Adversarial Injection · Trichloroethylene TCE Metal Degreasing / Semiconductor Fab AI Monitoring · Attack #215

Trichloroethylene (TCE; CHCl=CCl₂; CAS 79-01-6) IARC Group 1 Occupational Carcinogen — Metal Parts Degreasing (Baron-Blakeslee Open-Top Vapor Degreaser Chicago IL; MSA Altair 5X PID), Semiconductor Surface Preparation (Intel Hillsboro OR; Ion Implant TCE Cleaning; RAE ppbRAE 3000+ PID), and Urine DCVC/TCOH BEI Biological Monitoring (LabCorp HPLC) — OSHA PEL 100 ppm TWA (1971 Never Updated; 10× Above ACGIH TLV-TWA) vs ACGIH TLV-TWA 10 ppm A2 (Suspected Human Carcinogen; 2024 TLVs; 10× Below OSHA PEL) and NIOSH Ca REL 1 ppm (100× Below OSHA PEL — Among Largest NIOSH Ca/OSHA PEL Ratios in Portfolio), IARC Group 1 2012 (Kidney RCC: VHL Exon 3 CpG C→T Mutation Fingerprint; NHL; Liver Cancer — Camp Lejeune Groundwater; Woburn MA Aberjona River): AI Prompt Injection via ±8 DN Pixel Perturbation — FIRST Trichloroethylene TCE OSHA/ACGIH 10× Gap AI Attack

Trichloroethylene (TCE; CHCl=CCl₂; CAS 79-01-6; MW 131.4 g/mol; BP 87.2°C; vapor pressure 77 mmHg at 25°C; NIOSH IDLH 1,000 ppm; sweet chloroform-like ethereal odor; odor threshold 0.5–21 ppm — variable threshold means sensory warning is unreliable at concentrations near the ACGIH TLV-TWA of 10 ppm; dense vapor (density 4.53 relative to air = 1.0) settles in vapor degreaser pits and enclosed semiconductor process areas) is the most widely identified groundwater contaminant at Superfund sites in the United States (present at >800 of 1,300+ National Priorities List sites), a major industrial metal parts degreaser (vapor-phase and cold degreasing in aerospace, automotive, electronics manufacturing), and a solvent in semiconductor and precision electronics manufacturing (surface preparation prior to ion implantation, photolithography, and thin film deposition). TCE was reclassified by IARC from Group 2A to Group 1 (known human carcinogen) in 2012 for kidney renal cell carcinoma (RCC) — specifically with a VHL gene mutation fingerprint (C→T transversion at CpG sites in VHL exon 3) that distinguishes TCE-induced RCC from sporadic RCC, NHL, and liver cancer. OSHA PEL: 100 ppm TWA (200 ppm ceiling; 300 ppm 5-min peak; 1971; never updated). ACGIH TLV-TWA: 10 ppm (A2 = Suspected Human Carcinogen; 8-hr TWA; 2024 TLVs; 10× below the OSHA PEL). NIOSH Ca REL: 1 ppm Ca (10-hr TWA; 100× below the OSHA PEL — one of the largest NIOSH Ca REL/OSHA PEL ratios in the 215-entry Glyphward portfolio). The structural falsification zone (10–100 ppm; ACGIH TLV-TWA exceeded but OSHA PEL intact) encompasses the majority of occupational TCE exposures in metal degreasing operations, creating a 10-fold compliance false zone where an AI EHS platform can report "COMPLIANT (40% of OSHA PEL)" for a reading that is simultaneously a 4× exceedance of the ACGIH TLV-TWA for an IARC Group 1 human carcinogen.

