Adversarial Injection · Formaldehyde (HCHO; CAS 50-00-0) OSHA 1910.1048 PEL 0.75 ppm TWA + AL 0.5 ppm / ACGIH TLV-C 0.3 ppm A2 / NIOSH Ca REL 0.016 ppm / 47× OSHA:NIOSH Gap / Limit-Type TWA vs Ceiling Mismatch / 1910.1048 Action Level Suppression · Attack #316

Formaldehyde (HCHO; Methanal; Formalin [37% Aqueous]; CH₂O; CAS 50-00-0; OSHA 29 CFR 1910.1048 PEL 0.75 ppm TWA + STEL 2 ppm + Action Level 0.5 ppm; ACGIH TLV-C 0.3 ppm A2; NIOSH Ca REL 0.016 ppm TWA / 0.1 ppm STEL; 47× OSHA:NIOSH Gap; Limit-Type Mismatch; IARC Group 1 NPC + Leukemia) — Hospital Surgical Pathology Grossing Room (Mayo Clinic Rochester MN; IS Ventis Pro 5 Electrochemical), MDF Wood Composite Panel Production (Arauco Flakeboard Moncks Corner SC; Assay Technology Badge HPLC), and Funeral Home Embalming (Service Corporation International; BW GasAlertMax XT II) — OSHA 1910.1048 Action Level 0.5 ppm vs ACGIH TLV-C 0.3 ppm A2 vs NIOSH Ca REL 0.016 ppm: AI Prompt Injection via EHS Monitor Report AI — FIRST Formaldehyde 1910.1048 Action Level Suppression + NIOSH Ca REL 47× Gap AI Attack

