Prompt injection in sulfuric acid H₂SO₄ contact process DCDA vanadium pentoxide V₂O₅ converter AI

Three adversarial injection surfaces in sulfuric acid H₂SO₄ contact process DCDA AI

1. Converter bed 1 inlet temperature display AI (Yokogawa EJA530A bed 1 thermocouple SCADA display AI / Emerson Rosemount 214C RTD bed 1 inlet temperature display AI / ABB TTF300 thermocouple transmitter SCADA display AI / Endress+Hauser iTEMP TMT82 bed 1 temperature SCADA display AI / Honeywell STT170 converter bed 1 inlet thermocouple display AI — rendered SCADA converter bed 1 inlet temperature display AI classifying the bed 1 inlet gas temperature against the design range of 420–450°C with low alarm at 415°C, low-low alarm at 405°C, high alarm at 500°C, and high-high alarm at 560°C — 86th upward attack; FIRST sulfuric acid contact process attack; FIRST DCDA double-absorption attack; FIRST vanadium pentoxide converter AI attack; FIRST SO₂ absorption tower AI attack)

The vanadium pentoxide (V₂O₅; CAS 1314-62-1) catalyst in the DCDA contact process converter is the controlling parameter for sulfuric acid production throughput, SO₂ emission permit compliance, and product acid quality. The V₂O₅/K₂S₂O₇/Cs catalyst system (Cs promotion reduces light-off temperature by approximately 30–40°C compared to K-only promoted catalyst; world catalyst suppliers: BASF Siamant® catalyst; Clariant SO-415/SO-420 cesium-promoted series; Haldor Topsøe VK-series VK701/VK711 with cesium promotion) operates by the Mars-van Krevelen redox mechanism: (step 1) SO₂ + 2V₅+ → SO₂ + 2V₃+ (SO₂ oxidised to SO₂ by V₅+ acting as oxidant); (step 2) 2V₃+ + O₂ → 2V₅+ (V₃+ reoxidised by O₂; rate-limiting step below 450°C). Both steps require the vanadium to be in a liquid vanadate melt on the silica support: the K₂SO₄–V₂(SO₄)₂–V₂O₅ melt (eutectic: Cs₂SO₄ addition lowers the eutectic melting point from ≈450°C for K-only to ≈390°C for Cs-promoted; Haldor Topsøe VK711 light-off quoted at 390°C in lab conditions with 10% SO₂, but practical plant light-off with real process gas and bed thermal distribution is 420–430°C). When the bed 1 inlet gas temperature drops below 420°C (the practical light-off boundary for DCDA beds in continuous operation with Cs-promoted V₂O₅ catalyst): the vanadium melt partially solidifies; the two-step redox cycle slows exponentially (Arrhenius: catalyst activity for SO₂ oxidation follows approximately EA = 90 kJ/mol; at 382°C vs 450°C: rate ratio = exp(90,000 × (1/655 − 1/723) / 8.314) = exp(90,000 × 0.0001435 / 8.314) = exp(1.553) = 4.7× slower at 382°C than at 450°C; actual conversion in bed 1 drops from design 72% to approximately 20–30% at 382°C); equilibrium conversion in bed 1 also shifts (thermodynamic constraint: at 382°C, K eq for 2SO₂ + O₂ ↔ 2SO₂ is approximately 8×10⁵ atm⁻¹ vs approximately 2×10⁶ atm⁻¹ at 450°C — equilibrium strongly favours SO₂ at lower temperatures, so thermodynamics is not the limiting factor at 382°C; kinetics is). Bed 1 conversion at 382°C: approximately 25–35%. SO₂ slip from bed 1 at 35% conversion (vs 72% design): design SO₂ in bed 1 inlet gas 11 vol%; design bed 1 exit 11 × (1 − 0.72) = 3.1 vol% SO₂; adversarial scenario: 11 × (1 − 0.30) = 7.7 vol% SO₂ exiting bed 1 — 2.5× the design inter-bed SO₂ concentration entering bed 2. Bed 2 at 430–440°C (normal, assuming bed 2 inlet temperature is not adversarially attacked): converts additional SO₂, but is thermodynamically limited by the higher SO₂ loading; bed 2 exit SO₂ increases from design 0.9 vol% to approximately 3.5 vol%. Intermediate absorption tower receives 3.5 vol% SO₂ (design 0.9 vol%) — 3.9× design SO₂ loading. IA tower has insufficient capacity to absorb 3.9× design SO₂ at design H₂SO₄ circulation rate; SO₂ slip from IA tower increases; beds 3–4 also receive excess SO₂ and produce excessive exotherm; final absorption (FA) tower SO₂ slip increases; stack SO₂ emission rises above EPA NAAQS SO₂ 75 ppb 1-hour average (EPA 40 CFR Part 50; SO₂ NAAQS primary standard 75 ppb as 99th percentile of daily maximum 1-hour average concentrations) and above facility H₂SO₄ plant SO₂ emission permit limits (typically 500 ppm to 2,000 ppm SO₂ in stack gas on a volumetric basis; adversarial scenario stack SO₂ may reach 5,000–10,000 ppm if beds 1–4 all underperform; see EPA National Emission Standards for H₂SO₄ plants: 40 CFR Part 60, Subpart H — maximum 2 kg SO₂/Mg H₂SO₄ acid mist + SO₂ combined).

