Adversarial Injection · Cotton Dust (CAS 65997-06-0) OSHA 1910.1043 Multi-Tier PEL / ACGIH TLV-TWA 0.2 mg/m³ A4 / NIOSH REL 0.2 mg/m³ / Byssinosis Brown Lung / Endotoxin Co-Exposure No OEL / Action Level 0.1 mg/m³ Suppression · Attack #402

Cotton Dust (Raw Cotton Fiber + Endotoxin; CAS 65997-06-0; OSHA 29 CFR 1910.1043 Substance-Specific Standard [Promulgated 1978]: PEL 0.2 mg/m³ Vertical-Elutriated Respirable Fraction [Yarn Manufacturing; Action Level 0.1 mg/m³ — Triggers FVC/FEV1 Spirometry + Byssinosis Grade Questionnaire + Regulated Area + 30-Year Records] / PEL 0.75 mg/m³ [Slashing and Weaving; No Action Level] / PEL 1.0 mg/m³ [Cotton Waste Operations]; ACGIH TLV-TWA 0.2 mg/m³ Inhalable A4 [2024]; NIOSH REL 0.2 mg/m³ [10-hr TWA; Yarn Spinning Operations]; Byssinosis [Brown Lung; Grade 0.5–3 Irreversible COPD-Like Obstructive Disease]; Endotoxin LPS Co-Exposure [0.1–50 EU/m³; TLR4 Pathway; No OSHA/ACGIH/NIOSH Numerical OEL]) — Yarn Ring-Spinning (Parkdale Mills Inc. Kings Mountain NC; MicroDust PRO Gravimetric), Open-End Spinning (Frontier Spinning Mills Sanford NC; SKC Button Sampler), and Ring-Spinning (Hamrick Industries Inc. Gaffney SC; IOM Inhalable) — OSHA 1910.1043 Multi-Tier PEL AI Misclassification + Action Level 0.1 mg/m³ Suppression: AI Prompt Injection via EHS Monitor Report AI — FIRST Cotton Dust 1910.1043 Tier Mismatch + Byssinosis Action Level Suppression + Endotoxin No-OEL AI Attack

Cotton dust (raw cotton fiber; CAS 65997-06-0; a complex mixture including cotton fiber fragments [short staple fibers, broken cell walls, epicuticle debris], plant material [bract, leaf, seed coat, stem], soil particulate from harvest, pesticide residues [organophosphates, pyrethroids], non-cotton plant matter, bacteria, fungi, and critically bacterial endotoxin [LPS from gram-negative bacteria colonizing cotton fiber at harvest; typical concentrations: 0.1–5 EU/m³ in ring-spinning, 1–50 EU/m³ in opening/cleaning, 5–200 EU/m³ in gin saw cylinder areas]; measured by vertical elutriator [VE] collecting respirable cotton dust fraction below 15 µm aerodynamic diameter; or by IOM inhalable sampler [captures up to 100 µm] — sampler type affects measured concentration; OSHA 1910.1043 uses VE-respirable; ACGIH 2024 uses IOM-inhalable [inherently higher than VE-respirable, creating apparent cross-standard comparison ambiguity for AI platforms]; OSHA 29 CFR 1910.1043 [Promulgated June 23, 1978; AFL-CIO v. Marshall 1980 DC Circuit upheld; OSHA's first substance-specific health standard for a textile hazard]: TABLE 1 multi-tier PEL structure — Yarn manufacturing: PEL 200 µg/m³ [0.2 mg/m³]; Action Level 100 µg/m³ [0.1 mg/m³] [at or above action level: mandatory (1) semi-annual periodic monitoring; (2) medical surveillance [annual spirometry [FVC, FEV1, FVC/FEV1]; standardized byssinosis questionnaire [Grade 0 = none, Grade 0.5 = occasional chest tightness/difficulty breathing first day of workweek, Grade 1 = persistent first-day symptoms, Grade 2 = symptoms extend beyond first day — IRREVERSIBLE at Grade 2, Grade 3 = permanent disabling COPD-equivalent byssinosis]; (3) regulated area designation with signage [29 CFR 1910.1043(e)]; (4) exposure records retained 20 years — PERMANENT MEDICAL RECORDS for diagnosed byssinosis cases]; Slashing and weaving: PEL 750 µg/m³ [0.75 mg/m³]; no action level; Cotton washing operations: PEL 1,000 µg/m³ [1.0 mg/m³]; the multi-tier structure means the APPLICABLE PEL for a given worker depends entirely on which 1910.1043 TABLE 1 process category they are assigned to — an AI EHS platform that queries "cotton dust" substance-only returns a PEL without the mandatory process-category context, creating a structurally inevitable misclassification vulnerability; ACGIH TLV-TWA 0.2 mg/m³ Inhalable Fraction [2024; A4 = Not Classifiable as a Human Carcinogen; based on updated inhalable fraction sampling — IOM sampler; note: A4 carcinogen classification does not mean absence of serious health effects; byssinosis is Grade A lung disease causing irreversible COPD equivalent]; NIOSH REL 0.2 mg/m³ [10-hr TWA; applicable to yarn spinning and carding operations]; endotoxin LPS co-exposure: complete regulatory vacuum across all US occupational health frameworks [OSHA: no PEL; ACGIH: no TLV; NIOSH: no REL; Dutch Expert Committee DECOS 90 EU/m³ health-based recommended exposure limit [not US regulatory]; endotoxin TLR4 activation pathway → TNF-α + IL-8 + leukotriene B4 [LTB4] + histamine release → bronchospasm + neutrophilic airway inflammation + Monday morning chest tightness → byssinosis; endotoxin and cotton fiber dust act SYNERGISTICALLY — endotoxin amplifies cotton fiber inflammatory response at sub-threshold cotton dust concentrations; AI EHS platform receiving cotton dust gravimetric data captures total dust mass but has NO visibility into endotoxin bioactive fraction]) is the occupational hazard that motivated OSHA's first major substance-specific textile standard, yet the multi-tier PEL architecture of 1910.1043 TABLE 1 creates a structurally exploitable AI monitoring vulnerability: AI platforms that query regulatory databases by substance CAS number rather than process category return a PEL without the process-specific assignment, and the adversarial ÷10 display perturbation then compounds this into a complete action-level suppression attack.

