Primary audience
Refiners, physical suppliers, traders, supply planners, risk teams, and analysts who need planned and unplanned capacity events with evidence and confidence.
Energy data research
Experiment 3 · operational event evidence
Refinery outage data: a public-evidence feasibility experiment
Can public regulatory and operator evidence identify refinery outages quickly and accurately enough to support a useful event history without pretending that a proxy is a confirmed outage?
Status
collection probe
Initial scope
EIA refinery universe and methodology, Texas TCEQ emission-event reporting, and NOAA public incident records
Collection cadence
Hourly checks for fast public signals; daily evidence review; explicit closure tracking
Last reviewed
2026-08-11
This is not a live refinery-outage feed, alert service, or production dataset. Public signals shown here are research inputs, not confirmed operating status.
Decision context
What this experiment must decide
Decision question
Can a public-source event ledger reach commercially useful timeliness and accuracy, or is licensed/analyst-confirmed data required from the beginning?
Why collect now
Refinery outages have the strongest willingness-to-pay hypothesis in the opportunity screen, but also the largest risk of false positives, stale events, licensing cost, and misleading capacity claims.
P · Problem
Problem
A refinery outage is not one simple field. A whole-site shutdown, crude unit reduction, FCC maintenance, power interruption, flaring event, and planned turnaround have different market effects. Public reporting may describe only an emission, safety event, or maintenance activity—not the unit capacity that is actually offline.
Commercial services combine source access, analyst confirmation, facility and unit mapping, expected return dates, and revisions. Public sources are fragmented and can arrive after the market learns about an event. A fast but wrong alert can be worse than no alert for supply or trading workflows.
This experiment therefore begins as an evidence ledger. Every candidate event retains its source, first-seen time, claims, contradictions, confidence, and closure evidence. It does not transform an emission report automatically into an outage claim.
A · Approach
Approach
The collection design is intentionally narrower than a production data platform. It preserves evidence, makes uncertainty explicit, and produces a measurable build/no-build decision.
- 1
Define a refinery universe
Use an official facility and capacity reference so operator names, refinery identities, and baseline capacities do not depend on news text.
- 2
Collect public signals
Monitor legally accessible operator statements, regulatory emission-event records, incident data, and official local notices. Retain the source artifact and retrieval time.
- 3
Classify without overclaiming
Label each item as signal, probable event, confirmed unit event, confirmed site event, resolved, or contradicted. An emission event alone remains a signal.
- 4
Resolve entities and units
Map company, facility, unit aliases, and capacity only when an authoritative reference supports the relationship.
- 5
Track the full lifecycle
Record planned/unplanned status, start, expected return, updates, actual resolution, and uncertainty. Do not leave an event permanently open after the source stops discussing it.
- 6
Benchmark against trusted records
Compare the public ledger with design-partner logs or a licensed sample to measure detection rate, delay, false positives, and capacity error.
R · Research
Research
These are current findings and their decision implications. They are not promises about future OilPriceAPI coverage.
| Issue | Observed finding | Decision implication |
|---|---|---|
| EIA does not provide a direct outage series | EIA explains that its data do not reflect outages directly; significant drops in unit inputs can act as indirect evidence. | Input changes can support retrospective analysis but must not be labeled a direct event feed. |
| Texas exposes operational proxies | TCEQ’s event database includes incident location, compounds, quantities, duration, export, and an RSS feed updated hourly. | The source may offer timely signals for Texas facilities, but emissions events include maintenance, startup, shutdown, and other causes. |
| Final reports can arrive later | TCEQ describes initial notification requirements and a final report due after an event ends. | The ledger needs provisional and final versions instead of freezing the first report as truth. |
| Public incident archives are selective | NOAA publishes machine-readable data for selected incidents where its response office provided scientific support. | NOAA can contribute evidence, but absence from the archive cannot imply normal refinery operations. |
| Commercial products validate demand | The research scan found established refinery intelligence offerings from IIR, Kpler, Wood Mackenzie, Rystad, and S&P Global. | A paid category exists, but public-source economics must be compared with licensing or partnership rather than assumed superior. |
M · Measurement
Measurement
Primary decision metric
Confirmed-event recall at useful latency, measured against a design-partner or licensed reference set and reported alongside false-positive rate.