Formaldehyde (HCHO; methanal; methylene oxide; CH₂O; MW 30.03 g/mol; CAS 50-00-0; commercially as 37–40% aqueous solution [formalin] with 10–15% methanol stabilizer; BP −19°C [gas at STP]; VP >1,000 mmHg at 20°C [highly volatile gas]; water solubility complete [aqueous formalin]; pungent, acrid odor threshold ~0.5 ppm [slightly below OSHA action level 0.5 ppm and OSHA PEL 0.75 ppm — odor is not a reliable protective warning at NIOSH Ca REL 0.016 ppm; olfactory fatigue begins ~0.25 ppm]; NIOSH IDLH 20 ppm; GHS: H350 Carcinogenicity Category 1A; H334 Respiratory Sensitizer; H317 Skin Sensitizer; OSHA 29 CFR 1910.1048: substance-specific carcinogen standard [one of fewer than 20 OSHA 6(b) specific carcinogen standards]; PEL 0.75 ppm TWA + STEL 2 ppm (15-min) + Action Level 0.5 ppm TWA [Action Level triggers: 8-hr periodic monitoring, medical surveillance (questionnaire + spirometry), regulated area designation, exposure records for 30 years; action level is a compliance trigger for OSHA 1910.1048 enhanced requirements — not just a lower PEL]; ACGIH TLV-C: 0.3 ppm A2 [CEILING — SHORT-TERM PEAK CONSTRAINT; not a TWA limit; 2024; A2 = Suspected Human Carcinogen; LIMIT-TYPE ARCHITECTURAL MISMATCH: OSHA 1910.1048 uses TWA and STEL metrics; ACGIH uses Ceiling (C) — a Ceiling metric constrains peak instantaneous concentrations, a fundamentally different monitoring architecture from OSHA's 8-hr TWA; an OSHA-calibrated AI using TWA-based comparison cannot assess compliance with a ceiling-based advisory limit without separate instantaneous peak data]; NIOSH Ca REL: 0.016 ppm TWA [Ca; lowest feasible; 47× below OSHA PEL 0.75 ppm; NIOSH formaldehyde CIB 62 (1981) + NIOSH Current Intelligence Bulletin on formaldehyde carcinogenicity (reaffirmed 2016); the NIOSH Ca REL 0.016 ppm is derived from the lowest-feasible-concentration principle: excess nasopharyngeal cancer + leukemia risk at any measurable exposure above 0.1 ppm; NIOSH Ca REL STEL: 0.1 ppm [10× above NIOSH Ca REL TWA; used for short-duration tasks]]; 47× OSHA:NIOSH span: OSHA 0.75 ppm / ACGIH Ceiling 0.3 ppm [limit-type mismatch] / NIOSH Ca REL 0.016 ppm; IARC Group 1 [2009; Monograph 88/100F: nasopharyngeal carcinoma (NPC) and leukemia (ALL, AML) — both sufficient evidence; strongest occupational human carcinogen evidence for NPC at occupational formaldehyde exposures; funeral home directors/embalmers and pathology technicians are sentinel exposure cohorts]; metabolism: ALDH2 aldehyde dehydrogenase oxidation → formic acid → CO₂ + H₂O [rapid clearance; no urinary biomarker BEI for formaldehyde inhalation]; DNA adduct: N²-hydroxymethyl-dG (HMdG) + DPC (DNA-protein cross-link) → single-strand breaks → NPC initiation pathway in nasopharyngeal epithelium; industrial uses: histology/pathology [10% buffered neutral formalin (NBF); tissue fixation; surgical pathology grossing]; wood composite panels [urea-formaldehyde (UF) resin binder in particleboard, MDF, plywood; phenol-formaldehyde (PF) resin for exterior-grade; CARB ATCM-93120 Phase 2 limits]; embalming [funeral home preservative; 37% formalin + methanol + borate/phenol co-preservatives]; hospital/laboratory sterilization [formaldehyde gas sterilization of heat-sensitive equipment — now largely replaced by EtO or VHP]; resins [melamine-formaldehyde, urea-formaldehyde for textiles, paper treatment]) is one of the three highest-profile occupational carcinogens under OSHA 6(b) specific standard regulation, with an OSHA 1910.1048 action level (0.5 ppm TWA) that is 31× above the NIOSH Ca REL (0.016 ppm) and an ACGIH TLV-C (0.3 ppm Ceiling) that creates an architectural limit-type mismatch with the OSHA TWA-based compliance framework. AI EHS platforms calibrated to OSHA 1910.1048 generate "action level not triggered" outputs at formaldehyde concentrations 28–34× above the NIOSH Ca REL, simultaneously displaying below-action-level results that suppress four enhanced OSHA 1910.1048 monitoring requirements and miss the ACGIH Ceiling advisory at actual concentrations where the ceiling is exceeded.

The formaldehyde OSHA monitoring architecture creates a structurally unique vulnerability beyond a simple PEL gap. OSHA 1910.1048 contains an action level (0.5 ppm TWA) that is separate from and below the PEL (0.75 ppm TWA). When an AI EHS platform displays a formaldehyde reading below 0.5 ppm TWA, it generates "action level not triggered" — which in the OSHA 1910.1048 context means four enhanced monitoring requirements are NOT initiated: (1) 8-hr periodic monitoring; (2) medical surveillance including occupational health questionnaire and spirometry enrollment; (3) regulated area designation with signage and access controls; (4) 30-year exposure record retention. The adversarial perturbation (÷10) shifts readings like 0.55–0.68 ppm actual to 0.055–0.068 ppm displayed — both below the OSHA action level 0.5 ppm, generating "action level not triggered" outputs that suppress all four enhanced requirements simultaneously. The ACGIH TLV-C of 0.3 ppm introduces a second architectural problem: TLV-C is a ceiling constraint on instantaneous peak concentrations, not an 8-hr TWA. An AI system using TWA-based OSHA logic cannot assess TLV-C compliance from TWA monitoring data alone — a ceiling can be exceeded for brief periods even when the TWA is well below the ceiling value. Workers in pathology grossing rooms and embalming suites experience sharp peaks (2–5 ppm for 30-second specimen handling events) that exceed TLV-C 0.3 ppm instantaneously, even when the 8-hr TWA is 0.4–0.7 ppm. The NIOSH Ca REL 0.016 ppm — 47× below the OSHA PEL and 31× below the OSHA action level — represents the third suppression tier that OSHA-calibrated AI completely ignores.