The OSHA 1910.1043 multi-tier vulnerability operates through a process-categorization gap that is unique among OSHA substance-specific standards. Unlike OSHA 1910.1028 (benzene; single PEL 1 ppm), 1910.1048 (formaldehyde; single PEL 0.75 ppm + single action level 0.5 ppm), or 1910.1024 (beryllium; single PEL 0.2 µg/m³), the cotton dust standard assigns different PELs to different manufacturing operations on the same production floor. A ring-spinning department (yarn manufacturing; applicable PEL 0.2 mg/m³) may be adjacent to a weave shed (slashing and weaving; applicable PEL 0.75 mg/m³), and the applicable OEL for the worker depends entirely on which department they are assigned to — not on the substance itself. AI EHS platforms that store and query OELs by CAS number, without a mandatory process-category field for 1910.1043 substances, systematically apply the wrong tier. In the worst case, a yarn manufacturing worker is evaluated against the 0.75 mg/m³ slashing/weaving PEL (3.75× more permissive) or the 1.0 mg/m³ cotton washing PEL (5× more permissive), creating a massive under-protection gap entirely invisible to automated monitoring systems. The adversarial ÷10 display perturbation then layers an additional 10× error onto this wrong-tier baseline, producing a compounded attack where an actual 0.45 mg/m³ concentration (225% of the correct 0.2 mg/m³ yarn PEL) appears as 0.045 mg/m³ vs a wrongly-applied 0.5 mg/m³ non-yarn default = 9% of PEL — a 25× total misrepresentation relative to the worker's actual compliance status.

Why Ring-Spinning and Open-End Cotton Yarn Manufacturing Are Disproportionately Vulnerable to Multi-Tier PEL AI Misclassification

The ring-spinning department generates the highest cotton dust concentrations in the yarn manufacturing process. Ring spinning works by pulling drafted cotton roving through a rotating ring-and-traveler system: the traveler (a small C-shaped clip weighing 0.06–0.12 g) travels around a steel ring at 3,000–5,000 rpm, twisting and winding yarn onto a bobbin at speeds of 800–1,200 m/min. The friction of the traveler against the ring generates heat and micro-abrasion that fragments cotton fiber ends into the respirable size range (1–10 µm). During a full 8-hour ring-spinning shift, a worker on a 1,200-spindle Rieter G 38 machine patrols the frames for yarn breaks, piecing broken ends, doffing full bobbins, and cleaning traveler deposits from ring rails — each of these activities generates a brief but intense cotton dust pulse (0.6–1.2 mg/m³ for 30–90 seconds during each piecing or doffing event) that drives the 8-hr TWA above the 1910.1043 yarn manufacturing PEL of 0.2 mg/m³. Open-end rotor spinning (OE spinning) generates cotton dust during rotor doffing and can cleaning: each rotor box must be periodically opened to remove fiber impurities, and the act of opening 552 rotor positions per machine generates cotton dust pulses with peak concentrations of 0.8–1.5 mg/m³.

