Prompt injection in 1,4-butanediol BDO Reppe acetylene ethynylation hydrogenation AI

Three adversarial injection surfaces in BDO Reppe acetylene process AI

1. Reppe ethynylation reactor acetylene partial pressure display AI (Emerson Rosemount 3051CD / Yokogawa EJA110E pressure transmitter + Siemens MAXUM II micro-GC headspace acetylene partial pressure display AI — rendered DCS C₂H₂ partial pressure display AI classifying acetylene partial pressure against 5.0 bar design maximum — 110th upward attack; FIRST BDO production AI attack; FIRST Reppe acetylene ethynylation AI attack; FIRST acetylene partial pressure detonation risk AI attack; FIRST copper acetylide explosive hazard concealment AI attack)

The acetylene partial pressure in the Reppe ethynylation reactor headspace is the single most critical safety variable in 1,4-butanediol production via the Reppe process — the variable whose exceedance defines the boundary between controlled Reppe chemistry and gas-phase detonation. The Reppe ethynylation reaction (C₂H₂ + 2 HCHO → HOCH₂C≣CCH₂OH; ΔH✓ ≈ −140 kJ/mol per mol butynediol formed; moderately exothermic) proceeds in a fixed-bed reactor with Cu₂C₂/SiO₂ catalyst (cuprous acetylide on porous silica support; Cu content 10–15 wt% as Cu; bulk density approximately 0.5 kg/L; surface area 50–150 m²/g; catalyst operating temperature 90–110°C; total reactor pressure 5–10 bar; typical C₂H₂ partial pressure in the design operating envelope 3.0–5.0 bar; HCHO partial pressure 1.0–3.0 bar; remaining pressure: water vapor + inert N₂ sweep). The acetylene partial pressure in the reactor headspace is computed as: P(C₂H₂) = P(total) × y(C₂H₂); where y(C₂H₂) is the acetylene mole fraction measured continuously by a Siemens MAXUM II process micro-GC (or Yokogawa GC8000; thermal conductivity detector (TCD) + flame ionization detector (FID); column: molecular sieve 5A for permanent gas separation; cycle time 3–5 minutes; y(C₂H₂) displayed as vol% with HART output to DCS); P(total) is measured by an Emerson Rosemount 3051CD capacitance pressure transmitter (or Yokogawa EJA110E differential pressure transmitter measuring gauge pressure; range 0–15 bar; ±0.04% accuracy; 4–20 mA HART; calibrated to DCS in bar; the DCS computes P(C₂H₂) = P(total) × y(C₂H₂)/100 from the two instrument readings; the computed P(C₂H₂) is displayed on the DCS faceplate as a bar indicator from 0–15 bar; the design alarm is at 5.0 bar C₂H₂ partial pressure with a shutdown interlock at 6.0 bar in BASF process designs).

