README.md

---
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
language:
  - en
tags:
  - synthetic
  - regulatory-compliance
  - audit-trails
  - capa
  - permit-violations
  - environmental-reporting
  - safety-compliance
  - cybersecurity-compliance
  - oil-and-gas
  - iso-45001
  - ccps
  - phmsa
  - bsee
  - epa-cwa
  - api-rp-754
  - api-rp-1173
  - nist-csf
  - iec-62443
  - sox
  - grc
  - enforcement-risk
pretty_name: "OIL-037 — Synthetic Regulatory Compliance Dataset (Sample)"
size_categories:
  - 100K<n<1M
---

# OIL-037 — Synthetic Regulatory Compliance Dataset (Sample)

A schema-identical preview of **OIL-037**, the XpertSystems.ai synthetic
regulatory compliance dataset for upstream, midstream, and downstream oil &
gas operations. The full product covers 8,500 facilities across 16 regulatory
frameworks (OSHA PSM, EPA CAA/CWA, API RP 754 / 1173, ISO 14001 / 45001,
PHMSA, BSEE, GHGRP, SOX, NERC CIP, IEC 62443, NIST, SEC Climate, Corporate
ESG). This sample is the generator's `sample` mode (500 facilities, 3-year
window from 2023–2025) covering all 12 product tables.

> **Built by** XpertSystems.ai — Synthetic Data Platform
> **Contact** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai) · [xpertsystems.ai](https://xpertsystems.ai)
> **License** CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

## What's inside

12 CSV tables covering the complete GRC (governance / risk / compliance)
lifecycle: facility master → audit trails → inspection findings → permit
violations → CAPA workflows → environmental reporting → safety compliance →
training → contractor compliance → escalation chains → cybersecurity
compliance → pre-built ML labels.

| Table | Rows (sample) | What it represents |
|---|---:|---|
| `compliance_master.csv` | 500 | 7-type facility master with regulation type, base risk, maturity |
| `audit_trails.csv` | 2,250 | 6-type audits (internal, third-party, regulatory, ESG, cyber, PSM) |
| `inspection_findings.csv` | 2,700 | 7-class findings with severity score and root cause |
| `permit_violations.csv` | 8,800 | 7-type permits × 7 violation codes; shutdown/override/referral flags |
| `capa_workflows.csv` | 5,000 | ISO 45001 CAPA lifecycle: opened/target/actual close, aging, effectiveness |
| `environmental_reporting.csv` | 64,000 | 8-type exceedances (CO₂/CH₄/VOC/SOₓ/NOₓ/produced water/flaring/spill) × 6 agencies |
| `safety_compliance.csv` | 20,000 | OSHA recordable flags + API 754 Tier 1–4 PSE classification |
| `training_records.csv` | 30,000 | 8-module training × contractor flag × expiration tracking |
| `contractor_compliance.csv` | 3,000 | ISNetworld/Avetta-style pre-qualification audits |
| `escalation_chains.csv` | 2,500 | 6-level escalation (site → regional → legal → exec → regulator → board) |
| `cybersecurity_compliance.csv` | 6,000 | NIST CSF / NERC CIP / IEC 62443 / SOX ITGC × 7 control families |
| `compliance_labels.csv` | 500 | Pre-built ML labels: risk score, enforcement & shutdown probabilities, priority segment |

Total: ~145,000 rows, ~13 MB. The full OIL-037 product is ~6 million rows.

---

## Calibration sources

Every distribution and ratio is anchored to **named public references**. The
validation scorecard (see below) re-scores observed vs. target for 10
industry-anchored metrics, every one citing its source. Highlights:

- **ISO 45001:2018 Clause 10.2** — OH&S corrective action overdue rate benchmark.
- **CCPS Risk-Based Process Safety** — repeat audit finding rate maturity bands.
- **CCPS Auditing Guidelines** — CAPA closure aging distribution.
- **API RP 754** — Process Safety Performance Indicators (Tier 1–4 event mix).
- **API RP 1173** — Pipeline Safety Management System audit benchmarks.
- **ISO 19011** — Guidelines for auditing management systems.
- **BSEE / PHMSA inspection statistics** — critical finding severity ratio.
- **EPA NEI / TRI compliance reporting** — environmental exceedance baselines.
- **PHMSA HL pipeline + BSEE OCS enforcement data** — shutdown order frequency.
- **IPIECA / OGUK incident-management benchmarks** — escalation resolution rate.
- **ISNetworld / Avetta** — contractor pre-qualification insurance-validity baseline.