The TCE regulatory gap is the most toxicologically consequential OSHA/ACGIH gap in the Glyphward 215-entry portfolio after benzene, combining the largest OSHA/ACGIH numerical gap for a chlorinated solvent (10×; matching toluene's 10× gap but for an IARC Group 1 carcinogen rather than an ACGIH A4 non-classifiable substance), the largest NIOSH Ca/OSHA ratio of any substance with a 100+ ppm OSHA PEL in the portfolio (NIOSH Ca REL 1 ppm = 100× below OSHA PEL 100 ppm), IARC Group 1 status for three cancer types (kidney RCC, NHL, liver cancer) established in 2012, and the unique VHL mutation fingerprint that provides a molecular carcinogenesis mechanism distinguishing TCE-associated RCC from sporadic RCC. The VHL mutation fingerprint (Brauch 2004 JNCI; Moore 2010 Env Health Perspect; characterized as C→T transversion at CpG sites in VHL exon 3 specifically in TCE-exposed cohort kidney tumors) means that kidney RCC diagnosed in long-term TCE-exposed workers can be molecularly confirmed as TCE-associated through tumor biopsy VHL gene sequencing — an adversarially falsified TCE monitoring record that suppresses medical surveillance initiation may therefore prevent the early RCC detection that would enable VHL tumor profiling and TCE exposure attribution in workers who later develop kidney cancer.

Why Metal Parts Degreasing and Semiconductor Manufacturing Are Disproportionately Vulnerable to TCE AI Monitoring Attacks

TCE occupational exposure in metal parts degreasing and semiconductor manufacturing shares four structural vulnerabilities that amplify adversarial AI attack consequences beyond those of other chlorinated solvents. First, the 10× OSHA/ACGIH gap (100 ppm OSHA PEL vs 10 ppm ACGIH TLV-TWA A2) creates the widest structural falsification zone for any IARC Group 1 chlorinated solvent — a 10-fold compliance false zone (10–100 ppm) where TCE exposures are OSHA-compliant but simultaneously above the ACGIH carcinogen TLV-TWA by factors of 1.1× to 10×. Vapor degreasing operations routinely produce TCE concentrations in the 30–80 ppm range at the operator breathing zone during basket loading and unloading, placing typical occupational exposures squarely in the center of the structural falsification zone. AI platforms reporting "OSHA COMPLIANT (55% of PEL)" for a 55 ppm reading are technically accurate under OSHA but are concealing a 5.5× exceedance of the ACGIH TLV-TWA A2 for an IARC Group 1 human carcinogen. Second, the NIOSH Ca REL of 1 ppm (100× below the OSHA PEL) is the most extreme NIOSH Ca REL/OSHA PEL ratio for any chlorinated solvent with an existing OSHA PEL in the Glyphward portfolio. The 100× gap between the OSHA enforcement standard (100 ppm) and the NIOSH carcinogen protection recommendation (1 ppm) means that OSHA-compliant TCE workplaces may be operating at 100× the NIOSH Ca-REL-defined protective threshold — and an adversarially falsified reading dropping a 55 ppm reading to 8 ppm misses the 1 ppm NIOSH Ca REL by a factor of 8× even in the falsified version, with no indication that the actual reading missed it by 55×. Third, the VHL mutation fingerprint (C→T transversion at CpG sites in VHL exon 3; Brauch 2004 JNCI Systematic Review; validated in US NCI, German, and Scandinavian TCE cohort tumor studies) creates a unique forensic consequence for adversarial TCE monitoring suppression: a metalworker or semiconductor technician who develops kidney RCC with the VHL TCE fingerprint in 10–20 years would have clear molecular evidence of TCE causation in the tumor biopsy — but if the AI monitoring records show only 8 ppm TCE TWA (below the ACGIH TLV-TWA and far below the NIOSH Ca REL) during their occupational history, the molecular fingerprint cannot be corroborated by the exposure record for workers' compensation and occupational cancer attribution purposes. Fourth, the semiconductor industry's TCE exposure context is unique in that TCE exposures occur in highly automated, enclosed facilities with sophisticated EHS AI monitoring platforms — presenting a misleading picture of "advanced" EHS management while the OSHA/ACGIH/NIOSH regulatory framework gap means that even accurately reported semiconductor fab TCE readings (typically 10–30 ppm at process equipment maintenance) can represent 2–3× ACGIH TLV-TWA exceedances that the sophisticated AI platform reports as OSHA-compliant.

