xpertsystems/oil043-sample

OIL-043 — Synthetic Scenario Simulation Dataset (Sample)

A schema-identical preview of OIL-043, the XpertSystems.ai synthetic what-if scenario simulation dataset for oil & gas decision-support AI, business-continuity modeling, enterprise risk management (ERM), and executive-tier decision-support training. The full product covers 12,000 facilities × 250,000 scenarios across a 5-year horizon. This sample is the generator's sample mode (750 facilities × 8,000 scenarios) covering all 12 product tables.

Built by XpertSystems.ai — Synthetic Data Platform Contact pradeep@xpertsystems.ai · xpertsystems.ai License CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

What OIL-043 does that nothing else in the catalog does

OIL-043 is the catalog's first decision-support / what-if scenario SKU. Where OIL-042 (Digital Twin) models the steady-state operations of an oilfield, OIL-043 models the perturbations to those operations — price shocks, operational disruptions, equipment failure cascades, supply chain interruptions, inventory stress, logistics constraints, cyberattacks, emergency response, market recovery — each linked to a scenario_id with pre-built ML labels (disruption probability, resilience score, financial impact, decision priority).

This is the substrate that ERM, business-continuity, and executive decision-support AI teams have been waiting for: a coherent, joinable dataset where commodity shocks, OT cyber incidents, supply chain delays, and equipment failure cascades can be modeled together with shared severity, region, and decision-priority labels.

Buyer Persona	Use Case
Chief Risk Officer / ERM	Enterprise risk scoring across 9 scenario types
Business Continuity Director	Recovery time estimation, escalation modeling
C-suite Decision Support AI	Executive priority labels (low/medium/high/critical)
CISO / OT Security	ICS attack impact on operations (SCADA availability)
Strategic Planning / S&OP	Multi-scenario portfolio stress testing
Insurance / Reinsurance	Loss-severity distribution modeling for upstream

---

What's inside

12 CSV tables organized around a scenario_id master key: scenario master → price shocks → operational disruptions → equipment failure chains → production impacts → supply chain interruptions → inventory depletion → logistics constraints → cyberattack scenarios → emergency response → market recovery timelines → pre-built ML labels.

Table	Rows (sample)	What it represents
scenario_master.csv	8,000	9-class scenario type × 4-class severity × facility/region/duration
price_shock_events.csv	~12,000	7-commodity panel: WTI, Brent, HenryHubGas, Diesel, Gasoline, LNG_JKM, FuelOil
operational_disruptions.csv	~26,000	6-class disruption × 8-class root cause × throughput loss + downtime
equipment_failure_chains.csv	~19,000	8-class asset × 8-class failure mode × cascade level + spare availability
production_impacts.csv	8,000	Lost volume boe + revenue loss + ramp-down/up hours per scenario
supply_chain_interruptions.csv	~15,000	Route disruption with cost-increase + rerouting + supplier risk
inventory_depletion.csv	8,000	4-class stress level × depletion rate × days-to-stockout
logistics_constraints.csv	~4,500	5-class transport mode × congestion + demurrage cost
cyberattack_scenarios.csv	~1,100	5-class ICS attack × SCADA availability + manual operation flag
emergency_response.csv	~5,600	4-level escalation (site/regional/corporate/regulatory) + IC + exec brief
market_recovery_timelines.csv	8,000	Stabilization + full-recovery days + residual risk + lessons-learned
scenario_labels.csv	8,000	Pre-built ML labels: disruption prob + resilience + financial impact + decision priority

Total: ~123,000 rows, ~12 MB. The full OIL-043 product is ~4 million rows.

---

Calibration sources

Every distribution and ratio is anchored to named public references. Highlights:

• IPIECA Operating Risk Framework + IEA Black-Swan Scenario Library — scenario severity and rare-event distributions.
• IEA / EIA / S&P Platts commodity reference panels — 7-commodity price-shock taxonomy.
• ISO 14224:2016 + API RP 691 — rotating equipment failure-mode taxonomy.
• CCPS Bow-Tie + LOPA cascade analysis — equipment failure cascade depth ranges.
• ICS-CERT + NIST SP 800-82 — ICS/OT incident-impact SCADA-availability degradation bands.
• EIA / API midstream statistics — pipeline transport-mode share.
• IEA Energy Transport Network — 5-class logistics transport-mode taxonomy.
• OECD / IEA Scenario Recovery — disruption-event recovery timelines.
• CCPS Root-Cause Analysis + ASSE/ASSP — lessons-learned and corrective action norms.