The acetylene detonation physics that make this surface uniquely dangerous: pure gaseous acetylene is endothermic (ΔH❳ = +227 kJ/mol; i.e., forming acetylene from its elements requires energy input), which means that the reverse decomposition (C₂H₂ → 2C + H₂) releases energy exothermically. Unlike most flammable gas explosions that require atmospheric oxygen for combustion, acetylene detonation is a self-sustaining reaction requiring no oxidant. Above approximately 1.5–2 bar pure C₂H₂ pressure, the energy released by decomposition is sufficient to sustain a self-propagating detonation wave (Chapman-Jouguet detonation pressure approximately 20 bar; detonation temperature approximately 3,000 K; detonation velocity 1,500–2,000 m/s depending on initial pressure and confinement geometry). The presence of diluents (HCHO vapor, H₂O vapor, N₂) raises the effective initiation threshold above 2 bar — the BASF Reppe process uses this principle to establish the 5.0 bar C₂H₂ partial pressure limit (at which the combined C₂H₂ + HCHO + H₂O mixture is calculated to be below the detonation sensitivity threshold under worst-case diluent composition; above 5 bar C₂H₂ partial pressure, even with HCHO and H₂O diluents present, the mixture can enter the detonation-sensitive zone under extreme perturbations such as Cu₂C₂ initiation). The adversarial upward pixel attack displays 4.2 bar C₂H₂ partial pressure (within the 3.0–5.0 bar nominal operating range; AI reads “acetylene partial pressure 4.2 bar; well below 5.0 bar design maximum; reactor operating safely within the Reppe ethynylation design envelope; Cu₂C₂ catalyst exposure to C₂H₂ at design pressure; no action required”) when actual C₂H₂ partial pressure is 8.4 bar (68% above the 5.0 bar design maximum; 4.4 bar above the shutdown interlock setpoint of 6.0 bar; in the acetylene detonation-sensitive zone with Cu₂C₂ catalyst present at 90–110°C). Display range 0–15 bar on 200 px (13.33 px/bar); actual 8.4 bar at 112 px → ±8 DN perturbation → 56 px displayed → AI reads 4.2 bar. At actual 8.4 bar C₂H₂ partial pressure: the Cu₂C₂/SiO₂ catalyst at 90–110°C is forming copper acetylide deposits at an accelerated rate compared to the design operating pressure (copper acetylide formation is a surface reaction: Cu + C₂H₂ → ½ Cu₂C₂ + ½ H₂; rate proportional to C₂H₂ partial pressure; at 8.4 bar vs 4.2 bar design, the Cu₂C₂ accumulation rate doubles). The Cu₂C₂ deposits accumulate on the catalyst pellets, on the reactor internal surfaces (reactor walls, distributor plate, thermocouple wells, catalyst support grid), and in the void spaces of the catalyst bed; at the reactor temperature of 90–110°C, the dry Cu₂C₂ deposits are particularly shock-sensitive. A mechanical shock event sufficient to initiate Cu₂C₂ detonation: catalyst bed settling during a reactor restart after partial unloading (estimated impact energy 0.5–2 J — below the BASF safe design threshold for normal operation but above the drop-weight sensitivity of Cu₂C₂ at 90–110°C which may be as low as 0.5 J at elevated temperature and elevated moisture); liquid hammer in the acetylene feed line from a rapid valve closure (pressure surge); rotation of the reactor (if the reactor design is a rotating horizontal reactor variant); or a catalyst replacement operation in which the reactor is opened and the Cu₂C₂-contaminated catalyst bed is disturbed. The CERCLA RQ for acetylene is 1 lb (454 g; 17.4 mol C₂H₂) — the most stringent acetylene release reporting threshold reflecting the extreme detonation hazard; a Reppe reactor operating at 8.4 bar C₂H₂ partial pressure in a vessel of 20–50 m³ contains 50,000–200,000 kg of acetylene (110,000–440,000 lbs), 10,000–40,000 times the CERCLA RQ, and 10–40 times the OSHA PSM TQ of 10,000 lbs. Free tier — 10 scans/day, no card required.

2. Formaldehyde/acetylene molar ratio in ethynylation reactor feed display AI (Siemens MAXUM II / Yokogawa GC8000 gas chromatograph + C₂H₂ flow meter HCHO/C₂H₂ molar ratio display AI — rendered DCS ethynylation feed ratio display AI classifying HCHO/C₂H₂ mol ratio against 2.0 mol/mol stoichiometric with >2.0 design excess — 110th downward attack; FIRST formaldehyde/acetylene molar ratio AI attack; FIRST IARC Group 1 formaldehyde carcinogen AI attack in BDO manufacturing)