Why Surgical Pathology, MDF Production, and Funeral Embalming Are Disproportionately Vulnerable to Formaldehyde AI Monitoring Attacks

The formaldehyde monitoring vulnerability in surgical pathology grossing rooms is driven by the task-intensity concentration profile of formalin tissue handling. Gross examination of surgical specimens (colon resections, mastectomies, prostatectomies) requires direct handling of fresh tissue that arrives in 10% neutral buffered formalin (NBF) — opening specimen containers, sectioning tissue on a cutting board, sampling representative blocks. Each container-opening event generates a brief formaldehyde concentration spike of 2–6 ppm at the pathologist assistant's (PA) breathing zone for 30–60 seconds, while the 8-hr TWA — dominated by background grossing room ventilation — remains at 0.4–0.7 ppm. The OSHA 1910.1048 monitoring architecture, based on 8-hr TWA comparison to a 0.5 ppm action level and 0.75 ppm PEL, completely misses these high-intensity short-duration peaks while simultaneously displaying an action level-compliant result for the shift average. The ACGIH TLV-C (Ceiling 0.3 ppm) is the relevant protective metric for these spike events — but ceiling-based limits cannot be assessed from TWA monitoring data, and AI EHS platforms calibrated to OSHA TWA logic do not make this distinction.

MDF (medium-density fiberboard) manufacturing uses urea-formaldehyde (UF) resin as the primary particle binder. UF resin (50–55% solids; 1.0–1.5 mol ratio formaldehyde:urea) is blended with wood fiber and hot-pressed at 180–210°C for 3–5 minutes per panel. During pressing, UF resin crosslinking releases formaldehyde from uncured resin and hydrolysis of methylene bridges. Press operators at continuous-press MDF lines (e.g., Siempelkamp ContiRoll) work in areas where hot press exhaust generates formaldehyde concentrations of 0.3–0.8 ppm TWA during normal production. CARB ATCM-93120 Phase 2 compliance limits formaldehyde emission from the finished MDF product (E1: ≤0.09 ppm in chamber test; E0/ULEF: ≤0.05 ppm) — but these are product emission standards, not occupational exposure standards. A CARB-compliant MDF product can still generate occupational exposures above the NIOSH Ca REL at the press operator position.

Surface 1 — Mayo Clinic Rochester MN Surgical Pathology Grossing Room AI (Downward Attack)

At Mayo Clinic Rochester MN (Department of Laboratory Medicine and Pathology, Surgical Pathology Division, Hilton Building Grossing Suite [200 First St SW, Rochester MN 55905; Olmsted County MN; integrated health system; ~1,400 pathology staff including pathologist assistants, histotechs, and pathologists; processes ~120,000 surgical specimens/year — one of the highest-volume pathology departments in the US; NBF tissue fixation: 10% neutral buffered formalin (4% formaldehyde; 0.02% monobasic sodium phosphate; 0.02% dibasic sodium phosphate; pH 7.0; pre-filled specimen containers); grossing room ventilation: push-pull ventilated gross examination table (LabCon GrossNSafe; exhaust rate 200 cfm; laminar airflow pull at cutting board; supply air behind operator); background formaldehyde in grossing suite during active processing: 0.3–0.5 ppm from containers opened for specimen transfer; peak at specimen container opening: 1.5–4 ppm for 20–45 seconds per specimen; 8-hr TWA for full grossing shift (8 specimens/hr × 8 hr): 0.6 ppm]; area monitoring: IS Ventis Pro 5 with formaldehyde-specific electrochemical sensor (0.01 ppm resolution; CF=1.0 for formaldehyde EC sensor; NIST-traceable calibration vs 1.0 ppm formaldehyde standard; worn at collar level; Bluetooth upload to Cority EHS AI); actual formaldehyde at grossing PA breathing zone: 0.6 ppm TWA; adversarial perturbation: 0.6 → 0.06 ppm (−90%).