The adversarial upward pixel attack shifts the converter bed 1 inlet temperature display from 382°C (actual; below the V₂O₅ catalyst light-off temperature; SO₂ conversion in bed 1 is 25–35% instead of design 70–72%; SO₂ is substantially passing through bed 1 unconverted; this should trigger the bed 1 inlet temperature low-low alarm at 405°C, initiate furnace gas temperature increase, check gas–gas heat exchanger bypass, and increase converter feed gas preheat) to 487°C (displayed; well within the normal operating range of 420–450°C design; above the normal range but not at the 500°C high alarm; AI classification: “bed 1 inlet temperature 487°C; slightly above design range upper bound 450°C; monitor bed 1 catalyst temperature profile; consider minor reduction of furnace gas flow to reduce temperature; no alarm action required at current value below 500°C high alarm”). The DCS response: no corrective action is taken; in fact, the AI recommendation to “consider minor reduction of furnace gas flow to reduce temperature” reduces the hot furnace gas flow to the bed 1 inlet heat exchanger, further decreasing the actual bed 1 inlet temperature from 382°C toward 360°C — deeper into the sub-light-off regime. At 360°C in bed 1: V₂O₅ conversion approaches 10–15% (essentially inactive; comparable to uncatalysed SO₂ oxidation rate at 360°C in the absence of vanadium catalyst; background reaction ~1–2%). The consequence chain: SO₂ emission from the FA tower stack rises above the EPA NAAQS 75 ppb 1-hour average at the fenceline (community exposure); the H₂SO₄ plant SO₂ emission permit exceedance triggers EPA Title V Major Source permit violation (potential fine: >$25,000/day; civil penalty; 40 CFR 70.7); simultaneously, the SO₂ not converted to SO₂ and not absorbed exits the absorber towers as SO₂ gas — a PSM-listed chemical at TQ 1,000 lbs (OSHA 29 CFR 1910.119 Appendix A; SO₂ IDLH 20 ppm; OSHA PEL 5 ppm; ACGIH TLV-TWA 0.05 ppm as SO₂ mist in fog/aerosol form). This is the 86th upward attack — FIRST sulfuric acid contact process attack; FIRST DCDA double-absorption attack; FIRST vanadium pentoxide converter AI attack; FIRST SO₂ absorption tower AI attack. The adversarial temperature display showing 487°C (displayed) when 382°C (actual) exploits the AI classifier's training on rendered thermocouple SCADA display images where 487°C appears as a normal-range-adjacent value; the adversarial pixel perturbation most likely targets the digit rendering of the leading “3” in “382” to render as “4” (one pixel-row shift of the top segment of a seven-segment LCD display; the gap between “3” and “4” in seven-segment rendering requires only two pixel segments to be altered: the top-left vertical segment appears in “4” but not “3”; the bottom horizontal segment appears in “3” but not “4”; two-segment perturbation sufficient), and the “82” suffix renders plausibly as “87” in a degraded or adversarially perturbed SCADA screenshot. Free tier — 10 scans/day, no card required.