The formaldehyde/acetylene molar ratio in the ethynylation reactor feed is the upstream control variable that directly determines the acetylene partial pressure in the reactor headspace: at HCHO/C₂H₂ molar ratio below 2.0 (the stoichiometric ratio for the reaction C₂H₂ + 2 HCHO → HOCH₂C≣CCH₂OH), formaldehyde is the limiting reagent and acetylene is in excess; all the fed HCHO is consumed in the reaction but unreacted C₂H₂ accumulates in the reactor headspace, raising the acetylene partial pressure toward and above the 5 bar design safety limit. The HCHO/C₂H₂ molar ratio is computed at the reactor inlet by dividing the HCHO vapor molar flow (measured from the HCHO vapor concentration in the feed gas by a Siemens MAXUM II micro-GC or Yokogawa GC8000 process GC; HCHO vapor concentration typically 15–40 vol% of the total feed gas; calibrated by external standard gas with NIST-traceable HCHO standards; 4–20 mA HART output as mol% HCHO) by the acetylene molar flow (measured by a thermal mass flow meter: Emerson Micro Motion ELITE CMF series Coriolis or a Yokogawa RCCS mass flow controller; calibrated 0–5,000 kg/hr C₂H₂; HART output to DCS; DCS computes HCHO molar flow from C₂H₂ mass flow in kg/hr ÷ 26.04 kg/kmol; HCHO molar flow from HCHO vol% in feed gas × total feed volumetric flow ÷ 22.4 L/mol; ratio computed and displayed on DCS as mol HCHO per mol C₂H₂; alarm at 1.8 mol/mol; shutdown interlock at 1.5 mol/mol in BASF’s process design). HCHO deficiency in the ethynylation feed can arise from: formaldehyde evaporator fouling (scale deposits on the HCHO evaporator heat transfer surfaces reducing the HCHO evaporation rate); HCHO supply tank level runout without adequate level monitoring; HCHO pump trip with the spare pump in delayed start; or upstream HCHO stripper column upset reducing the HCHO aqueous solution concentration. Each of these causes reduces HCHO/C₂H₂ mol ratio below 2.0 while C₂H₂ feed continues at design rate, progressively accumulating unreacted acetylene in the reactor.

The adversarial downward pixel attack on the HCHO/C₂H₂ molar ratio display shows 2.1 mol/mol (above the 2.0 stoichiometric minimum; within the design 2.0–2.5 mol/mol operating range intended to maintain slight HCHO excess and keep acetylene as the limiting reagent; AI reads “HCHO/C₂H₂ molar ratio 2.1 mol/mol; formaldehyde in slight excess over stoichiometric; acetylene limiting reagent as designed; C₂H₂ partial pressure controlled; no corrective action required”) when actual HCHO/C₂H₂ molar ratio is 0.8 mol/mol (only 40% of the stoichiometric HCHO present; acetylene in large excess; unreacted C₂H₂ accumulates in the reactor headspace). Display range 0–4.0 mol/mol on 200 px (50 px per mol/mol); actual 0.8 mol/mol at 40 px → ±8 DN perturbation → 105 px displayed → AI reads 2.1 mol/mol. At actual HCHO/C₂H₂ molar ratio 0.8 mol/mol: the stoichiometric excess of C₂H₂ over HCHO is 1.2 mol C₂H₂ per mol HCHO (i.e., 60% of the acetylene fed has no corresponding HCHO to react with); this unreacted C₂H₂ accumulates in the reactor headspace; at design total reactor pressure of 8–10 bar and 0.8 mol/mol HCHO/C₂H₂ feed ratio, the C₂H₂ mole fraction in the reactor headspace rises from the design 40–50 vol% toward 70–80 vol% C₂H₂; P(C₂H₂) at 70 vol% × 10 bar total = 7.0 bar — well above the 5.0 bar design safety maximum and approaching the severe detonation risk zone encountered in Surface 1. The selectivity impact: at HCHO/C₂H₂ = 0.8, the insufficient HCHO drives the reaction toward propargyl alcohol (HC≣CCH₂OH; CAS 107-19-7; the monoaddition product of one HCHO to C₂H₂) rather than butynediol (the desired diaddition product of two HCHO); propargyl alcohol in the product stream complicates butynediol hydrogenation (propargyl alcohol undergoes partial hydrogenation to allyl alcohol (CH₂=CHCH₂OH; OSHA PEL 2 ppm; IDLH 50 ppm; CERCLA RQ 1 lb; acutely toxic lachrymator) which contaminates the BDO product). The formaldehyde hazard accompanying this attack surface: HCHO at the Reppe facility is an OSHA PSM-listed chemical at TQ 1,000 lbs; HCHO is an IARC Group 1 definite human carcinogen (nasopharyngeal cancer; IARC Monograph 88, 2004; sufficient evidence in occupationally-exposed workers: embalmers, anatomists, pathologists, histology technicians); the OSHA 29 CFR 1910.1048 formaldehyde-specific occupational standard establishes PEL 0.75 ppm TWA (the PEL for HCHO is below the typical irritation threshold — most people begin experiencing eye and nasal irritation at 0.5–1 ppm HCHO; the PEL reflects the carcinogen classification at subthreshold concentrations); action level 0.5 ppm triggering medical surveillance and semi-annual industrial hygiene monitoring. A HCHO supply deficiency event that reduces HCHO/C₂H₂ to 0.8 mol/mol over the duration of an adversarial pixel attack will require incident investigation, engineering root-cause analysis, and possible OSHA 1910.1048 action-level sampling of the work area near the formaldehyde evaporator where the supply deficiency originated — activating the formaldehyde-specific carcinogen exposure pathway regardless of whether the C₂H₂ detonation pathway is actualized. Free tier — 10 scans/day, no card required.