---

Validation scorecard

The wrapper ships a 10-metric scorecard (validation_scorecard.json) that re-scores the dataset on every generation. Default seed 42 result:

ID	Metric	Target	Observed	Source
M01	Scenario-Type Taxonomy (floor)	≥ 9	9	IPIECA / IEA
M02	Commodity Panel Coverage (floor)	≥ 7	7	IEA / EIA / Platts
M03	Failure-Mode Taxonomy (floor)	≥ 8	8	ISO 14224 / API RP 691
M04	Critical-Severity Scenario Share	0.04–0.08	0.067	IPIECA Operating Risk
M05	Cascade Level (mean)	1.5–3.5	2.41	CCPS Bow-Tie / LOPA
M06	Cyber-Active SCADA Availability %	55–85	72.9	ICS-CERT / NIST 800-82
M07	Transport-Mode Taxonomy (floor)	≥ 5	5	IEA Energy Transport
M08	Pipeline Transport Share	0.30–0.50	0.38	EIA / API midstream
M09	Full Recovery Days (median)	0–60	21.2	OECD / IEA Scenario
M10	Lessons Learned (mean)	3–7	4.97	CCPS RCA / ASSE

Grade: A+ (100/100). Verified across seeds 42, 7, 123, 2024, 99, 1.

---

Suggested use cases

• Decision-support AI training — scenario_labels.csv provides 4-class decision priority labels (low / medium / high / critical) plus a binary model_label calibrated against disruption probability + financial impact. Train executive priority-classification models with ~27% positive class density.
• Enterprise risk scoring (ERM) — disruption_probability, resilience_score, financial_impact_score, and operational_risk_score are per-scenario continuous-valued ML targets. Train regression models for portfolio-wide risk scoring.
• Multi-modal scenario impact modeling — join across all 11 event tables on scenario_id to train models that predict downstream impact (production loss, recovery time) from upstream signals (price shock, cyber event, equipment failure).
• Cascading failure modeling — equipment_failure_chains.csv has cascade_level (1–6) for upstream → downstream failure propagation. Train graph-neural-network or Bow-Tie analysis models.
• Cyber-physical impact estimation — cyberattack_scenarios.csv × operational_disruptions.csv × production_impacts.csv enable Industroyer / TRITON / Colonial Pipeline-class incident impact modeling.
• Supply chain stress testing — scenario portfolios with linked inventory depletion + logistics constraints + cost increase enable multi-tier supply-chain network resilience modeling.
• Black-swan rare-event modeling — is_rare_event flag identifies critical-severity scenarios with explicit rare-event injection.
• Cross-vertical scenario validation — the 9-class scenario taxonomy applies analogously to other XpertSystems verticals (Insurance, Healthcare, Cybersecurity); buyers can use OIL-043 as the framework for building their own scenario libraries.

---

Loading

from datasets import load_dataset

scenarios = load_dataset(
    "xpertsystems/oil043-sample",
    data_files="scenario_master.csv",
    split="train",
)
labels = load_dataset(
    "xpertsystems/oil043-sample",
    data_files="scenario_labels.csv",
    split="train",
)
disruptions = load_dataset(
    "xpertsystems/oil043-sample",
    data_files="operational_disruptions.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/oil043-sample",
    filename="market_recovery_timelines.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)

All 12 tables share scenario_id as the master join key. Most tables also carry facility_id for cross-cutting joins. Aggregation patterns:

• scenario_master ⨝ scenario_labels (1:1) — every scenario has labels
• scenario_master ⨝ operational_disruptions (1:N) — multiple disruptions per scenario
• scenario_master ⨝ equipment_failure_chains (1:N) — failure cascades
• scenario_master ⨝ cyberattack_scenarios (1:0–1) — cyber-only scenarios
• scenario_master ⨝ market_recovery_timelines (1:1) — every scenario has recovery

---

Schema highlights

scenario_master.csv — scenario_id, facility_id, scenario_type (9-class: price_shock / equipment_failure / operational_disruption / supply_chain_interruption / inventory_stress / cyberattack / weather_disruption / geopolitical_event / regulatory_shutdown), severity_level ∈ {low, medium, high, critical}, region (8-class), facility_type (8-class), start_timestamp, duration_hours, is_rare_event, dependency_count, baseline_capacity_boe_per_day, scenario_complexity_score ∈ [0, 1].

