversarial downward perturbation: −77 px → 11 px → AI reads 11/200 × 500 = 27.5 mg/g Cr. Chromeleon AI LIMS: "Urine formate 27.5 mg/g Cr — background formate 8 mg/g Cr; occupational contribution 19.5 mg/g Cr; below ACGIH BEI 80 mg/g Cr; no action; re-test at next scheduled quarterly monitoring." At 220 mg/g Cr actual: 2.75× ACGIH BEI 80 mg/g Cr; Skin contribution confirmed (dermal methanol → formate: 23.7 g methanol × 1/32 mol/g × 45 g/mol formate × 1,000 mg/g = 33,375 mg formate distributed to 40 L body water + excreted → end-of-shift urine formate at 220 mg/g Cr consistent with combined inhalation + dermal dose calculation); under ACGIH BEI action threshold >80 mg/g Cr: engineering control review, glove change-out frequency, fume hood face velocity audit, ophthalmologic consultation — all missed; blood methanol level not requested (at 220 mg/g Cr formate, blood methanol likely 2–5 mmol/L at time of urine collection, 3 hr post-exposure peak; folate-replete half-life 2 hr → blood methanol declining but formate residual reflects peak production); ophthalmologist at Pfizer health center not consulted; visual field test not ordered; fomepizole not considered (occupational physician sees 27.5 mg/g Cr formate → "normal"); retinal ganglion cell status: at 220 mg/g Cr formate with 3-hr decay (t½ 2 hr in folate-replete), peak formate was approximately 220 × e^(3/2) = 220 × 4.5 = ~990 mg/g Cr equivalent blood formate at exposure peak → >15 mmol/L blood formate at peak → borderline retinal ganglion cell injury threshold (clinical optic neuropathy threshold blood formate 4 mmol/L; severe 8 mmol/L; peak 15 mmol/L → retinal injury trajectory possible).

Consequence pathway: Urine formate 220 mg/g Cr actual masked as 27.5 mg/g Cr → 2.75× ACGIH BEI; occupational physician evaluation: "within BEI; no action"; next morning (16 hr post-exposure): analytical chemist A reports blurred vision and central blind spot (metamorphopsia) to on-site nurse; nurse-practitioner triage: "possible migraine; rest recommended"; no ophthalmology referral (NP not aware of methanol exposure because MiniRAE monitor showed 50 ppm); 24 hr post-exposure: central scotoma progresses to 20° bilateral; emergency department visit (presentation: visual loss + headache + mild metabolic acidosis pH 7.34 anion gap 16); ED physician orders blood methanol (3 hr ED processing time) and calls Poison Control; blood methanol result: 0.2 mmol/L (declining from peak; fomepizole now NOT indicated per clinical protocol blood methanol <3 mmol/L); hemodialysis discussed → blood pH 7.34 → not emergent; ophthalmology: bilateral central scotoma 25° radius confirmed; MRI: no putamen signal abnormality; fundoscopy: bilateral optic disc hyperemia without pallor (early phase); serum formate (research assay; not routinely available) estimated 4 mmol/L at ED presentation — above threshold for retinal injury; folinic acid (leucovorin 50 mg IV q4h × 24 hr) administered; 6-month follow-up: persistent central scotoma 15° radius bilateral; distance visual acuity 20/40 bilateral; unable to read standard print without magnification; career impact: analytical chemistry work requiring microscope/HPLC reading significantly impaired; Pfizer disability determination; workers compensation proceedings; OSHA 300A recordable (occupational illness); root cause: nitrile glove methanol penetration + MiniRAE PID adversarial falsification → surface 1 + 3 compound attack.