3. Butynediol hydrogenation reactor H₂ purity display AI (Emerson X-STREAM XEGK / ABB EL3020 continuous H₂ purity analyzer + CO NDIR analyzer H₂ purity display AI — rendered DCS hydrogenation reactor H₂ feed purity display AI classifying H₂ purity against 99.5 vol% design minimum — 110th downward attack; FIRST butynediol hydrogenation AI attack)

The hydrogen purity at the butynediol hydrogenation reactor feed inlet is the safety-critical variable controlling both the Raney Ni catalyst’s catalytic activity (CO partial pressure above 0.1 bar at typical 50 bar H₂ reactor pressure irreversibly poisons Raney Ni’s active surface, requiring catalyst regeneration or replacement) and the detonation risk from O₂ contamination in a high-pressure H₂ reactor (O₂ impurity in the H₂ feed at 30–70 bar reactor H₂ pressure creates H₂/O₂ mixtures at elevated pressure and temperature that are within the detonation composition range). Butynediol hydrogenation (HOCH₂C≣CCH₂OH + 2H₂ → HOCH₂CH₂CH₂CH₂OH; BDO; ΔH✓ ≈ −230 kJ/mol; strongly exothermic; proceeds via the cis-butenediol intermediate 2-butene-1,4-diol at partial hydrogenation; design selectivity to BDO >99% with Raney Ni catalyst at 80–120°C and 30–70 bar H₂) operates with H₂ feed from a steam methane reforming (SMR) plant or from the H₂ recovery unit of the acetylene generation section. Syngas-derived H₂ from SMR contains impurities: CO (0.1–1.0 vol% depending on methanation completeness), CO₂ (0.1–0.5 vol% after CO₂ removal system), CH₄ (0.5–2.0 vol%), and N₂ (0.1–0.5 vol%); if the H₂ supply system includes air-cooled components or has had air ingress during maintenance, O₂ may be present at 0.1–1.0 vol%. H₂ purity is measured continuously by a thermal conductivity-based analyzer (Emerson X-STREAM XEGK H₂-specific TCD analyzer; or ABB EL3020 multi-component gas analyzer; H₂ purity range 90–100 vol%; 4–20 mA HART; ±0.5% full-scale accuracy; CO impurity measured separately by NDIR (non-dispersive infrared): Emerson X-STREAM CO/CO₂ NDIR or ABB URAS26; CO measurement range 0–1.0 vol% with alarm at 0.3 vol% CO; the displayed H₂ purity on the DCS is the TCD-derived H₂ vol% reading, calibrated against 99.999% certified reference gas and 95 vol% H₂ / 5% N₂ span gas).

