Buckeye Oracle · by Octarine Analytics
Interrogate the docket.
Get the record back.
Buckeye Oracle is a research assistant for digging through PUCO filings: applications, testimony, staff reports, and public comments. Ask a question about a case and Buckeye Oracle loads the actual documents, then answers with direct quotes linked to the official record.
In practice
A question in, a cited answer out
No keyword roulette, no downloading forty PDFs. Below is an illustrative exchange: a portfolio question spanning two dozen dockets, plus a follow-up, answered with outcomes from the record and citations that link to the official filings.
Data centers (Amazon, Crown56, Viking, Standard Power) filed a cluster of AEC applications in 2023–2024 with mixed results — several were denied, which is relatively unusual historically.Case record, 24-0143-EL-AEC ↗
"The Applicant will add material load to the grid (18 MW, nearly continuously) while employing 12 individuals at an average wage of $16.00 per hour. At this time, Staff does not believe that this level of investment justifies the use of a special rate arrangement."Staff also noted that FirstEnergy's pending ESP case (23-0301-EL-SSO) might produce similar transmission billing options through normal ratemaking, making a special arrangement potentially unnecessary. The Commission agreed that Glorious Sky had not met its public-interest burden and denied the application on May 1, 2024 — the same day it denied Viking Data Center and SAI US. Staff Review and Recommendation, filed 2024-04-17 ↗
Illustrative session, condensed for length. Citations link to the actual PUCO DIS filings.
Why it's useful
Hours of docket-diving, done in minutes
A single rate case can span dozens of filings and thousands of pages. Buckeye Oracle turns that record into something you can interrogate.
Answers from the record, not from memory
Every response is grounded in the extracted text of actual filings. Key positions come back as direct quotes with an attribution line that links to the document's official PUCO DIS page.
From a live session: pinpointed Staff's recommended revenue requirement and ROE range from a 384-page staff report, with a working DIS citation link.The whole docket, not just the order
Applications, testimony, staff reports, motions, and public comments are all searchable, including intervenor positions across every party in a contested case.
From a live session: mapped rate-design positions of nine parties (OCC, OPAE, CUB Ohio, City of Dayton, OELC, OMAEG, Buckeye Power, Walmart, and AES Ohio), with quotes from each.Trends across cases, not just one docket
Ask a question that spans dozens of cases: every filing of a given type, how the Commission ruled, and how that's shifted year over year. You get the pattern, not just a list.
From a live session: inventoried 24 reasonable-arrangement (AEC) cases filed 2023–2026 with outcomes by year, surfacing a cluster of data-center denials in 2023–24.Designed to stay neutral
It describes what parties filed, quotes rather than paraphrases, and labels inferences as inferences. Ask "what's your read?" and you get analysis, clearly flagged as such.
Honest about its limits
If a case isn't in the database, a document has no extracted text, or a question needs access it doesn't have, it says so plainly, rather than guessing.
A record you can keep
Download any session as a text file (your questions and the cited answers), ready to share with a colleague or use as a verified-research starting point.
How it works
It researches the way an analyst would
Under the hood, the assistant runs a disciplined research workflow against a database of extracted PUCO filing text — the same steps a careful analyst takes, just faster.
Resolve the case
Give it a full case number like 25-0958-EL-AIR or a partial like 25-958, and it confirms the exact docket before reading anything.
Inventory the filings
It maps what's in the docket: which documents exist, which have extracted text, and where the testimony, staff reports, and applications are.
Search & read
Targeted keyword search narrows to the relevant passages; then it reads the full text of the documents that matter, prioritizing testimony and staff analysis.
Report with citations
You get a prose answer that leads with the finding, quotes the record verbatim, and links each citation to the official DIS page for verification.
Who it's for
Built for people who live in the docket (and those who'd rather not)
- Regulatory attorneys Scope a new docket, locate the testimony that matters, and get quotable passages with pinpoint sources before diving into the PDFs.
- Intervenors & consumer advocates Track party positions across concurrent cases without a dedicated research staff.
- Utility & energy analysts Pull outcome trends, rider histories, and ROE comparisons across years of filings.
- Journalists & researchers Separate original public comment from form-letter campaigns, and quote the record accurately.
FAQ
Questions? Asked and answered.
What exactly is Buckeye Oracle?
Buckeye Oracle is a conversational research assistant over a database of PUCO (Public Utilities Commission of Ohio) DIS (Docketing Information System) filings: case applications, testimony, staff reports, motions, and public comments. You ask questions in plain English; it reads the extracted document text and answers with quotes and citations. In other words, it's a lot of boring database infrastructure with a shiny and intuitive user interface. Built and operated by Octarine Analytics and powered by Anthropic's Claude models.
Is this legal advice?
No. Buckeye Oracle is first and foremost a research tool. It surfaces what's in the filings; it should not be used for legal conclusions or recommendations, and it's not a substitute for an analyst or licensed attorney. Outputs are AI-summarized and can contain errors, so every quote and citation should be verified against the source document before you rely on it. That's why every citation links directly to the official DIS record.
How accurate are the answers?
The assistant is instructed to quote rather than paraphrase for key positions, to label anything inferred rather than stated, and to link every citation to its source so you can validate with one click. That said, it's an AI system: treat its output as a well-organized starting point for verification, not as verified fact.
What cases and documents does it cover?
The database indexes 38,331 PUCO cases across utility types: electric (EL, ~22,700 cases), gas (GA, ~3,750), telephone, and wireless. Case metadata is well-populated across all of them.
Full extracted document text — what powers quote-level research — is concentrated in the electric dockets. For case types with thinner text coverage, the assistant will tell you up front what's readable, so you know the yield before you dig. We are regularly updating the database and index to expand the search and research capabilities of the application.
Do you have plans to expand coverage to other utilities or states?
To begin, we want to ensure that Buckeye Oracle is the best tool and database search assistant it can be for electricity dockets at the PUCO. If this goal is achieved and the application proves to be useful to folks and is sustainable for us to maintain, then yes! Do let us know which utilities or states you'd like to see next.
How current is the data?
The filing database and index are refreshed weekly, so recently filed documents in active dockets appear on that cycle. For anything filed in the last few days, check the PUCO DIS directly.
Can it compare multiple cases or spot trends?
Yes. Buckeye Oracle can compare party positions across dockets, tag and group cases by topic, and compute aggregate views (outcome trends, case volumes by year) across a whole class of cases. For overly broad questions it may ask you to narrow the scope (by case type, date range, or utility) rather than guess.
What happens when it doesn't know?
It should say so directly. If a case number isn't in the database, it reports that and asks for a corrected reference. If a document has no extracted text, it flags the gap where it matters. It's built to flag a gap rather than guess. But like any AI system, always validate against the actual filing before you rely on it. Buckeye Oracle is a research assistant, not a replacement for expert analysis.
Can I save my research?
Yes. Click Download session (or type /download) at any point to export the full conversation (your questions and the cited answers) as a text file.
Is the AI agent trained on case filings?
No. The core of this project is database and indexing infrastructure; the AI functions as a user interface and summary tool to analyze the query results from the database. In other words, none of these cases live in the 'memory' of the agent and each session will be completely fresh.
Are my questions confidential?
The plan is for paid subscriptions to run zero data retention by default: your queries and analysis wouldn't be kept around once the session ends, so download a transcript before you close the tab if you want a record. For the demo/beta, we're currently logging queries and outputs so we can improve the tool. Full details will live in the Privacy Policy once that's finalized.
How do I get access?
Accounts are currently provisioned by invitation while the tool is in its beta phase. Get in touch to request a demo or discuss access for your team. Seats are currently limited.