Detection recall
Reference events detected by the public ledger divided by reference events.
False-positive rate
Candidate events rejected after review divided by all candidates.
First-signal latency
Time between reference event start and first retained public signal.
Confirmation latency
Time between event start and evidence sufficient for confirmed status.
Capacity error
Difference between estimated and validated offline capacity, when a comparable number exists.
Resolution completeness
Confirmed events with supported end state divided by confirmed events.
Contradiction rate
Events with materially conflicting sources and how quickly conflicts are resolved.
Analyst minutes
Manual research time required per confirmed event.
Continue when
- ✓Public evidence detects at least 80% of buyer-relevant events in the validation set with an agreed useful latency.
- ✓False-positive status remains below 10% after the defined confirmation window, with every status change auditable.
- ✓At least one buyer funds a pilot or one licensed partner permits a validation sample before production build-out.
Stop or redesign when
- ×Public signals cannot distinguish operationally meaningful outages from emissions, maintenance, or reporting noise at acceptable accuracy.
- ×Required source licenses and analyst review exceed expected contract economics or cannot support the intended customer use.
Initial collection
Started
2026-08-11T21:35:40Z
Current status
All three public evidence and methodology surfaces returned HTTP 200 in the first bounded capture.
Baseline observations
- •EIA explicitly cautions that its data do not reflect outages directly, though unit-input changes may reveal outages retrospectively.
- •TCEQ exposes searchable emissions and maintenance-event reports with durations and spreadsheet export; its RSS feed is described as hourly.
- •NOAA offers raw incident data only for selected responses where its office provided scientific support.
- •None of these sources alone supplies a complete, confirmed national refinery-outage feed.
Next capture
Select ten refineries, map their official identities, and log public signals for fourteen days before comparing against a trusted event set.
Proposed record schema
The schema is a test contract, not a public endpoint specification. Unknown values remain unknown, and every derived field must retain source evidence.
- facility
- Canonical refinery identity and location.
- unit
- Affected process unit when evidenced; otherwise unknown.
- event_type
- Planned maintenance, unplanned outage, reduction, startup, shutdown, or unresolved signal.
- status
- Signal, probable, confirmed, resolved, or contradicted.
- capacity_affected
- Reported or estimated capacity with method and confidence.
- first_seen_at
- When the collector first retained evidence.
- event_start_at
- Claimed start time, separate from discovery.
- confidence
- Rule-based level supported by listed evidence, not an unexplained score.
Sources and scope
OilPriceAPI Research prepared this page to test a product decision. The method combines direct-source review, immutable collection probes, and explicit decision gates. External links establish what each source publishes; they do not grant OilPriceAPI redistribution rights or prove complete coverage.
Refinery outages: description and potential impact
U.S. Energy Information Administration
Supports: Official explanation of outage effects and the limitation that EIA data do not directly report outages.
Limitation: The analysis is methodological and historical, not a current event feed.
Reports of air emissions events
Texas Commission on Environmental Quality
Supports: Public search/export fields, duration information, and hourly RSS description.
Limitation: An emissions, maintenance, startup, or shutdown report is a signal—not proof of market-relevant capacity loss.
Raw incident data
NOAA Office of Response and Restoration
Supports: Machine-readable public incident records for selected supported responses.
Limitation: Coverage is selective; missing records cannot be interpreted as no outage.
Help validate the next decision
If this dataset is part of your workflow, tell us which sources matter, what failure looks like, and what evidence would make the data useful enough to buy. A conversation is evidence; a page view is not.