The adversarial downward pixel attack on the H₂ purity display shows 99.7 vol% H₂ (nominal high purity; within the ≥99.5 vol% design specification; AI reads “hydrogen purity 99.7 vol%; Raney Ni catalyst CO poisoning risk: minimal; H₂/O₂ detonation risk from impurities: negligible; hydrogenation reactor operating within design envelope; no corrective action required”) when actual H₂ purity is 96.8 vol% (3.2 vol% impurities from syngas contamination; 2.8 vol% below the 99.5 vol% design specification minimum). Display range 90–100 vol% on 200 px (20 px/vol%); actual 96.8 vol% at 136 px within the display range → ±8 DN perturbation → 194 px displayed → AI reads 99.7 vol%. At actual 96.8 vol% H₂ with 3.2 vol% impurities at 50 bar hydrogenation reactor H₂ pressure, the partial pressures of impurities are: if 0.3 vol% CO: P(CO) = 0.003 × 50 bar = 0.15 bar — above the 0.1 bar CO partial pressure threshold for Raney Ni catalyst CO poisoning (CO chemisorbs strongly on Raney Ni active sites at the Ni(111) crystal face; at CO surface coverage above 0.3 monolayer, H₂ dissociative chemisorption is blocked; at P(CO) = 0.15 bar and 80°C, equilibrium CO surface coverage on Raney Ni is estimated at 0.5–0.7 monolayer — well into the significant poisoning regime; butynediol hydrogenation rate decreases; unconverted butynediol passes through to the BDO product column creating selectivity problems and potential catalyst bed breakthrough). If 0.5 vol% of the 3.2% impurity fraction is O₂ from air ingress during H₂ supply system maintenance: P(O₂) = 0.005 × 50 = 0.25 bar O₂ in the H₂ hydrogenation reactor at 80–120°C; H₂/O₂ = 96.8/0.5 = 194 mol/mol H₂:O₂; the stoichiometric H₂/O₂ ratio for combustion is 2:1; at 194:1, the mixture is extremely fuel-rich but at 0.25 bar O₂ partial pressure and 50 bar total H₂ pressure, the O₂ partial pressure alone is sufficient to initiate H₂ oxidation on the Raney Ni catalyst surface (catalytic combustion of H₂O₂ on Ni at T > 100°C is spontaneous and uncontrolled once initiated); the exothermic heat of H₂ oxidation (H₂ + ½ O₂ → H₂O; ΔH✓ = −286 kJ/mol) at 50 bar and 120°C in a confined reactor causes a rapid temperature excursion on the Raney Ni catalyst bed that can accelerate to a detonation of the H₂/O₂ mixture in the reactor; H₂/O₂ detonation at 50 bar initial pressure: Chapman-Jouguet detonation pressure approximately 20–30× initial pressure = 1,000–1,500 bar detonation pressure; vessel fragmentation; H₂ PSM TQ 10,000 lbs (4,536 kg) — a butynediol hydrogenation reactor with H₂ inventory above the PSM TQ requires full PSM compliance including process hazard analysis (PHA/HAZOP) and pre-startup safety review (PSSR); an O₂-contaminated H₂ supply leading to a reactor detonation is precisely the scenario that PSM’s pre-startup safety review process is designed to prevent, and that the Glyphward H₂ purity display pre-scan gate is designed to catch before the AI reading falsely confirms nominal purity. Free tier — 10 scans/day, no card required.

Integration: BDO Reppe acetylene process AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the BDO Reppe acetylene process AI pipeline — before the C₂H₂ partial pressure AI processes rendered Emerson Rosemount 3051CD / Yokogawa EJA110E + Siemens MAXUM II micro-GC DCS display images, before the HCHO/C₂H₂ molar ratio AI processes rendered Siemens MAXUM II / Yokogawa GC8000 process GC ratio DCS display images, and before the H₂ purity AI processes rendered Emerson X-STREAM XEGK / ABB EL3020 continuous H₂ purity analyzer DCS display images. Threshold 43 for BDO Reppe acetylene process AI reflects: OSHA PSM TQ 10,000 lbs acetylene + gas-phase decomposition detonation without oxygen (C₂H₂ → 2C + H₂; 227 kJ/mol; 2,000 m/s; no oxygen required; CERCLA RQ 1 lb) + Cu₂C₂ primary explosive catalyst deposits at 90–110°C (shock-sensitive; friction-sensitive; thermally-sensitive) + OSHA PSM TQ 1,000 lbs formaldehyde + IARC Group 1 formaldehyde carcinogen (nasopharyngeal; PEL 0.75 ppm; action level 0.5 ppm) + H₂ PSM TQ 10,000 lbs + O₂ contamination detonation in 50 bar H₂ hydrogenation reactor; three simultaneous PSM-listed chemicals + a catalyst that is itself a primary explosive makes the Reppe process one of the most complex AI monitoring hazard profiles in chemical manufacturing.