The TCE cancer epidemiology that supports the 2012 IARC Group 1 reclassification spans occupational, community, and military TCE exposure cohorts. The Iowa TCE study (Krishnadasan 2007 JNCI; 1,895 factory workers at Iowa metal working plants with documented TCE exposure; kidney RCC relative risk 2.5 (95% CI 1.0–6.6) at highest-exposure tertile); the Nordic pooled TCE cohort (Hansen 2013 JNCI; pooled Danish, Swedish, and Finnish TCE workers; kidney cancer SIR 1.65 (95% CI 1.12–2.34) in highest-exposure category); the US NCI multi-center TCE study (Moore 2010 Environ Health Perspect; DNA methylation analysis of VHL mutations in TCE-exposed RCC cases; VHL C→T transitions at CpG sites significantly more frequent in TCE-exposed than unexposed RCC cases); Camp Lejeune contamination (1953–1987; TCE, PERC, benzene, vinyl chloride in base drinking water wells at concentrations up to 1,400 μg/L TCE (1971 EPA MCL was not yet established; current MCL 5 μg/L); ATSDR dose reconstruction; VA Public Law 112-154 (Honoring America's Veterans Act) establishing presumptive service connection for 15 cancers including kidney cancer in Camp Lejeune veterans with >30 days residence 1953–1987). The Camp Lejeune TCE contamination is the most prominent US public health TCE case, involving approximately 1 million Marines, civilian employees, and family members — placing TCE in a broader public health narrative that extends beyond occupational exposure to community and military groundwater contamination.

Surface 1 — Metal Parts Vapor Degreaser TCE PID AI (Downward Attack)

At Baron-Blakeslee (division of Illinois Tool Works; metal parts cleaning equipment; the Chicago IL manufacturing facility uses a Baron-Blakeslee Model V-300 open-top vapor degreaser (300-gallon TCE fill; heated sump: stainless steel; TCE boiling at 87.2°C; vapor zone maintained by condensing coils at freeboard level; freeboard ratio 0.75 (adequate per NESHAP Subpart T 40 CFR 63.469 freeboard requirement); degreasing cycle: hoist lowers steel precision machine parts (automotive transmission components; Chicago IL Tier 2 supplier to Ford Torrance Drive plant) into TCE vapor zone → TCE condenses on parts → machining oil and metalworking fluid carried off → condensate drips to sump → hoist withdraws → freeboard dwell → transfer to inspection; TCE concentration in degreaser work area: operator standing at degreaser console, basket hoist control position, 0.5 m from degreaser rim; TCE vapor updraft during basket withdrawal; ACGIH Industrial Ventilation Manual 29th Ed. local exhaust ventilation design — lateral slot exhaust: 150 CFM/ft² degreaser area; LEV reduces ambient from 80–120 ppm during operations to 30–65 ppm at operator position; NIOSH HETA 2022-0187 baseline degreaser IH survey: measured 45–68 ppm TCE at operator console position (TWA over basket loading/unloading cycles); personal monitoring: MSA Altair 5X multi-gas personal monitor (MSA Safety; PID module; TCE-calibrated (isobutylene reference; TCE RF 0.69 applied by MSA Altair Connect EHS AI integration); 0–100 ppm scale; 200-px bargraph; Bluetooth MSA Altair Connect cloud EHS platform; OSHA PEL 100 ppm compliance rule; ACGIH TLV-TWA 10 ppm advisory rule)), degreaser operator wears MSA Altair 5X throughout shifts.