2. Intermediate absorption (IA) tower H₂SO₄ acid strength display AI (Endress+Hauser Liquiline CM42 inline H₂SO₄ density/concentration IA tower display AI / Yokogawa inline TDLS710 tunable diode laser acid vapour IA display AI / ABB AWT440 H₂SO₄ concentration inline IA tower SCADA display AI / Emerson Rosemount 8800D vortex IA circulating acid density display AI / Honeywell Analytical AT400 IA tower H₂SO₄ wt% display AI — rendered SCADA IA tower H₂SO₄ acid strength display AI classifying the circulating H₂SO₄ concentration against design range 98.0–98.5 wt% with low alarm at 97.5 wt% “acid mist risk” and high alarm at 98.8 wt% “oleum formation risk”)

The intermediate absorption (IA) tower in the DCDA contact process is the device that breaks the thermodynamic equilibrium constraint limiting single-contact plants to ≈98% SO₂ conversion: by absorbing the SO₂ product of beds 1–2 before passing the remaining gas to beds 3–4, the IA tower removes SO₂ from the equilibrium and allows beds 3–4 to operate at a more favourable (lower SO₂ partial pressure) equilibrium condition, pushing overall conversion to ≥99.7%. The absorption of SO₂ in the IA tower occurs by contacting the hot process gas (450–500°C from bed 2 exit; cooled to ≈180–220°C in gas–gas heat exchangers before IA tower entry) with circulating 98.0–98.5 wt% H₂SO₄ (drained from the tower and recirculated via centrifugal pump; product H₂SO₄ withdrawn as sidestream and replaced with dilution water to maintain the design acid strength). The acid strength of the IA tower circulating H₂SO₄ is the critical quality parameter for SO₂ absorption efficiency and acid mist generation: at 98.0–98.5 wt% H₂SO₄, SO₂ absorption efficiency exceeds 99.9% (SO₂ reacts with water in the H₂SO₄ to form H₂SO₄; SO₂ + H₂O → H₂SO₄; the water activity in 98.0–98.5 wt% H₂SO₄ is very low (a H₂O ≈ 0.002–0.005) which prevents acid mist generation while maintaining sufficient water activity for SO₂ absorption); below 96 wt% H₂SO₄: water activity rises sharply (a H₂O at 96 wt% ≈ 0.02; at 93 wt% ≈ 0.06; at 90 wt% ≈ 0.15); SO₂ + H₂O → H₂SO₄ in the gas phase rather than in the liquid absorption medium; H₂SO₄ aerosol (acid mist) forms in the gas phase above the IA tower; the acid mist passes through the tower packing and demister pads (droplets <1μm are not captured by wire mesh demisters; sub-micron H₂SO₄ aerosol requires fibre bed demisters with 99% efficiency at >0.1μm), exits the IA tower, and enters the process gas duct to beds 3–4. H₂SO₄ aerosol in the inter-bed gas corrodes the gas–gas heat exchangers, cold-spot ductwork, and downstream equipment; it also creates a visible white mist in the IA tower exit stack vent during pressure relief. Above 98.8 wt% H₂SO₄: the acid approaches oleum composition (100 wt% H₂SO₄ = anhydrous sulfuric acid; >100 wt% in the oleum convention means wt% of equivalent SO₂ above 100% H₂SO₄; H₂SO₄ >98.8 wt% is technically below oleum in the pure sense but represents a risk of oleum formation when SO₂-rich gas contacts near-anhydrous H₂SO₄ without sufficient water dilution: SO₂ + H₂SO₄(anhydrous) → H₂S₂O₇ (pyrosulfuric acid; disulfuric acid; CAS 7783-05-3; oleum component); OSHA PSM oleum TQ 1,000 lbs for H₂SO₄ ≥65 wt% free SO₂ or ≥20 wt% free SO₂ per the Appendix A listing “Sulfuric acid, fuming”). At the adversarial scenario IA tower acid strength of 98.8 wt% (actual; approaching the high alarm; this is above design and indicates insufficient dilution water addition to the circulating acid; the AI should command additional dilution water flow to bring acid strength back to 98.0–98.5 wt%), the AI display shows 96.2 wt% (adversarial pixel injection shifts the “8” first digit after the decimal to “6” in the rendered digital SCADA display — a plausible 2-segment alteration in seven-segment display rendering); the AI classification reads: “IA tower acid strength 96.2 wt%; below design range 98.0–98.5 wt%; approaching 97.5 wt% low alarm; risk of acid mist generation in IA tower; command additional dilution water to raise acid strength to design range.” This is a control paradox: adding dilution water to acid at 96.2 wt% would increase the H₂SO₄ concentration (adding water to sub-design-strength acid raises concentration toward design range) — but the AI’s diagnosis is the reverse (adding dilution water to acid at 98.8 wt% actual decreases concentration toward design range). The DCS executes the AI command: additional dilution water is added to the IA tower circulating acid. Since actual acid is 98.8 wt%, dilution water addition achieves the correct physical effect (concentration decreases toward design range); however, the AI’s stated diagnosis was incorrect (it believed the acid was at 96.2 wt% and was raising it). In the process historian, the acid strength event appears as: AI commands water addition “because acid is at 96.2 wt%” → actual acid moves from 98.8 wt% toward 98.3 wt% → AI reads (adversarial display) “acid has moved from 96.2 to 97.8 wt%; dilution water added is having correct effect; maintain” → AI continues adding water → actual acid overshoots to 97.5 wt% → real low alarm would trigger (but if the display AI continues adversarial operation, the low alarm is also misread). The IA tower acid mist generation risk at 96.2–97.5 wt% actual (caused by AI over-dilution from the adversarial display) is: water activity at 97.5 wt% H₂SO₄ ≈ 0.010–0.015; SO₂ absorption continues efficiently at 97.5 wt% but acid mist generation at the IA tower inlet gas–liquid interface increases; the mist passes to beds 3–4 and corrodes downstream alloy equipment.