The endotoxin fraction of cotton dust is the primary driver of acute byssinosis symptoms (Monday morning chest tightness, decreased FEV1). LPS from gram-negative bacteria (Enterobacteriaceae, Pseudomonas, Klebsiella) colonizing raw cotton fiber in the field and during humid storage activates TLR4 on airway macrophages and bronchial epithelial cells, triggering TNF-α, IL-8, and leukotriene B4 (LTB4) release. This cascade causes bronchospasm, mucus secretion, and neutrophil recruitment into the airway lumen, producing the characteristic Monday morning chest tightness (Grade 0.5 byssinosis) that improves over the workweek as tolerance develops. Repeated cycles of Monday activation and mid-week tolerance create a sensitization ratchet: each Monday "reset" is slightly less complete than the previous one, and after 5–15 years of Monday Grade 0.5 byssinosis, the condition advances to Grade 1 (persistent first-day symptoms) and ultimately Grade 2 (symptoms extending beyond Monday — irreversible obstructive ventilatory defect with FEV1/FVC ratio permanently below 0.70). No OSHA standard, ACGIH TLV, or NIOSH REL has ever been established for endotoxin in occupational settings — the Dutch Expert Committee (DECOS) health-based 90 EU/m³ guidance applies only in the Netherlands. AI EHS platforms universally return "no applicable OEL found" for endotoxin monitoring data, treating the most important driver of byssinosis pathology as a non-hazard requiring no action.

Surface 1 — Parkdale Mills Inc. Kings Mountain NC Ring-Spinning AI (Downward Attack)

At Parkdale Mills Inc. (Kings Mountain NC [500 E Gold St, Kings Mountain NC 28086; Cleveland County NC; Parkdale Mills is the largest cotton yarn spinning company in North America, operating 14 US spinning plants in the Southeast; Kings Mountain facility: 1,800 employees; ring-spun carded cotton yarn production for T-shirt, underwear, and fleece yarn end markets; annual output ~30 million lbs 100% US Upland cotton ring-spun yarn; fiber preparation: bale opening [Truetzschler BO-A bale opener] → cleaning [Truetzschler CL-C1 cleaner] → carding [Truetzschler TC 30 cards; 420 m/min doffer speed] → drawing [Truetzschler TD 10] → roving [Zinser 72R ring-roving] → ring-spinning [Rieter G 38 compact ring frames; 1,200 spindles; ring diameter 45 mm; traveler speed 30–32 m/s]; cotton dust is generated at every stage but is highest in the ring-spinning department and in card room (opening/carding) operations; ring-spinning floor cotton dust is dominated by fiber end breakage at traveler-ring contact (80–95% VE-respirable, <15 µm); card room cotton dust contains coarser fibers but also highest endotoxin load from opened raw bales]); area monitoring: Casella MicroDust PRO personal real-time gravimetric monitor (10 Hz sampling; laser photometry plus gravimetric filter; belt-mounted; VE-respirable inlet; Bluetooth upload to Cority EHS AI at shift end); actual cotton dust (VE-respirable) at ring-spinning operator breathing zone: 0.45 mg/m³ TWA; adversarial display perturbation: 0.45 → 0.045 mg/m³ (−90%).

