Reference design

Energy intelligence for governed operational decisions

A reference design for drilling performance and contracts, land and lease obligations, and operational health. The patterns are examples. They are not customer deployments and they do not report measured savings.

A documented design: the data, rules, architecture, controls, and review steps a build would follow. Not a delivered system or a customer deployment.

Energy operating site: land parcels, a drilling and production asset, and processing and storage infrastructure, with a person reviewing a governed result.

Three patterns, shown together. Each one reaches a governed result, then runtime meaning, then a person. They do not have to run in order. Example reference design, not a deployed system.

Start with the worked example

Two downtime records overlap on the same asset. Added together they count 4 hours, but the shared hour is counted once, so they cover 3 elapsed hours. Commercial exposure is not calculated until the business agrees how the shared hour is treated.

Worked example: overlapping downtime records

Worked example.

Equipment08:00–10:00

Weather09:00–11:00

Both

Union of supplied intervals — 3 elapsed hours.

Calculated from the example timestamps; operational meaning awaits review.

4 hours across two records

1 shared hour, counted once

Operations: validate what the records represent.

Commercial owner: establish allocation and applicable treatment.

Commercial exposure: not calculated.

Awaiting review

Read the worked example

The three patterns below are reference designs: drilling performance and contracts, land and lease, and operational health.

How Energy connects to the business functions

Energy is an industry, not a sixth business function. An industry pattern applies the shared architecture to decisions specific to a sector, and one decision can span more than one business function. The links below connect each Energy decision to the functions its documented review involves.

One pattern at a time

Each pattern uses the same Agentic AI Runtime and a separate human review. The runtime explains the governed result. It does not establish an executed Energy agent, invent a cause, or take the action.

Show
Drilling and contract review. Design example, laid out like a review timeline. Not a customer deployment and not a measured result.

Sources

Example categories

Rule and context

Approved and versioned

Inside the Rule and context boundary

Governed result

Reviewer view

Runtime interpretation

Shared runtime

Human review

Before it is a finding

Flow stage 3: Governed result

Review list

A short list of periods for a person to look at, with the governing fields attached so the reviewer can see why each one is listed.

  1. SourcesThe records the decision depends on, at a declared grain and period.
  2. Rule and contextAn approved rule and its context decide what is listed. The model does not set them.
  3. Governed resultThe result carries the fields that produced it, and missing data stays labeled as missing.
  4. Agentic AI Runtime interpretationThe runtime reads the approved result and its context through approved access, explains the declared fields, and proposes a finding. It adds no calculation, cause, or missing fact.
  5. Human reviewA person confirms the proposed finding before it is treated as a finding. Nothing is confirmed automatically.
  • Records or result that crosses a stage
  • Context, approved access, or policy
  • Human review
  • Rule and context boundary
  • Where Orisdale designs the decision, context, or validation
Every relationship as text
  • Well performance records to Approved exception rule (data or result): Well-day records reach the approved rule.
  • Contract terms to Period and rate context (data or result): Contract terms reach the context definitions.
  • Approved exception rule to Review list (data or result): The approved rule produces the governed result (Exception rule applied).
  • Period and rate context to Review list (workflow, query, or policy): The context definitions travel with the result (Period and rate fields).
  • Review list to Agentic AI Runtime (data or result): The governed result and its context go to the Agentic AI Runtime. The runtime reads only these approved outputs, not the raw sources.
  • Agentic AI Runtime to Review list (workflow, query, or policy): The runtime reads the governed result through approved tools and queries. It does not open the raw sources or run a calculation of its own.
  • Agentic AI Runtime to Analyst confirms the exception (human review): The runtime presents a Proposed exception for a person to review. It is not treated as a finding until they confirm it.

The runtime stage is the same Agentic AI Runtime described in the shared reference architecture. In this reference design it describes an intended responsibility. It does not establish an executed or deployed Energy agent. Request a Strategy Session

Land and lease obligation review. Example parcel schematic, not a map. Not a customer deployment and not a measured result.