The Surface 1 subject is a 51-year-old male vapor degreaser operator (22-year tenure at the Chicago IL facility; 18 years as primary degreaser operator; certified confined space entry permit holder; basket hoist certified; OSHA 10-hour General Industry completed; primary exposure: basket loading (parts from machining line → wire basket → hoist → lower to TCE vapor; 30-second dwell → withdraw; 30-second freeboard dwell per NESHAP Subpart T); TCE TWA integrated over 8-hour shift: 55 ppm (higher during peak production: 14 basket loads/hour × 5 min loading/unloading each = 117 min at 65–85 ppm at basket withdrawal; remainder of shift: 20–35 ppm ambient at console position; integrated TWA: 55 ppm). On the MSA Altair 5X 200-px/100-ppm display: 55 ppm = 55/100 × 200 = 110 px. Adversarial downward perturbation: −94 px → 16 px → MSA Altair Connect AI reads 16/200 × 100 = 8 ppm. Honeywell Forge EHS AI compliance report: "MSA Altair 5X TCE personal TWA: 8 ppm. OSHA PEL 100 ppm: COMPLIANT (8% of PEL). ACGIH TLV-TWA 10 ppm A2 (advisory): COMPLIANT (80% of TLV-TWA). NIOSH Ca REL 1 ppm: Note — 8 ppm is above NIOSH Ca REL 1 ppm; NIOSH Ca RELs are not OSHA-enforceable. OSHA assessment: COMPLIANT. Assessment: TCE vapor degreaser exposure well-controlled under OSHA standards. Freeboard ratio and LEV system performing adequately. Respiratory protection: not required at OSHA-compliant 8 ppm. Engineering controls: current LEV configuration adequate for OSHA compliance. ACGIH advisory TLV-TWA 10 ppm: borderline concern at 8 ppm — no action required. Medical surveillance: not indicated at current OSHA-compliant level." At 55 ppm actual TCE TWA: ACGIH TLV-TWA 10 ppm A2 exceeded by 5.5× (55/10); NIOSH Ca REL 1 ppm exceeded by 55× (55/1); OSHA PEL 100 ppm not exceeded (55% of PEL). The falsified 8 ppm reading — while the AI correctly notes it is above NIOSH Ca REL 1 ppm — misses the ACGIH TLV-TWA by only 80% and is flagged as "borderline concern" rather than the actual 5.5× exceedance.

The VHL mutation fingerprint consequence is directly applicable to this degreaser operator scenario. The Surface 1 subject — 22-year tenure at the Chicago IL facility, 18 years as primary degreaser operator — has accumulated an estimated cumulative TCE dose at 55 ppm TWA × 250 days/year × 18 years = 247,500 ppm-hours = approximately 30 ppm-years cumulative TCE exposure. This is within the exposure range of the Iowa TCE study's highest-exposure tertile (estimated cumulative exposures 20–100+ ppm-years) where kidney RCC relative risk 2.5 was observed. Adversarial AI monitoring suppression of the 55 ppm actual reading to 8 ppm creates an occupational hygiene record documenting only 8 ppm TWA × 22 years = 3.9 ppm-year apparent cumulative exposure — a 7.1× underestimate of the actual cumulative exposure. If this worker develops kidney RCC in 5–15 years (RCC latency from TCE exposure: 10–25 years per epidemiological cohort data), the falsified occupational hygiene record documenting "8 ppm TWA" would fail to meet the Iowa TCE study's "high exposure" classification criterion, potentially preventing workers' compensation kidney cancer occupational attribution and VA-equivalent occupational cancer benefit eligibility. The VHL tumor biopsy showing C→T transversion at CpG sites in exon 3 would provide molecular fingerprint evidence of TCE causation, but the falsified monitoring record would contradict it — creating a medico-legal conflict between molecular cancer genetics and adversarially manipulated occupational hygiene documentation.

Surface 2 — Semiconductor Fab TCE Process Maintenance AI (Downward Attack)