price_shock_events.csv — commodity (7-class IEA/EIA panel), shock_direction ∈ {up, down}, shock_magnitude_pct, volatility_regime ∈ {normal, elevated, stressed, crisis}, spread_impact_bps, mean_reversion_days.

equipment_failure_chains.csv — asset_type (8-class: compressor / pump / valve / pipeline_segment / turbine / heat_exchanger / storage_tank / separator), failure_mode (8-class ISO 14224), cascade_level ∈ {1, …, 6} (CCPS Bow-Tie), mtbf_hours_before_failure, estimated_repair_hours, spare_part_available (links to OIL-041 spare-parts demand), failure_probability.

cyberattack_scenarios.csv — attack_type (5-class: scada_lockout / ransomware / sensor_spoofing / data_exfiltration / network_segmentation_failure), ot_network_impact_score, scada_availability_pct, manual_operation_required, containment_hours, estimated_cyber_loss_usd.

scenario_labels.csv — pre-built ML labels: disruption_probability ∈ [0, 1], resilience_score ∈ [0, 1], financial_impact_score ∈ [0, 1], operational_risk_score ∈ [0, 1], recommended_decision_priority ∈ {low, medium, high, critical}, requires_executive_action (binary), model_label (binary, high+critical = 1).

---

Calibration notes & limitations

In the spirit of honest synthetic data, a few things buyers of the sample should know:

1. Throughput loss median is 33% — well above industry-mature 5–15%. The operational_disruptions.csv table is biased toward stressed-scenario training utility: throughput losses are sampled as 0.08 + sev × 0.55 plus noise. The dataset is designed to give ML models trainable positive-class density for severe scenarios, not to estimate routine operations. For routine-disruption analytics, filter to severity_level == 'low' (33% of records) to recover median throughput loss ~10%.

2. SCADA availability ~73% on cyber-active scenarios. This is the conditional availability during an active cyber incident — not the steady-state SCADA quality (which is ~99.9% in OIL-042's scada_telemetry.csv). The 73% figure is anchored to ICS-CERT incident reports (55–85% degradation band) and is the metric of interest for cyber-impact modeling.

3. Critical severity rate 6.7%, rare event flag 4.8%. The is_rare_event flag is stricter than severity_level == 'critical' — it fires only when severity == 'critical' AND random < 0.72. This models the IPIECA distinction between "high-severity scenario" (any crit) and "tail-risk / black-swan" (truly novel + catastrophic). Use is_rare_event for black-swan modeling, severity_level == 'critical' for general high-severity work.

4. Cyber-attack scenarios are sparse (~1,100 rows). Calibrated to IPIECA's cyber-attack base rate of ~6% of scenarios (with cyberattack_probability config flag). For dense cyber-attack ML training, use the full product (prod mode → ~34,000 cyber scenarios) or oversample with weights from attack_type.

5. Logistics constraints sparse (~4,500 rows). Only fires on supply_chain / weather / geopolitical scenarios + 40% random others. For dense logistics ML, filter to those 3 scenario types directly.

6. Spare-part availability ~72%, not OIL-041's industry-mature 85%+. In OIL-043, spare availability is conditional on stressed scenarios — it degrades as severity increases by design. Use OIL-041 for steady-state spare-parts inventory analytics; use OIL-043 for crisis- scenario spare-parts unavailability modeling.

7. Equipment failure mode taxonomy is 8-class here, vs OIL-038's 10-class generator and OIL-042's 10-class. The 8 modes are a subset (the 2 dropped: wax_deposition, scale_blockage — which are more process-side than mechanical). Cross-SKU joins on failure_mode may need value normalization.

8. Operational disruption types: 6-class. Smaller than the 18-class OIL-038 failure modes — by design (operational disruptions are at the event level, not the mechanical mode level).

9. Deterministic seeding. All 12 tables are deterministic on --seed. Catalog default is seed 42. Seed sweep verifies Grade A+ across {42, 7, 123, 2024, 99, 1}.

---

Commercial / full product

The full OIL-043 product covers 12,000 facilities × ~250,000 scenarios across a 5-year horizon (4 million rows total), with dense coverage of all categorical taxonomies including the rare cyber-attack scenarios (~34,000), heavy-tail black-swan injection at IPIECA-specified rates, and configurable scenario-portfolio composition for industry- specific stress testing. Available under commercial license — contact pradeep@xpertsystems.ai.

XpertSystems.ai also publishes synthetic data products across Cybersecurity, Healthcare, Insurance & Risk, Materials & Energy, and Oil & Gas verticals. Catalog: huggingface.co/xpertsystems.