---

## 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 | CAPA Overdue Rate (ceiling) | ≤ 0.25 | **0.184** | ISO 45001:2018 |
| M02 | Repeat Audit-Finding Rate (ceiling) | ≤ 0.25 | **0.067** | CCPS RBPS |
| M03 | Critical Finding Share (ceiling) | ≤ 0.075 | **0.034** | BSEE / PHMSA |
| M04 | Env Exceedance Rate (ceiling) | ≤ 0.035 | **0.026** | EPA NEI / TRI |
| M05 | Enforcement Shutdown Rate (ceiling) | ≤ 0.055 | **0.017** | PHMSA / BSEE |
| M06 | API 754 PSE Tier 1+2 Share | 0.035–0.065 | **0.049** | API RP 754 |
| M07 | CAPA Aging Days (median) | 30–60 | **42** | CCPS / ISO 45001 |
| M08 | Escalation Resolution Rate | 0.62–0.78 | **0.709** | IPIECA / OGUK |
| M09 | Contractor Insurance Valid (floor) | ≥ 0.89 | **0.928** | ISNetworld / Avetta |
| M10 | Audit Score Median (floor) | ≥ 80 (0–100) | **93.8** | API RP 1173 / ISO 19011 |

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

---

## Suggested use cases

- **Enforcement-risk prediction** — train classifiers that predict
  `enforcement_probability` or `regulatory_shutdown_probability` from facility
  context, finding history, environmental exceedances, and CAPA closure
  patterns. Pre-built labels in `compliance_labels.csv`.
- **CAPA closure-aging forecasting** — regression on `aging_days` from finding
  severity, facility maturity, contractor ratio, and effectiveness-check
  history. Practical for compliance-team capacity planning.
- **GRC operational dashboards** — `compliance_master.csv` × `compliance_labels.csv`
  on `facility_id` builds a complete priority-segmented portfolio view across
  500 facilities for portfolio-level risk benchmarking.
- **Audit-finding root-cause modeling** — 7-class `root_cause_category` ×
  4-class severity ladder on ~2,700 findings enables HUMAN_ERROR vs.
  PROCEDURE_GAP vs. EQUIPMENT_CONDITION decomposition for audit-program
  improvement studies.
- **Escalation pathway analysis** — `escalation_chains.csv` has 6-level
  escalation ladder + response-time-hours + whistleblower/legal flags,
  enabling temporal-process-mining and bottleneck analytics.
- **Multi-framework cyber compliance benchmarking** — NIST CSF / NERC CIP /
  IEC 62443 / SOX ITGC × 7 control-family decomposition supports cross-
  framework gap-analysis modeling.
- **Environmental exceedance early-warning** — 8 exceedance types × 6
  agencies × 64K reporting periods enables agency-specific exceedance and
  late-submission classifiers.

---

## Loading

```python
from datasets import load_dataset

facilities = load_dataset(
    "xpertsystems/oil037-sample",
    data_files="compliance_master.csv",
    split="train",
)
findings = load_dataset(
    "xpertsystems/oil037-sample",
    data_files="inspection_findings.csv",
    split="train",
)
labels = load_dataset(
    "xpertsystems/oil037-sample",
    data_files="compliance_labels.csv",
    split="train",
)
```