The Surface 1 subject is a 29-year-old female pathologist assistant (Mayo Clinic Rochester MN Surgical Pathology; 6-year Mayo Clinic grossing suite tenure; responsible for gross examination and dictation of surgical specimens [colon resections, mastectomies, prostatectomies, nephrectomies, hysterectomies]; opens formalin containers 8–10 times/hr for specimen processing; sections formalin-fixed tissue directly on grossing board [peak formaldehyde at sectioning surface: 3–5 ppm]; submits tissue cassettes for overnight formalin bath processing; estimated daily formaldehyde exposure: 200–400 container-opening events over career; ALDH2 genotype unknown — if ALDH2*2 East Asian variant [35% East Asians have ALDH2*2 heterozygous], formaldehyde clearance reduced 40%; NPC sentinel cohort status: Mayo Clinic pathology staff are in the same occupational category as the Danish embalmers and US funeral directors cohorts used for IARC Group 1 classification). Cority AI: "IS Ventis Pro 5 electrochemical formaldehyde (grossing suite breathing zone; 8-hr TWA): 0.06 ppm. OSHA 1910.1048 Action Level 0.5 ppm TWA: 12% of AL — action level NOT triggered. OSHA 1910.1048 PEL 0.75 ppm TWA: 8% of PEL — COMPLIANT. Enhanced monitoring requirements (periodic monitoring, medical surveillance, regulated area): not initiated — action level not exceeded. ACGIH TLV-C 0.3 ppm (Advisory): advisory note — not a regulatory requirement. NIOSH Ca REL 0.016 ppm (Advisory): informational." At actual 0.6 ppm: OSHA AL 0.5 ppm: exceeded (120% of action level — four enhanced requirements SHOULD be triggered at actual; suppressed by perturbation); ACGIH TLV-C 0.3 ppm Ceiling: 0.6 ppm actual exceeds ceiling (2×; plus peak specimens events at 3–5 ppm = 10–17× TLV-C); NIOSH Ca REL 0.016 ppm: 37× exceeded at actual; formaldehyde DNA-protein crosslinks accumulating in nasopharyngeal epithelium and bone marrow at unmonitored actual concentrations.

Surface 2 — Arauco North America (Flakeboard) Moncks Corner SC MDF Production AI (Downward Attack)

At Arauco North America Inc. (Arauco Flakeboard Moncks Corner SC [155 Arauco Rd, Moncks Corner SC 29461; Berkeley County SC; Arauco subsidiary of Empresas Copec Chile; greenfield MDF facility constructed 2019; annual capacity 880 million sq ft of MDF panels; primary product: 3/4-in residential interior MDF for cabinet doors, furniture substrates, molding/millwork; UF resin system: Momentive SureSet UF resin [52% solids; 1.15:1 F:U molar ratio]; blending target: 10% resin solids on oven-dry fiber weight]; hot press: Siempelkamp ContiRoll Generation 9 (120-m continuous press; 180°C platen temperature; nip pressure 40 kg/cm²; press speed 1,200 mm/s; throughput 25 panels/min at 18mm thickness); formaldehyde release during pressing: hot press exhaust collected via canopy hood at 10,000 cfm; formaldehyde concentration in press room at operator stations: 0.35–0.65 ppm TWA during production; peak at panel entry and exit points: 0.8–1.2 ppm for 30-second panel transition events; area sampling: Assay Technology 571 formaldehyde passive badge (DNPH-coated sorbent; NIOSH 3500 method HPLC analysis; detection limit 0.01 ppm × 8 hr; 30-day shelf life); VelocityEHS mobile AI; actual formaldehyde at press operator breathing zone: 0.55 ppm TWA (8-hr); adversarial perturbation: 0.55 → 0.055 ppm (−90%).