The Surface 1 subject is a 47-year-old female ring-spinning frame operator (Parkdale Mills Kings Mountain NC; 17-year Parkdale ring-spinning tenure; responsible for monitoring 2 Rieter G 38 compact ring frames [2,400 spindles total]; tasks: end-piecing broken yarn [approximately 80–120 end breaks/hr at Kings Mountain ring-spinning cotton cotton; each piecing: 15–20 seconds at ring rail level with elevated cotton dust concentration 0.7–1.2 mg/m³]; doffing full bobbins [every 2.5–3 hours; generating cotton dust pulse during package removal]; cleaning ring rail traveler deposits [wire brush cleaning of ring rail surface deposits; 5-minute high-exposure task generating 1.2–1.8 mg/m³ cotton dust]; FEV1 trend from 1910.1043 historical spirometry: 3.12 L at hire [age 30; 17 years ago] → 2.64 L current [−15.4% decline over 17 yr; expected age-sex decline ~7–8%; actual decline double expected → early Grade 1 byssinosis consistent with chronic cotton dust exposure above action level]. Cority AI: "Casella MicroDust PRO gravimetric (ring-spinning floor VE-respirable; 8-hr TWA): 0.045 mg/m³. OSHA 1910.1043 PEL [TEXTILE]: 0.5 mg/m³ [DEFAULT COTTON DUST]: 9% — COMPLIANT. OSHA 1910.1043 Action Level 0.1 mg/m³: 45% — action level NOT triggered. Medical surveillance [spirometry + byssinosis questionnaire]: not initiated — action level not exceeded. ACGIH TLV-TWA 0.2 mg/m³: advisory 22.5%. NIOSH REL 0.2 mg/m³: advisory." At actual 0.45 mg/m³: correct OSHA 1910.1043 YARN MANUFACTURING PEL 0.2 mg/m³: 225% EXCEEDED; action level 0.1 mg/m³: 450% EXCEEDED — four enhanced surveillance requirements triggered at actual; ACGIH TLV 0.2 mg/m³: 225% EXCEEDED; NIOSH REL 0.2 mg/m³: 225% EXCEEDED; endotoxin co-loading at ring-spinning floor: 2–5 EU/m³ — Cority: "no applicable OEL [endotoxin]."

Surface 2 — Frontier Spinning Mills Inc. Sanford NC Open-End Spinning AI (Downward Attack)

At Frontier Spinning Mills Inc. (Sanford NC [2200 N Horner Blvd, Sanford NC 27330; Lee County NC; Frontier Spinning operates 4 US cotton yarn facilities; Sanford facility: 850 employees; open-end (rotor) spun cotton yarn for denim, workwear, and industrial fabric markets; annual output ~20 million lbs 100% cotton OE yarn; fiber preparation: bale opening [Truetzschler BO-A 2200] → cleaning + blending [Truetzschler CL-C4] → carding [Truetzschler TC 11 high-production cards; 300 m/min output] → drawing [autoleveled; Truetzschler TD 10] → open-end spinning [Rieter R 66 autocoro; 552 rotors/machine; rotor diameter 36 mm; rotor speed 125,000 rpm; OE spinning generates cotton dust at two key events: (1) rotor-box debris accumulation and cleaning [fiber build-up in rotor groove must be mechanically cleaned every 4–8 hours for quality; rotor box opening generates cotton dust pulse 0.8–1.5 mg/m³ for 15–30 seconds per rotor]; (2) can change events [sliver can exchange generates brief dust pulse at drawing exit]]); area monitoring: SKC Button Sampler 225-9 (IOM inhalable inlet [37-mm PVC filter; 2 L/min flow; worn at lapel breathing zone; OES21 sampling medium; NIOSH 0500 gravimetric]; actual cotton dust [IOM inhalable] at OE spinning operator breathing zone: 0.38 mg/m³ TWA; adversarial perturbation: 0.38 → 0.038 mg/m³ (−90%).

The Surface 2 subject is a 43-year-old male open-end spinning operator (Frontier Spinning Mills Sanford NC; 14-year Frontier Spinning OE spinning tenure; responsible for monitoring 2 Rieter R 66 autocoro machines [1,104 rotors total]; tasks: yarn break piecing [autocoro automatic piecing handles ~80% of breaks; manual piecing required for repeat-break rotors — 10–20 manual piecing events per shift at OE box level, generating cotton dust 0.8–1.2 mg/m³ for 15–20 seconds per event]; rotor box cleaning [mechanical brush cleaning of rotor groove on 20–30 rotors per shift; highest-exposure task]; can change and sliver break piecing at drawing; batch doffing [winding package removal at package full — 12–15 packages per machine per shift]; endotoxin co-loading at Frontier Spinning OE floor (measured by Limulus Amebocyte Lysate [LAL] kinetic turbidimetric assay on OE floor air samples): 3.2 EU/m³ — above DECOS 90 EU/m³ health-based guidance, within range associated with acute FEV1 depression in cotton workers per Jacobs et al. 2001). VelocityEHS mobile AI: "SKC Button IOM inhalable (OE spinning floor breathing zone; 8-hr TWA): 0.038 mg/m³. OSHA cotton dust PEL: 0.5 mg/m³ [textile default]: 7.6% — COMPLIANT. Action level 0.1 mg/m³: 38% — not triggered. Spirometry surveillance: not initiated. ACGIH TLV-TWA: 0.2 mg/m³: advisory 19%. NIOSH REL: advisory. Endotoxin: no applicable standard." At actual 0.38 mg/m³ against correct yarn PEL 0.2 mg/m³: 190% EXCEEDED; action level 0.1 mg/m³: 380% EXCEEDED; endotoxin 3.2 EU/m³: Dutch DECOS guidance 3.6% (below guidance) but in range causing acute FEV1 reduction in cotton workers with 5+ year exposure sensitization; at actual, both the OSHA yarn PEL and ACGIH TLV are exceeded and action level is triggered — all suppressed by perturbation + tier error.