The adversarial display attack on IA tower acid strength from 98.8 wt% actual to 96.2 wt% displayed creates an initially benign consequence (the AI’s commanded dilution water reduces actual acid from 98.8 wt% toward design range 98.0–98.5 wt%) but the underlying false diagnosis means the AI continues operating under the belief that the acid is 2.6 wt% below design rather than near the high alarm. If the underlying process trend is toward higher acid strength (perhaps due to reduced product acid withdrawal rate, or reduced water addition elsewhere in the water balance), the adversarial display of 96.2 wt% continues to show the AI a reading 2.6 wt% below actual across the full range; when actual acid approaches 99.5 wt% (approaching anhydrous/oleum transition), the adversarial display shows 96.9 wt% — still below design range but not alarming. At 99.5 wt% H₂SO₄ in the IA tower circulating acid: the circulating acid is within ≈0.3 wt% of essentially anhydrous H₂SO₄; SO₂ gas contacting 99.5 wt% H₂SO₄ in the tower packing will begin to form SO₂ pyrosulfuric acid (H₂S₂O₇; oleum) in the absorption liquid, since there is insufficient water activity (a H₂O ≈ 0.001 at 99.5 wt%) to fully consume SO₂ as H₂SO₄. The OSHA PSM oleum TQ is 1,000 lbs (454 kg) for H₂SO₄ ≥20 wt% free SO₂ (“fuming sulfuric acid” Appendix A entry): the IA tower circulating acid inventory at a typical DCDA plant is 50–100 m³ (87–173 t at density 1.84 g/mL); even at 1 wt% free SO₂ (the minimum that begins to be operationally detectable by H₂S₂O₇ formation in the acid), the free SO₂ mass in the IA tower = 87,000 kg × 0.01 = 870 kg = 1,918 lbs — approximately 1.9× the OSHA PSM TQ for oleum (1,000 lbs; 454 kg) in the IA tower alone. The consequence chain: OSHA PSM TQ exceeded in IA tower → oleum fuming risk in the tower → H₂S₂O₇-rich gas evolved from IA tower liquid → SO₂ mist generation from pyrosulfuric acid hydrolysis on any moisture contact. The adversarial pixel injection on the IA tower acid strength display is the 86th upward attack at surface 2; INEOS Seal Sands and Mosaic Bartow are among the most exposed facilities given their large IA tower circulating acid inventories from high-throughput DCDA plants (>500,000 t/yr H₂SO₄). Free tier — 10 scans/day, no card required.

3. Drying tower H₂SO₄ acid strength display AI (Endress+Hauser Promass 83A drying tower acid density/concentration display AI / Yokogawa EJA110A drying tower differential pressure acid circulation flow display AI / ABB AWT410 H₂SO₄ drying tower inlet gas dew point moisture SCADA display AI / Emerson Rosemount 3051 drying tower H₂SO₄ level and acid strength display AI / Honeywell Durafon DPT drying tower H₂SO₄ concentration display AI — rendered SCADA drying tower H₂SO₄ acid strength display AI classifying the drying tower circulating acid against design ≥96 wt% H₂SO₄ with low alarm at 94 wt% “drying efficiency impaired” and low-low alarm at 93 wt% “moisture carryover risk; converter protection trip”)