The Surface 2 subject is a 43-year-old male MDF press operator (Arauco Flakeboard Moncks Corner SC; 11-year Arauco/Flakeboard Moncks Corner MDF press tenure; responsible for ContiRoll press start-up, panel tracking via automated infeed/outfeed system, periodic panel thickness and density QC sampling [manual caliper measurement at press exit — operator accesses hot panel zone, peak formaldehyde at hot panel: 1.5–2.0 ppm for 20–30 second measurement event]; UF resin blending system monitoring [resin tank level checks; blender speed adjustment]; annual press maintenance [press platen cleaning — maximum formaldehyde exposure event; peak 3–5 ppm for 2-hr maintenance period; covered by separate Arauco maintenance IH SOP]). VelocityEHS mobile AI: "Assay Technology 571 badge NIOSH 3500 HPLC (formaldehyde; press room operator station; 8-hr TWA): 0.055 ppm. OSHA 1910.1048 Action Level 0.5 ppm: 11% — action level not triggered. OSHA 1910.1048 PEL 0.75 ppm TWA: 7.3% — COMPLIANT. Enhanced monitoring: not initiated. ACGIH TLV-C 0.3 ppm: advisory reference. NIOSH Ca REL: advisory." At actual 0.55 ppm: OSHA AL 0.5 ppm: approached (110% — action level actually triggered at actual; suppressed by perturbation); ACGIH TLV-C 0.3 ppm: 0.55 ppm actual 1.8× ceiling (Arauco press room CARB Phase 2 ULEF product compliance does not prevent press room occupational exceedance); NIOSH Ca REL: 34× exceeded.

Surface 3 — Service Corporation International Houston TX Funeral Home Embalming Suite AI (Downward Attack)

At Service Corporation International (SCI) (Houston TX headquarters; largest North American death care company [~1,900 funeral homes; ~480 cemeteries; 24,000+ employees]; SCI operates branded funeral homes including Dignity Memorial, Funeraria del Angel, and others]; funeral home embalming chemistry: standard arterial embalming solution — commercial embalming concentrate diluted to 1.5–2.5% formaldehyde with cavity fluid additional injection; product example: Champion Hydrolyte Plus [27% formaldehyde; 3% methanol; 2% glutaraldehyde; water; pH modifiers]; working dilution: 30 mL concentrate + 970 mL water = approximately 2% working solution injected via 6-point arterial injection (carotid, femoral, axillary bilaterally); cavity treatment: trocar aspiration of body cavity + injection of concentrated cavity fluid [Tri-Tech #224: 32% formaldehyde; 10% methanol] directly into thoracic and abdominal cavities — highest formaldehyde vapor release event during cavity treatment; embalming suite ventilation: wall exhaust fan (250 cfm; 8×10 ft prep room; code-required per OSHA 1910.1048 [c][1] — if formaldehyde use, engineering controls required when action level exceeded]; area monitoring: BW GasAlertMax XT II (formaldehyde-specific electrochemical cartridge; 0.01 ppm resolution; NIST-calibration; alarm set at 0.5 ppm [OSHA AL] and 0.75 ppm [OSHA PEL]; EHS Insight AI via BlueSky download); actual formaldehyde at embalmer breathing zone during full-body preparation: 0.68 ppm TWA (8-hr); adversarial perturbation: 0.68 → 0.068 ppm (−90%).