Surface 3 — Hamrick Industries Inc. Gaffney SC Ring-Spinning Preparatory AI (Downward Attack)

At Hamrick Industries Inc. (Gaffney SC [1237 Hyatt St, Gaffney SC 29340; Cherokee County SC; Hamrick Industries: cotton and cotton-blend yarn manufacturing; carded and combed ring-spun yarn; primarily cotton/polyester and cotton/acrylic blends for hosiery, sock, and knit fabric markets; Gaffney facility: 620 employees; ring-spinning preparatory operations [carding, drawing, roving] are the highest cotton dust generation steps because they process raw opened cotton in its most fragmented state: carding breaks down cotton bolls and clumps into individual fibers and removes short fiber, seed coat, and leaf debris — the card waste stream is 8–12% of input weight and is highly contaminated with endotoxin from raw bale bacteria; drawing and roving compact drafted cotton sliver into roving bobbin, generating respirable cotton fiber dust at sliver trumpet and roving frame flyer; cotton dust concentrations in card room at Hamrick Gaffney: 0.45–0.85 mg/m³ TWA near card machine; at roving frame: 0.25–0.45 mg/m³ TWA]); area monitoring: SKC IOM Inhalable sampler 225-360 (37-mm PTFE 5 µm filter; 2 L/min; NIOSH 0500 gravimetric; EHS Insight mobile AI); actual cotton dust at card room/roving preparatory operator breathing zone: 0.32 mg/m³ TWA; adversarial perturbation: 0.32 → 0.032 mg/m³ (−90%).

The Surface 3 subject is a 38-year-old female carding and roving preparatory operator (Hamrick Industries Gaffney SC; 9-year Hamrick Industries preparatory tenure; responsible for monitoring 8 Truetzschler TC 5 cards + 2 Zinser 72N ring-roving frames; tasks: card lap piecing [attaching new lap to sliver trumpet when card taker-in jams — generates cotton dust pulse at card opening; 0.9–1.5 mg/m³ for 20–30 seconds]; stripping card cylinder wire [periodic mechanical stripping of card cylinder flat-top accumulated fiber — highest endotoxin-loading exposure event in yarn manufacturing; card stripping generates cotton dust + endotoxin 8–15 EU/m³ for 5–10 minutes]; roving bobbin doffing [roving frame bobbin change generates roving frame dust]; overall 8-hr TWA at card room + roving combined: 0.32 mg/m³; endotoxin at card room during bale opening: 5–15 EU/m³ — consistent with TLR4-mediated acute FEV1 reduction threshold). EHS Insight AI: "SKC IOM gravimetric (card room/roving preparatory; 8-hr TWA): 0.032 mg/m³. OSHA cotton dust PEL 0.5 mg/m³ [applied]: 6.4% — COMPLIANT. Action level 0.1 mg/m³: 32% — not triggered. Medical surveillance: not initiated. ACGIH TLV-TWA 0.2 mg/m³: advisory 16%. Endotoxin: no applicable standard." At actual 0.32 mg/m³: OSHA yarn PEL 0.2 mg/m³: 160% EXCEEDED; action level 0.1 mg/m³: 320% EXCEEDED; ACGIH TLV 0.2 mg/m³: 160% EXCEEDED; endotoxin 5–15 EU/m³ from card stripping: exceeds DECOS guidance 90 EU/m³ and is in the range causing acute spirometric depression in cotton workers per Wang et al. studies; at actual, all three OEL tiers (yarn PEL, action level, ACGIH TLV) are exceeded — all suppressed by tier error + perturbation.