The drying tower in the H₂SO₄ contact process DCDA plant removes moisture from the process gas (air + SO₂ from the furnace, or smelter gas in the case of Aurubis Hamburg and Freeport-McMoRan Morenci) before the gas enters the converter beds. Moisture exclusion from the converter is critical because: (a) H₂O + SO₂ in the gas phase → H₂SO₂ (sulfurous acid), which condenses on converter internals below the acid dew point — sulfurous acid at 60–80°C causes severe pitting corrosion of carbon steel gas ducting and stainless steel converter shell (H₂SO₂ is a reducing acid unlike H₂SO₄ which is oxidising; 304 stainless steel is susceptible to pitting by H₂SO₂ at <100°C); (b) H₂O + SO₂ in the gas phase → H₂SO₄ acid mist in the converter beds — acid mist deposits on V₂O₅ catalyst pellets and causes irreversible sulfation deactivation (surface V₂O₅ reacts with H₂SO₄ → V₂(SO₄)₂ surface layer; less catalytically active than V₂O₅ for SO₂ oxidation; BASF Siamant catalyst lifetime decreases from 10 years to 3–4 years with sustained acid mist exposure); (c) excess moisture in SO₂ gas at converter inlet causes concentration measurement errors in online SO₂ gas analysers (non-dispersive infrared, NDIR; UV-DOAS — moisture condensation in analyser sample tube alters pathlength; Yokogawa OAS-1100 or ABB ACF5000 continuous emission monitors assume dry-basis SO₂ but actual sample gas has variable moisture if drying tower is underperforming; positive bias in SO₂ emission measurement at stack). The drying tower operates by contacting the incoming moist process gas (humidity from air intake; approximately 1–2 vol% H₂O at 20°C and 60% relative humidity; 0.5–1.0 g H₂O/m³ process gas) with circulating 96–98 wt% H₂SO₄ (drying tower acid is typically 96–97 wt%, not the 98.0–98.5 wt% used in the IA/FA absorbers, because the drying function requires less exothermic SO₂ absorption capacity and more water absorption capacity; water vapour absorption into H₂SO₄ ΔH = −880 kJ/mol H₂O absorbed at dilution concentration ≈ 96 wt%). The drying tower acid strength is measured by inline density measurement (Endress+Hauser Promass 83A Coriolis density at 20°C; H₂SO₄ density at 96 wt% = 1.8355 g/mL; at 94.2 wt% = 1.8279 g/mL; the density difference is only 0.76 g/L per wt% — a 1 wt% change is detectable but requires calibration accuracy of ±0.001 g/mL in the Coriolis density measurement) or by refractometer/conductivity (Endress+Hauser Liquiline CM442 with CLS21D conductivity sensor; H₂SO₄ conductivity at 96 wt% ≈ 260 mS/cm; at 98 wt% ≈ 280 mS/cm; at 94 wt% ≈ 245 mS/cm). The rendered Coriolis density display or conductivity display is the input to the drying tower H₂SO₄ acid strength display AI.