Or with pandas directly:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/oil037-sample",
    filename="capa_workflows.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)
```

All 12 tables join on `facility_id`. Cross-table joins also available on
`audit_id` (audits ↔ findings), `finding_id` (findings ↔ violations ↔ capa),
`violation_id` (violations ↔ capa ↔ safety ↔ escalation), and `capa_id`
(capa ↔ escalation).

---

## Schema highlights

**`compliance_master.csv`** — `facility_id`, `facility_type` (7-class:
REFINERY / OFFSHORE_PLATFORM / LNG_TERMINAL / PIPELINE_SYSTEM / TANK_FARM /
DRILLING_SITE / PETROCHEMICAL_PLANT), `region` (9-class), `operator_entity`,
`regulation_type` (16-class), `operational_maturity_score` ∈ [0.05, 0.99],
`base_compliance_risk` ∈ [0, 1], `contractor_ratio`,
`environmental_sensitivity`, `cyber_maturity_score`, `compliance_status` ∈
{COMPLIANT, WATCH, NON_COMPLIANT, ENFORCEMENT_RISK}, `last_major_audit_date`,
`active_permit_count`, `employee_count`.

**`inspection_findings.csv`** — `finding_id`, `audit_id`, `facility_id`,
`detected_date`, `finding_type` (7-class: DOCUMENTATION_GAP / TRAINING_EXPIRY
/ EQUIPMENT_CERTIFICATION / PERMIT_BREACH / ENV_EXCEEDANCE /
CYBER_CONTROL_GAP / PROCESS_SAFETY_GAP), `severity_level` ∈ {LOW, MEDIUM,
HIGH, CRITICAL}, `severity_score` ∈ [0, 1], `repeat_finding_flag`,
`regulation_type` (16-class), `root_cause_category` (7-class),
`required_remediation_due_date`, `inspector_variability_score`.

**`safety_compliance.csv`** — `osha_recordable_flag`,
`process_safety_tier` ∈ {NONE, TIER_4, TIER_3, TIER_2, TIER_1} (API RP 754
performance indicators), 7-class `compliance_gap`, 6-class
`corrective_action`.

**`cybersecurity_compliance.csv`** — 5-class `framework` ∈ {NERC_CIP,
IEC_62443, NIST_CSF, SOX_ITGC, CORPORATE_CYBER}, 7-class `control_family`
(ACCESS_CONTROL / ASSET_INVENTORY / CHANGE_MANAGEMENT / LOGGING /
INCIDENT_RESPONSE / VULNERABILITY_MGMT / NETWORK_SEGMENTATION),
`compliance_gap` ∈ {NONE, MINOR, MAJOR}, `audit_result` ∈ {PASS,
CONDITIONAL_PASS, FAIL}, `gap_score`, `ot_asset_exposure_flag`,
`remediation_sla_days`.

**`compliance_labels.csv` — pre-built ML labels per facility**:
`risk_score`, `enforcement_probability`, `regulatory_shutdown_probability`,
`compliance_grade` ∈ {A, B, C, D, F}, `priority_segment` ∈ {LOW_RISK,
WATCHLIST, HIGH_RISK, CRITICAL_RISK}.

---

## Calibration notes & limitations

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

1. **Training-record certification status skew.** In `training_records.csv`,
   the `CURRENT` rate at sample scale is ~13% with ~86% `EXPIRED`, well below
   the API RP T-1 mature target (≥ 80% current). This is a generator quirk
   in the expiration-date sampling logic (random expirations uniformly drawn
   from -180 to +365 days around random in-window dates skew most flags to
   the EXPIRED side). For ML utility, **use `completion_score` and
   `days_until_expiry` as primary features** rather than the
   `certification_status` enum, or filter to `days_until_expiry > 0` and treat
   the rest as a separate compliance-debt subset. The full OIL-037 product
   ships a re-calibrated training mode-pack.

2. **`compliance_labels.csv` grade-class skew.** ~91% of facilities receive
   grade "F" because the grade is computed as `compliance_grade(1 - risk_score)`
   but the risk-score formula at sample scale concentrates between 0.55–0.75,
   mapping to F under the threshold ladder. For **balanced multi-class training**,
   use `priority_segment` (4-class: LOW_RISK / WATCHLIST / HIGH_RISK /
   CRITICAL_RISK), which is well-spread at sample scale, or rebuild a custom
   grade column from `risk_score` directly with quintile thresholds.

3. **Cybersecurity compliance pass rate.** At sample scale, the cyber audit
   PASS rate is ~25% and FAIL ~21% — these reflect the generator's strict
   cyber_gap formula and are **below NIST CSF mature-program baselines** of
   ≥60% PASS. The 7-control-family taxonomy and 5-framework distribution are
   accurate; the categorical pass/fail thresholds are intentionally
   conservative for ML utility. The validation scorecard does **not** validate
   cyber audit results against NIST CSF maturity claims for this reason.

4. **OSHA recordable rate.** Sample-scale `osha_recordable_flag` rate (~3.8%)
   is incident-dense relative to BLS oil & gas TRIR (~0.8–1.5% in raw event
   terms). This is intentional ML-utility scaling (similar to OIL-035) so the
   500-facility sample has trainable positive-class density. The full product
   recovers realistic upstream BLS rates.

5. **Date-window length.** Default window is 2023-01-01 to 2025-12-31 (3
   years). For longer-horizon trend modeling, use the full product or
   override `--start-date` / `--end-date` on the underlying generator.

6. **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-037** product covers 8,500 facilities across a configurable
multi-year horizon (~6 million rows total), with re-calibrated training
certification status logic, balanced compliance-label distributions, and
optional NIST CSF maturity-aligned cyber-audit mode-packs. Available under
commercial license — contact
[pradeep@xpertsystems.ai](mailto: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](https://huggingface.co/xpertsystems).