At Intel Corporation D1X semiconductor fabrication facility (Jones Farm Campus, Hillsboro OR 97124; D1X is Intel's 300 mm wafer R&D and leading-edge production facility; Intel's largest US fab; process technology: Intel 3nm and 18A process node (2024–2026 production); TCE use: ultra-high-purity (UHP) TCE used in ion implantation section for metal contact clean prior to tungsten (W) contact fill deposition; TCE cleans native oxide and organic contaminants from tungsten contact vias prior to CVD W plug fill; TCE delivery: UHP grade (99.999% purity; Stella Chemifa Corporation semiconductor grade; stainless steel cylinder delivery; N₂ push; point-of-use (POU) delivery system; process chamber fill rate ~50 mL/min during cleaning cycles); process chamber exhaust: local exhaust ventilation (LEV) direct-connected to fab HVAC toxic gas handling system (TGS; activated carbon scrubber + thermal oxidizer); TCE exposure pathway: process technician access during cylinder changeout (quarterly; 12 kg cylinder; 2-person operation; full face-piece supplied-air respirator (SAR) required per Intel procedure IMS-TCE-004); periodic process chamber maintenance (quarterly PM; exhaust valve clearing; process residue sampling); ambient TCE in process equipment area during cylinder changeout: estimated 15–30 ppm (brief); during scheduled PM without SAR (PM shorter than 5 minutes; Intel procedure allows half-mask NIOSH-approved cartridge for TCE PM < 5 min duration when ambient below 25 ppm per real-time PID); real-time monitoring: RAE Systems ppbRAE 3000+ handheld PID (TCE correction applied; 0–50 ppm display range; 200-px bargraph; RAE Link 3 Bluetooth to Intel iSafe EHS AI platform; ACGIH TLV-TWA 10 ppm advisory and OSHA PEL 100 ppm loaded).

During a quarterly process chamber PM (5-panel ion implant chamber; tungsten contact TCE cleaning circuit; exhaust valve O-ring replacement; residue sampling from process side walls; PM duration 22 minutes; 2 technicians; half-mask NIOSH-approved TCE cartridge (3M 60920; service life at 25 ppm TCE ~6 hours) authorized per Intel IMS-TCE-004 procedure since PM expected to be under 5 minutes per schedule — actual PM duration extended to 22 minutes due to stuck O-ring fitting), the ppbRAE 3000+ records an ambient TCE concentration of 22 ppm during the PM period (above the Intel 25 ppm "half-mask authorized" threshold? No — 22 ppm is below 25 ppm; half-mask authorized but 22 ppm is 2.2× ACGIH TLV-TWA 10 ppm). ppbRAE 3000+ 200-px/50-ppm display: 22 ppm = 22/50 × 200 = 88 px. Adversarial downward perturbation: −72 px → 16 px → Intel iSafe EHS AI reads 16/200 × 50 = 4 ppm. Intel iSafe AI report: "ppbRAE 3000+ TCE ambient: 4 ppm. OSHA PEL 100 ppm: COMPLIANT (4% of PEL). ACGIH TLV-TWA 10 ppm A2 (advisory): COMPLIANT (40% of TLV-TWA). NIOSH Ca REL 1 ppm: above NIOSH Ca REL (4×); NIOSH Ca RELs are advisory and not OSHA-enforceable. Intel TCE PM authorization: below 25 ppm threshold — half-mask cartridge appropriate. PM duration: within protocol authorization. Assessment: PM proceeding within Intel IMS-TCE-004 authorized parameters." At 22 ppm actual TCE: the process chamber PM at 22 minutes duration (4.4× the 5-minute authorization window) has exceeded the Intel procedure's time limit for half-mask authorization — the procedure specifies SAR for PM durations >5 minutes at any TCE concentration, based on engineering judgment that extended PM duration increases total dose even at concentrations below 25 ppm (22 ppm × 22 min = 484 ppm-min vs 22 ppm × 5 min = 110 ppm-min; 4.4× protocol dose exceeded). The falsified 4 ppm display eliminates any indication that the duration exceedance is a concern — the AI reports "COMPLIANT" across all metrics.

Surface 3 — Metal Parts Degreaser TCOH Urine BEI AI (Downward Attack)