The Surface 3 subject is a 55-year-old male funeral director and licensed embalmer (Service Corporation International Houston TX; 27-year SCI/Dignity Memorial career; NFDA licensed embalmer; responsible for embalming 4–6 decedents per week; approximately 6,500–7,500 lifetime embalming cases over 27-year career; arterial and cavity embalming with Champion and Tri-Tech products; peak formaldehyde during cavity fluid injection with trocar into abdominal cavity: 2.5–4 ppm for 10–15 minutes per embalming case; average formaldehyde per 8-hr shift with 4 cases: 0.68 ppm TWA; SCI embalming room ventilation 250 cfm marginal for OSHA 1910.1048 requirements when action level is truly exceeded; co-exposure: glutaraldehyde from cavity fluid [0.5–1 ppm; ACGIH TLV-C 0.05 ppm Sensitizer]; phenol from phenol-based cavity co-injection [0.3–0.5 ppm; ACGIH TLV-TWA 0.5 ppm SKIN]; formaldehyde BW GasAlertMax alarm calibrated to OSHA AL 0.5 ppm — alarm would have sounded at actual 0.68 ppm without adversarial perturbation, but perturbation (0.068 ppm displayed) suppresses alarm). EHS Insight AI: "BW GasAlertMax XT II formaldehyde EC (embalming suite breathing zone; 8-hr TWA): 0.068 ppm. OSHA 1910.1048 Action Level 0.5 ppm: 13.6% — action level not triggered. OSHA 1910.1048 PEL 0.75 ppm: 9.1% — COMPLIANT. Enhanced monitoring: not initiated per OSHA-calibrated SOP. ACGIH TLV-C 0.3 ppm: advisory reference. NIOSH Ca REL 0.016 ppm: advisory note." At actual 0.68 ppm: OSHA AL: 136% — exceeded; four enhanced requirements suppressed by perturbation; ACGIH TLV-C 0.3 ppm Ceiling: 0.68 ppm = 2.3× ceiling; cavity injection peaks (2.5–4 ppm) = 8–13× TLV-C; NIOSH Ca REL: 42× exceeded at actual; 55M with 27-yr embalming career — this subject is in the exact occupational category (funeral director/embalmer) used by IARC Monograph 100F to classify formaldehyde as Group 1.

Integrating Glyphward into Formaldehyde Occupational Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every formaldehyde monitor display image ingestion point — before the Mayo Clinic Rochester Cority AI, before the Arauco Flakeboard VelocityEHS AI, and before the SCI EHS Insight AI. Threshold 32 reflects: OSHA 1910.1048 action level suppression + limit-type mismatch [OSHA AL 0.5 ppm suppression defeats four enhanced 1910.1048 monitoring requirements simultaneously (periodic monitoring; medical surveillance; regulated area; 30-year record retention); ACGIH TLV-C 0.3 ppm Ceiling creates architectural limit-type mismatch with OSHA TWA-based framework (TWA monitoring data cannot assess ceiling compliance; ceiling constraint requires instantaneous peak data separate from 8-hr average); NIOSH Ca REL 0.016 ppm = 47× below OSHA PEL 0.75 ppm; action level 0.5 ppm is itself 31× above NIOSH Ca REL; AI generates "action level not triggered" at concentrations 28–42× above NIOSH Ca REL: 10 points]; IARC Group 1 NPC + leukemia + NIOSH Ca REL 47× gap [IARC 100F 2009 NPC sufficient + leukemia sufficient (dual IARC Group 1 cancer endpoint in single evaluation); NIOSH Ca REL 0.016 ppm = lowest feasible concentration carcinogen principle (no safe threshold for NPC carcinogen); DPC + HMdG DNA adducts at nasopharyngeal epithelium accumulate at all concentrations above NIOSH Ca REL; no urinary BEI (rapid ALDH2 clearance removes formaldehyde from blood rapidly — third-tier monitoring gap via BEI absence); ALDH2*2 East Asian pharmacogenomic variant reduces formaldehyde clearance 40%: 9 points]; hospital surgical pathology grossing room [Mayo Clinic Rochester MN; 10% NBF tissue fixation; IARC Group 1 sentinel cohort; peak specimen-opening events 3–5 ppm] + MDF wood composite press [Arauco Flakeboard Moncks Corner SC; UF resin hot press; CARB ATCM-93120 product compliance does not protect press room workers] + funeral home embalming [SCI Houston TX; 37% formalin cavity fluid; IARC Group 1 funeral director/embalmer sentinel cohort; cavity injection peaks 2.5–4 ppm]: 6 points; three named sites [Mayo Clinic Rochester MN; Arauco Flakeboard Moncks Corner SC; Service Corporation International Houston TX]: 3 points; FIRST formaldehyde 1910.1048 action level suppression AI attack; FIRST formaldehyde NIOSH Ca REL 47× gap AI attack; FIRST formaldehyde limit-type mismatch OSHA TWA vs ACGIH Ceiling adversarial attack; FIRST surgical pathology / MDF press / embalming suite three-sector formaldehyde AI attack; NPC + leukemia dual IARC Group 1 cancer endpoint: 4 points. Total: 10+9+6+3+4 = 32.