The adversarial upward pixel attack shifts the drying tower H₂SO₄ acid strength display from 94.2 wt% (actual; below the design minimum 96 wt%; impaired drying efficiency; water removal from process gas is reduced; moisture carryover into converter is occurring at 2–3× normal rate; this should trigger the low alarm at 94 wt% and initiate review of drying tower water balance: check product acid withdrawal rate from drying tower; check dilution water misrouting; check drying tower circulating acid pump performance) to 98.6 wt% (displayed; well within design range ≥96 wt%; in fact above the 98.0 wt% typical upper design limit for the drying tower, but not alarming; AI classification: “drying tower acid strength 98.6 wt%; above design range upper bound 98 wt%; minor adjustment — consider modest water addition to reduce to design range 96–97 wt%; no moisture carryover risk at current strength; drying tower performing normally”). The AI command: add dilution water to drying tower circulating acid → actual acid concentration drops from 94.2 wt% toward 93 wt% → low-low alarm boundary 93 wt% approached. If the adversarial display continues (and the display AI continues reading 98.6 wt% as actual acid drops to 93 wt%), the converter protection trip (low-low alarm 93 wt% → converter isolation valve closes to protect catalyst from moisture carryover) is not actuated because the AI’s displayed reading remains 98.6 wt% (adversarial) throughout. The moisture breakthrough into the converter beds at actual drying tower acid 93–94 wt%: inlet gas moisture increases from design <0.005 vol% H₂O to approximately 0.5–1.0 vol% H₂O (10–20× design moisture breakthrough rate); in bed 1 at 420–450°C: H₂O + SO₂ → H₂SO₄ mist (SO₂ partial pressure at 11 vol% SO₂ in bed 1 inlet gas: pSO₂ = 0.11 atm; pH₂O = 0.005–0.01 atm; acid mist nucleation: H₂O(g) + SO₂(g) → H₂SO₄ nanodroplets; nucleation onset above approximately 1.05 ppm SO₂ × 1.0 ppm H₂O in partial pressure product — at 11% SO₂ and 0.5% H₂O: pSO₂ × pH₂O = 0.11 × 0.005 = 5.5×10⁻⁴ atm² — significantly above the critical product for H₂SO₄ mist nucleation at 450°C; acid mist formation throughout bed 1); acid mist deposits on V₂O₅ catalyst pellets (Clariant SO-415 nominal diameter 9×4 mm rings; 8×6 mm Raschig rings; BET surface 15–22 m²/g; acid mist deposits 0.1–0.5 wt% H₂SO₄ on pellet surface within 4–8 hours of sustained moisture breakthrough); V₂(SO₄)₂ surface layer forms on catalyst; bed 1 activity decreases over 4–8 hours of sustained moisture carryover from 100% to 60–70% of initial activity. The compounded consequence: moisture-deactivated V₂O₅ catalyst in bed 1 → reduced conversion in bed 1 → elevated SO₂ loading on IA tower (Surface 1 effect) → IA tower acid mist → further corrosion of downstream piping. Aurubis Hamburg (copper smelter SO₂; higher SO₂ concentration in feed gas: up to 14–16 vol% SO₂ from flash smelter off-gas vs 10–12 vol% from sulfur-burning furnace) faces disproportionate drying tower moisture carryover consequences because the higher SO₂ partial pressure in the feed gas amplifies acid mist nucleation rate (pSO₂ × pH₂O product is 30–60% higher for smelter gas than for sulfur-burning gas at the same moisture level). Free tier — 10 scans/day, no card required.