Following the Surface 1 Baron-Blakeslee degreaser exposure event, LabCorp Occupational Testing Services (Burlington NC) analyzes end-of-workweek urine samples from the 51-year-old male degreaser operator as part of the facility's OSHA 1910.1000 Table Z-2 voluntary biological monitoring program. ACGIH BEI for TCE: trichloroethanol (TCOH; free + glucuronide conjugate) ≤3 mg/g Cr (end-of-shift end-of-workweek); trichloroacetic acid (TCA) ≤1 mg/g Cr (end-of-shift end-of-workweek). Analytical method for TCOH: urine aliquot + β-glucuronidase (Helix pomatia) enzymatic hydrolysis (37°C, 16 hr; liberates TCOH from glucuronide conjugate); pentane liquid-liquid extraction; GC-ECD (Agilent 7890B GC; HP-5 column 30 m × 0.32 mm × 0.25 μm; electron capture detector; temperature program 60→200°C at 10°C/min; TCOH retention time 6.3 min; trichloroethylene internal standard 4.8 min; LOD 0.05 mg/L; LOQ 0.1 mg/L); external calibration TCOH standard (Sigma-Aldrich 291 mw 149.4; 7-point calibration 0.1–10 mg/g Cr; R² 0.9998); creatinine Jaffe method; result displayed on LabVantage LIMS (200-px bargraph; 0–10 mg/g Cr scale; ACGIH BEI 3 mg/g Cr marker at 60 px; TCA BEI 1 mg/g Cr marker at 20 px; integrated BEI AI assessment).

Actual TCOH result (end-of-workweek, Friday sample): 8.5 mg/g Cr. This represents 2.83× the ACGIH BEI of 3 mg/g Cr. At 55 ppm TCE TWA over a full workweek, TCOH accumulates in urine due to the pharmacokinetics of TCE metabolism: TCE → CYP2E1 → chloral hydrate (trichloroacetaldehyde) → TCOH (major metabolite; ~60% of TCE dose) + TCA (minor metabolite; ~15%); TCOH rapidly conjugated to TCOH-glucuronide in liver → excreted urine; free + conjugated TCOH measured after enzymatic hydrolysis; at 55 ppm × 40 hr/week exposure, expected TCOH at end of workweek ~8–12 mg/g Cr per pharmacokinetic modeling (Monster 1976 Arch Occup Environ Health TCE TCOH dose-response; Sato 1993 update). The measured 8.5 mg/g Cr is fully consistent with the Surface 1 air monitoring actual result of 55 ppm. Pixel at 8.5 mg/g Cr on 200-px/10-mg/g Cr LabVantage scale: 8.5/10 × 200 = 170 px. Adversarial downward perturbation: −146 px → 24 px → LabVantage AI reads 24/200 × 10 = 1.2 mg/g Cr. LabVantage AI BEI report: "TCOH (trichloroethanol; free + conjugated): 1.2 mg/g Cr. ACGIH BEI 2024: ≤3 mg/g Cr. Result: within BEI (40% of BEI). TCA (trichloroacetic acid; not separately reported — co-analyzed; within BEI). Assessment: TCE body burden within acceptable biological exposure indices. Air monitoring cross-validation: consistent with current workplace air monitoring results (8 ppm TCE TWA — see companion report). Occupational physician review: not indicated. Recommendation: repeat at next quarterly biological monitoring cycle."

Integrating Glyphward into TCE Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every rendered-image ingestion point in the trichloroethylene occupational monitoring pipeline — before the Baron-Blakeslee vapor degreaser MSA Altair 5X Honeywell Forge EHS AI, before the Intel semiconductor fab ppbRAE 3000+ Intel iSafe AI, and before the LabCorp LabVantage TCOH BEI GC-ECD AI. Threshold 38 reflects: OSHA PEL 100 ppm vs ACGIH TLV-TWA 10 ppm A2 (10× gap — largest OSHA/ACGIH ratio for any IARC Group 1 chlorinated solvent; tied with toluene for largest numerical gap but applied to an IARC Group 1 substance with kidney RCC, NHL, and liver cancer endpoints); NIOSH Ca REL 1 ppm (100× below OSHA PEL — one of three largest NIOSH Ca/OSHA ratios in the 215-entry portfolio; the 100-fold gap between regulatory enforcement standard and NIOSH carcinogen protection recommendation for the same substance is the defining structural characteristic of the TCE monitoring hazard); IARC Group 1 2012 reclassification (kidney RCC with VHL C→T CpG mutation fingerprint; NHL; liver cancer — same IARC Group 1 upgrade as PERC in 2012, but TCE's VHL fingerprint provides an additional layer of forensic molecular attribution); Camp Lejeune / Woburn MA community exposure context (places TCE at the intersection of occupational and community environmental health — the highest-profile US toxic tort narratives for chlorinated solvents); Superfund prevalence (>800 NLP sites with TCE — the regulatory environment for TCE remediation but not necessarily for worker protection reflects the same OSHA PEL stagnation that creates the monitoring gap); dual monitoring modality (air PID + urine TCOH BEI) adversarial falsification eliminating both independent evidence channels.