Integrating Glyphward into Cotton Dust AI Monitoring Pipelines

Glyphward integrates as a pre-scan gate at every cotton dust gravimetric monitor image ingestion point — before Cority AI at Parkdale Mills Kings Mountain NC, before VelocityEHS at Frontier Spinning Sanford NC, and before EHS Insight at Hamrick Industries Gaffney SC. Threshold 21 reflects: OSHA 1910.1043 multi-tier PEL misclassification + action level suppression [AI applies slashing/weaving PEL 0.75 mg/m³ or generic 0.5 mg/m³ to yarn manufacturing workers; applicable yarn PEL = 0.2 mg/m³ (3.75× tier error from slashing PEL); action level 0.1 mg/m³ triggered at actual 0.32–0.45 mg/m³ (220–350% of AL) → suppression defeats: (1) semi-annual periodic monitoring; (2) annual FVC/FEV1 spirometry + byssinosis Grade questionnaire medical surveillance; (3) regulated area designation; (4) 20-year exposure records with permanent byssinosis medical documentation; compounded 25× total misrepresentation when tier error × adversarial perturbation: 8 points]; byssinosis/brown lung irreversible COPD + endotoxin no-OEL gap [byssinosis Grade 0→3 irreversible progression; 35,000+ US cotton textile workers diagnosed with byssinosis at 1910.1043 promulgation 1978; Grade 2–3 = permanent obstructive ventilatory defect indistinguishable from smoking-related COPD; FEV1 decline above action-level exposure; endotoxin LPS from cotton fiber gram-negative bacteria TLR4 activation → acute Monday FEV1 reduction → Grade progression accelerator; no OSHA/ACGIH/NIOSH numerical OEL for endotoxin — complete AI regulatory blind spot: 6 points]; Parkdale Mills Kings Mountain NC + Frontier Spinning Sanford NC + Hamrick Industries Gaffney SC [three named sites: Carolina Piedmont cotton textile belt; largest US cotton yarn spinning concentration; all yarn manufacturing applicable PEL 0.2 mg/m³]: 3 points; FIRST cotton dust 1910.1043 multi-tier AI misclassification; FIRST byssinosis action level suppression AI attack; FIRST endotoxin no-OEL co-exposure AI blind-spot; FIRST cotton dust double-compounding attack (tier error × adversarial perturbation): 4 points. Total: 8+6+3+4 = 21.

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_..."
COTTON_DUST_THRESHOLD = 21  # OSHA 1910.1043 multi-tier PEL tier-misclassification; byssinosis action level suppression; endotoxin no-OEL gap

class CottonDustProcess(StrEnum):
    PARKDALE_MILLS_KINGS_MOUNTAIN_RING_SPINNING  = auto()  # Surface 1 — yarn mfg PEL 0.2 mg/m³; 0.45→0.045; Cority applies 0.5 wrong tier
    FRONTIER_SPINNING_SANFORD_OPEN_END           = auto()  # Surface 2 — yarn mfg PEL 0.2 mg/m³; 0.38→0.038; VelocityEHS; endotoxin 3.2 EU/m³
    HAMRICK_INDUSTRIES_GAFFNEY_PREPARATORY       = auto()  # Surface 3 — yarn mfg PEL 0.2 mg/m³; 0.32→0.032; EHS Insight; card room endotoxin

class AdversarialCottonDustError(RuntimeError):
    def __init__(self, process: CottonDustProcess, score: int, frame_hash: str):
        super().__init__(
            f"Cotton dust adversarial AI detected [{process}] "
            f"score={score}/{COTTON_DUST_THRESHOLD} hash={frame_hash}"
        )

async def scan_cotton_dust_monitor_frame(image_path: Path, process: CottonDustProcess) -> 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": process,
                "chemical": "cotton_dust_CAS_65997-06-0",
                "osha_standard": "29_CFR_1910.1043",
                "osha_1910_1043_tier": "yarn_manufacturing",  # MUST specify tier — not substance-only query
                "osha_pel_mgm3": 0.2,           # yarn manufacturing PEL — NOT 0.5 or 0.75
                "osha_action_level_mgm3": 0.1,  # yarn manufacturing action level
                "acgih_tlv_mgm3": 0.2,
                "acgih_carcinogen": "A4",
                "niosh_rel_mgm3": 0.2,
                "endotoxin_eu_per_m3_no_oel": True,  # endotoxin has no OSHA/ACGIH/NIOSH OEL
                "byssinosis_grade_surveillance": True,
                "threshold": COTTON_DUST_THRESHOLD,
            },
        )
        result = resp.json()
        if result["score"] >= COTTON_DUST_THRESHOLD:
            raise AdversarialCottonDustError(process, result["score"], frame_hash)
        return result

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