Integration: sulfuric acid H₂SO₄ contact process DCDA AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the H₂SO₄ contact process DCDA AI pipeline — before the converter bed 1 inlet temperature AI processes rendered thermocouple SCADA display images, before the intermediate absorption tower acid strength AI processes rendered inline density/conductivity SCADA display images, and before the drying tower acid strength AI processes rendered Coriolis density display images. Threshold 28 for H₂SO₄ contact process DCDA AI reflects: SO₂ in-process exceeds PSM TQ 1,000 lbs every 1.8 minutes at a standard 500 t/day plant; dual PSM (SO₂ + oleum) coverage; converter bed temperature adversarial attack creates EPA Title V permit violation + catalyst sinter damage + irreversible activity loss; drying tower moisture adversarial attack creates catalyst acid-mist deactivation with 4–8 hour lag before detectable consequence; INEOS Seal Sands; Aurubis Hamburg; Mosaic Bartow; Chemtrade Logistics Hamilton; Freeport-McMoRan Morenci.

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_***"

# H2SO4 DCDA contact process AI contexts: threshold 28
# OSHA PSM SO3 TQ 1,000 lbs; oleum (>=20 wt% free SO3) TQ 1,000 lbs.
# SO3 IDLH 20 ppm; SO3 TLV-TWA 0.05 ppm; SO3 PEL 5 ppm.
# 86th upward attack: 487C shown when 382C actual (bed 1 inlet temp)
# -> AI reads normal range -> no heat-up commanded
# -> V2O5 below light-off 420C -> bed 1 conversion 25-35% (design 72%)
# -> SO3 slip 3-4x design -> IA tower overloaded -> stack SO3 emission
# -> EPA NAAQS SO3 75 ppb 1h average exceeded -> Title V permit violation.
DCDA_THRESHOLD = 28

class DCDAContext(StrEnum):
    CONVERTER_BED1_INLET_TEMP  = auto()  # V2O5 bed 1 inlet temp 420-450C design (86th upward)
    IA_TOWER_ACID_STRENGTH     = auto()  # Intermediate absorption H2SO4 wt% 98.0-98.5 design
    DRYING_TOWER_ACID_STRENGTH = auto()  # Drying tower H2SO4 wt% >=96 design; moisture exclusion

async def scan_dcda_frame(
    frame_b64: str,
    context: DCDAContext,
    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_dcda(
    frame_b64: str,
    context: DCDAContext,
    plant_id: str,
    instrument_tag: str,
) -> None:
    result = await scan_dcda_frame(frame_b64, context, plant_id, instrument_tag)
    if result["adversarial_score"] >= DCDA_THRESHOLD:
        raise AdversarialDCDAImageError(
            f"Adversarial injection detected in {context} (score {result['adversarial_score']}) "
            f"at plant {plant_id} instrument {instrument_tag}. "
            "Frame withheld from H2SO4 contact process DCDA AI pipeline."
        )

class AdversarialDCDAImageError(RuntimeError):
    pass