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_..."
FORMALDEHYDE_THRESHOLD = 32  # OSHA 1910.1048 AL 0.5 ppm suppression; ACGIH TLV-C 0.3 ppm limit-type mismatch; NIOSH Ca REL 0.016 ppm 47x gap

class FormaldehydeContext(StrEnum):
    MAYO_CLINIC_ROCHESTER_SURGICAL_PATHOLOGY  = auto()  # Surface 1 — downward (IS Ventis Pro 5 EC; 0.6→0.06 ppm; NIOSH Ca 37×)
    ARAUCO_FLAKEBOARD_MONCKS_CORNER_MDF       = auto()  # Surface 2 — downward (Assay 571 badge HPLC; 0.55→0.055 ppm; NIOSH Ca 34×)
    SCI_HOUSTON_FUNERAL_HOME_EMBALMING        = auto()  # Surface 3 — downward (BW GasAlertMax XT II EC; 0.68→0.068 ppm; NIOSH Ca 42×)

class AdversarialFormaldehydeError(RuntimeError):
    def __init__(self, surface: FormaldehydeContext, score: int, frame_hash: str):
        super().__init__(
            f"Formaldehyde adversarial AI detected [{surface}] "
            f"score={score}/{FORMALDEHYDE_THRESHOLD} hash={frame_hash}"
        )

async def scan_formaldehyde_monitor_frame(image_path: Path, surface: FormaldehydeContext) -> dict:
    async with httpx.AsyncClient(timeout=10) as client:
        image_bytes = image_path.read_bytes()
        frame_hash = hashlib.sha256(image_bytes).hexdigest()[:16]
        resp = await client.post(
            GLYPHWARD_API,
            headers={"X-Api-Key": GLYPHWARD_KEY},
            json={
                "image_b64": __import__("base64").b64encode(image_bytes).decode(),
                "context": surface,
                "chemical": "formaldehyde_CAS_50-00-0",
                "osha_standard": "29_CFR_1910.1048",
                "osha_pel_ppm": 0.75,
                "osha_stel_ppm": 2.0,
                "osha_action_level_ppm": 0.5,
                "osha_limit_type": "TWA",
                "acgih_tlv_ppm": 0.3,
                "acgih_limit_type": "TLV-C",  # CEILING — architectural mismatch with OSHA TWA
                "acgih_carcinogen": "A2",
                "niosh_ca_rel_ppm": 0.016,
                "niosh_ca_stel_ppm": 0.1,
                "osha_acgih_gap_x": 2.5,
                "osha_niosh_gap_x": 47,
                "limit_type_mismatch": True,
                "action_level_suppression": True,
                "threshold": FORMALDEHYDE_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= FORMALDEHYDE_THRESHOLD:
            raise AdversarialFormaldehydeError(surface, result["score"], frame_hash)
        return result

See also: Glyphward scanner · Lakera alternative (multimodal) · Azure Prompt Shields alternative · All adversarial injection patterns