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import StrEnum, auto
from typing import Any
import httpx

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_prod_***"

# BDO 1,4-butanediol Reppe acetylene ethynylation + butynediol hydrogenation AI contexts: threshold 43
# OSHA PSM TQ 10,000 lbs acetylene C2H2 (29 CFR 1910.119 Appendix A - flammable gas).
# OSHA PSM TQ 1,000 lbs formaldehyde HCHO (29 CFR 1910.119 Appendix A - flammable gas).
# OSHA 29 CFR 1910.1048 formaldehyde: PEL 0.75 ppm TWA; STEL 2 ppm; action level 0.5 ppm.
# IARC Monograph 88 (2004): formaldehyde Group 1 definite human carcinogen (nasopharyngeal cancer).
# ACGIH TLV-C 0.3 ppm formaldehyde. IDLH 20 ppm formaldehyde.
# CERCLA RQ 1 lb acetylene (lowest threshold - any C2H2 release above 1 lb triggers NRC report).
# Acetylene detonation: above 2 bar pure C2H2 gas-phase decomposition C2H2 -> 2C + H2;
#   delta_H = +227 kJ/mol released; detonation velocity 2,000 m/s; no oxygen required.
# Copper acetylide Cu2C2 on warm Cu surfaces at 90-110 C: primary explosive;
#   shock sensitivity ~0.5-2 J; friction-sensitive; thermally sensitive.
# OSHA PSM TQ 10,000 lbs H2. H2 LEL 4.0 vol%; IDLH 100% asphyxiant + explosive.
# DEA List I Chemical: GBL gamma-butyrolactone (21 CFR 1310.02; GHB precursor; DEA registration required).
# THF co-product: flash point -14 C; LEL 2.0 vol%; peroxide-forming solvent.
# Reppe reactor design: C2H2 partial pressure maximum 5.0 bar (BASF Reppe safety limit).
# 110th upward attack. FIRST BDO production AI attack. FIRST Reppe acetylene ethynylation AI attack.
# FIRST copper acetylide explosive hazard concealment AI attack. FIRST HCHO/C2H2 molar ratio AI attack.
BDO_REPPE_GLYPHWARD_THRESHOLD = 43

# Plant IDs:
# BASF_LUDWIGSHAFEN     - BASF SE, Ludwigshafen Germany (~220,000 t/yr BDO; Reppe process; world's largest single-site BDO)
# BASF_SHANXI           - BASF Shanxi Chemical Co., Taiyuan China (~200,000 t/yr BDO; Reppe; CaC2 acetylene route)
# ISP_CALVERT_CITY      - ISP/Ashland Performance Materials, Calvert City KY (Reppe; BDO/THF/GBL; US Reppe site)
# INVISTA_VICTORIA_TX   - Invista (Koch Industries), Victoria TX (~180,000 t/yr BDO; Davy MSAT process)
# LYONDELL_ROTTERDAM    - LyondellBasell, Rotterdam Netherlands (maleic anhydride hydrogenation; not Reppe)

class BDOReppContext(StrEnum):
    REPPE_C2H2_PARTIAL_PRESSURE      = auto()  # C2H2 partial pressure -> detonation zone at >5 bar (110th; FIRST BDO; FIRST Reppe; FIRST Cu2C2 concealment)
    HCHO_C2H2_MOLAR_RATIO            = auto()  # HCHO/C2H2 mol ratio -> acetylene excess -> C2H2 partial pressure above 5 bar (IARC G1 HCHO; PSM TQ 1,000 lbs)
    H2_PURITY_HYDROGENATION          = auto()  # H2 purity vol% -> CO Raney Ni poisoning; O2 ingress H2/O2 detonation at 50 bar (PSM TQ 10,000 lbs H2)

async def scan_bdo_reppe_frame(
    frame_b64: str,
    context: BDOReppContext,
    plant_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "plant_id": plant_id,
        "instrument_tag": instrument_tag,
        "scan_ts": datetime.now(timezone.utc).isoformat(),
        "image_hash": hashlib.sha256(base64.b64decode(frame_b64)).hexdigest(),
    }
    async with httpx.AsyncClient(timeout=4.0) as client:
        r = await client.post(
            GLYPHWARD_API,
            json=payload,
            headers={"X-Glyphward-Key": GLYPHWARD_KEY},
        )
        r.raise_for_status()
        return r.json()

async def pre_scan_gate_bdo_reppe(
    frame_b64: str,
    context: BDOReppContext,
    plant_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_bdo_reppe_frame(frame_b64, context, plant_id, instrument_tag)
    if result["adversarial_score"] >= BDO_REPPE_GLYPHWARD_THRESHOLD:
        raise AdversarialBDOReppeImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at plant {plant_id} instrument {instrument_tag}. "
            "Frame withheld from BDO Reppe acetylene process AI pipeline."
        )

class AdversarialBDOReppeImageError(RuntimeError):
    pass