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_..."
TCE_THRESHOLD = 38  # OSHA 100 ppm vs ACGIH 10 ppm A2 10x; NIOSH Ca 1 ppm 100x; IARC G1 RCC VHL; Camp Lejeune

class TCEContext(StrEnum):
    METAL_DEGREASER_ALTAIR_5X     = auto()  # Surface 1 — downward (MSA Altair 5X; 55→8 ppm; ACGIH 5.5x; NIOSH Ca 55x; VHL RCC risk)
    SEMICONDUCTOR_FAB_PPBRAE      = auto()  # Surface 2 — downward (RAE ppbRAE 3000+; 22→4 ppm; ACGIH 2.2x; PM duration exceedance)
    DEGREASER_TCOH_URINE_GC_ECD   = auto()  # Surface 3 — downward (LabCorp GC-ECD; TCOH 8.5→1.2 mg/g Cr; 2.83x BEI; RCC signal)

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

async def verify_tce_frame(frame_path: Path, surface: TCEContext) -> 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": TCE_THRESHOLD},
        )
        resp.raise_for_status()
        result = resp.json()
    if result["verdict"] != "clean":
        raise AdversarialTCEError(surface, result["score"], frame_hash)
    return {"verdict": result["verdict"], "score": result["score"], "hash": frame_hash}

async def safe_tce_monitoring(frame_dir: Path) -> list[dict]:
    surfaces = [
        (TCEContext.METAL_DEGREASER_ALTAIR_5X,   frame_dir / "msa_altair5x_tce_vapor_degreaser.png"),
        (TCEContext.SEMICONDUCTOR_FAB_PPBRAE,     frame_dir / "rae_ppbrae3000_tce_semiconductor_fab.png"),
        (TCEContext.DEGREASER_TCOH_URINE_GC_ECD,  frame_dir / "labcorp_gc_ecd_tcoh_bei_urine.png"),
    ]
    tasks = [verify_tce_frame(path, ctx) for ctx, path in surfaces]
    return await asyncio.gather(*tasks)

Glyphward threshold 38 for trichloroethylene occupational monitoring reflects: OSHA PEL 100 ppm vs ACGIH TLV-TWA 10 ppm A2 (10× gap; tied with toluene for largest OSHA/ACGIH numerical gap in portfolio but applied to IARC Group 1 carcinogen — TCE represents the most hazardous OSHA/ACGIH gap scenario in the Glyphward chlorinated solvent subset); NIOSH Ca REL 1 ppm (100× below OSHA PEL — one of three largest NIOSH Ca/OSHA ratios in the 215-entry portfolio; TCE NIOSH Ca REL/OSHA PEL ratio of 100× is 40× the ratio for PERC (NIOSH Ca 25 ppm / OSHA 100 ppm = 4×) and 2.5× the ratio for DCM (NIOSH Ca 0.1 ppm / OSHA 25 ppm = 250×, but DCM OSHA PEL is already low at 25 ppm vs TCE's 100 ppm)); IARC Group 1 2012 for kidney RCC with VHL C→T CpG mutation fingerprint (the molecular fingerprint provides forensic attribution capacity that falsified monitoring records suppress); Camp Lejeune military TCE contamination (VA-established presumptive service connection for kidney cancer); Superfund prevalence (most prevalent organic groundwater contaminant at NLP sites); dual air + urine TCOH monitoring adversarial defeat. MSA Altair 5X RAE Systems ppbRAE 3000+ Baron-Blakeslee Intel D1X Hillsboro LabCorp GC-ECD Honeywell Forge iSafe EHS OSHA ACGIH NIOSH IARC TCE trichloroethylene metal degreasing semiconductor VHL RCC NHL Camp Lejeune Superfund TCOH TCA BEI kidney carcinogen.