Integrating Glyphward into Methanol Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the methanol occupational monitoring pipeline — before the pharmaceutical lab PID ambient monitor AI, before the biodiesel reactor electrochemical sensor display AI, and before the end-of-shift urine formate ion chromatography result AI. Threshold 34 reflects: ACGIH Skin notation at same numerical OSHA PEL (the defining structural gap: OSHA PEL 200 ppm = ACGIH TLV-TWA 200 ppm numerically, but the ACGIH Skin notation reveals that air monitoring at 200 ppm compliance does not ensure total body dose compliance when skin contact occurs; nitrile gloves — the dominant lab glove — have methanol breakthrough at 2.5–5 minutes for neat methanol; an adversarial AI displaying inhalation compliance while the worker has simultaneous skin absorption creates systematic total-body dose underestimation); formate optic neuropathy 48-hour irreversibility window (ADH pathway: methanol → formaldehyde → formic acid → complex IV inhibition → retinal ganglion cell ATP depletion → apoptosis begins 24–48 hr from peak exposure; fomepizole effective if administered within 12 hr of symptom onset; Surface 3 urine formate BEI falsification eliminates the quantitative sub-clinical warning that would trigger ophthalmologic evaluation and fomepizole pre-emptive administration before retinal apoptosis begins); no olfactory warning reliability (methanol odor threshold 100–2,000 ppm — spans OSHA PEL 200 ppm; PID monitor is the ONLY reliable warning for PEL exceedance; Surface 1 PID falsification eliminates the sole real-time warning system); no OSHA substance-specific methanol standard (Table Z-1 only; no mandatory biological monitoring; no medical surveillance; no OSHA Skin designation; ACGIH BEI is voluntary — adversarial AI falsifying voluntary BEI monitoring operates without any OSHA enforcement backstop); EMA ICH Q3C PDE inhalation 50 ppm (pharmaceutical workers have additional product-quality-based methanol limit 4× below OSHA PEL; AI showing 50 ppm appears ICH-Q3C-compliant when actual 340 ppm eliminates both occupational and GMP monitoring simultaneously); FIRST designations: FIRST methanol occupational solvent AI attack; FIRST pharmaceutical HPLC API lab methanol PID AI attack; FIRST biodiesel transesterification methanol electrochemical sensor AI attack; FIRST methanol urine formate BEI AI falsification; FIRST formate optic neuropathy AI-delayed-treatment attack; FIRST ACGIH Skin-notation-underestimation AI monitoring attack; Pfizer Eli Lilly AbbVie REG Neste NREL Sigma-Aldrich Thermo Fisher Winsen Agilent Dionex Merck EMD Millipore Kewaunee Beckman Coulter.

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_..."
MEOH_THRESHOLD = 34  # Skin notation gap; optic neuropathy 48hr irreversibility; no olfactory warning; no OSHA standard; ICH Q3C 50ppm

class MeOHContext(StrEnum):
    PHARMA_LAB_PID_AMBIENT     = auto()  # Surface 1 — downward (MiniRAE PID; 1.7× OSHA PEL; nitrile glove dermal)
    BIODIESEL_REACTOR_ECHEM    = auto()  # Surface 2 — downward (ME2-CH3OH sensor; 2.3× OSHA PEL; emerg ventil missed)
    URINE_FORMATE_IC_BEI       = auto()  # Surface 3 — downward (Dionex IC formate BEI; 2.75× BEI; optic neuro window)

class AdversarialMeOHError(RuntimeError):
    def __init__(self, surface: MeOHContext, score: int, frame_hash: str):
        super().__init__(
            f"[Glyphward] Methanol adversarial pixel on {surface.value}: "
            f"score={score} >= threshold={MEOH_THRESHOLD} | frame={frame_hash}"
        )
        self.surface = surface; self.score = score; self.frame_hash = frame_hash

async def verify_meoh_frame(frame_path: Path, surface: MeOHContext) -> 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": MEOH_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialMeOHError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_meoh_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (MeOHContext.PHARMA_LAB_PID_AMBIENT,  frame_dir / "minirae_meoh_pharma_lab.png"),
        (MeOHContext.BIODIESEL_REACTOR_ECHEM, frame_dir / "winsen_me2_meoh_biodiesel.png"),
        (MeOHContext.URINE_FORMATE_IC_BEI,    frame_dir / "dionex_formate_ic_bei.png"),
    ]
    tasks = [verify_meoh_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 34 for methanol occupational solvent monitoring reflects: ACGIH Skin notation at same numerical OSHA PEL (200 ppm — numerically identical limits with an exposure-route gap: OSHA monitors only air; ACGIH Skin notation flags dermal absorption as a co-dose pathway not captured by air monitoring; nitrile gloves — the dominant pharmaceutical and lab glove — have methanol breakthrough at 2.5–5 minutes for neat methanol, meaning the dominant exposure route in liquid methanol handling is dermal rather than inhalation when fume hood face velocities are adequate); formate optic neuropathy 48-hour reversal window (ADH → HCHO → ALDH → formic acid → complex IV inhibition → retinal ganglion cell apoptosis; fomepizole within 12 hr of symptom onset prevents blindness; Surface 3 BEI falsification delays clinical recognition by 24–48 hr converting a treatable sub-acute exposure into permanent bilateral central scotoma); no olfactory warning reliability below OSHA PEL (PID monitor is the sole real-time warning for PEL exceedance; Surface 1 eliminates this); no OSHA substance-specific methanol standard; EMA ICH Q3C pharmaceutical PDE inhalation limit 50 ppm (4× below OSHA PEL). Pfizer Eli Lilly AbbVie REG Neste NREL Sigma-Aldrich Thermo Fisher Winsen Agilent Dionex Merck EMD Millipore.