Sources

Example categories

Rule and context

Approved and versioned

Inside the Rule and context boundary

Governed result

Reviewer view

Runtime interpretation

Shared runtime

Human review

Before it is a finding

Flow stage 3: Governed result

Obligation queue

A queue that shows the governed status, the relevant date, and the fields the status was derived from.

  1. SourcesThe records the decision depends on, at a declared grain and period.
  2. Rule and contextAn approved rule and its context decide what is listed. The model does not set them.
  3. Governed resultThe result carries the fields that produced it, and missing data stays labeled as missing.
  4. Agentic AI Runtime interpretationThe runtime reads the approved result and its context through approved access, explains the declared fields, and proposes a finding. It adds no calculation, cause, or missing fact.
  5. Human reviewA person confirms the proposed finding before it is treated as a finding. Nothing is confirmed automatically.
  • Records or result that crosses a stage
  • Context, approved access, or policy
  • Human review
  • Rule and context boundary
  • Where Orisdale designs the decision, context, or validation
Every relationship as text
  • Lease record to Approved status rule (data or result): Lease and tract reach the approved rule.
  • Obligation and as-of dates to Lease and date context (data or result): Dates reach the context definitions.
  • Approved status rule to Obligation queue (data or result): The approved rule produces the governed result (Status rule applied).
  • Lease and date context to Obligation queue (workflow, query, or policy): The context definitions travel with the result (Status and date meaning).
  • Obligation queue to Agentic AI Runtime (data or result): The governed result and its context go to the Agentic AI Runtime. The runtime reads only these approved outputs, not the raw sources.
  • Agentic AI Runtime to Obligation queue (workflow, query, or policy): The runtime reads the governed result through approved tools and queries. It does not open the raw sources or run a calculation of its own.
  • Agentic AI Runtime to Land staff confirm a status change (human review): The runtime presents a Proposed status change for a person to review. It is not treated as a finding until they confirm it.

The runtime stage is the same Agentic AI Runtime described in the shared reference architecture. In this reference design it describes an intended responsibility. It does not establish an executed or deployed Energy agent. Request a Strategy Session

Operational health review. Example review queue. An incomplete record stays labeled incomplete. Not a customer deployment and not a measured result.

Sources

Example categories

Rule and context

Approved and versioned

Inside the Rule and context boundary

Governed result

Reviewer view

Runtime interpretation

Shared runtime

Human review

Before it is a finding

Flow stage 3: Governed result

Prioritized review list

Each row carries the governed priority and the fields that produced it. An incomplete record is listed as incomplete, not completed by a model.

  1. SourcesThe records the decision depends on, at a declared grain and period.
  2. Rule and contextAn approved rule and its context decide what is listed. The model does not set them.
  3. Governed resultThe result carries the fields that produced it, and missing data stays labeled as missing.
  4. Agentic AI Runtime interpretationThe runtime reads the approved result and its context through approved access, explains the declared fields, and proposes a finding. It adds no calculation, cause, or missing fact.
  5. Human reviewA person confirms the proposed finding before it is treated as a finding. Nothing is confirmed automatically.
  • Records or result that crosses a stage
  • Context, approved access, or policy
  • Human review
  • Rule and context boundary
  • Where Orisdale designs the decision, context, or validation
Every relationship as text
  • Assets and work orders to Documented priority rule (data or result): Asset and work order reach the approved rule.
  • Downtime and history records to Category definitions (data or result): Downtime records reach the context definitions.
  • Documented priority rule to Prioritized review list (data or result): The approved rule produces the governed result (Priority rule applied).
  • Category definitions to Prioritized review list (workflow, query, or policy): The context definitions travel with the result (Category meaning).
  • Prioritized review list to Agentic AI Runtime (data or result): The governed result and its context go to the Agentic AI Runtime. The runtime reads only these approved outputs, not the raw sources.
  • Agentic AI Runtime to Prioritized review list (workflow, query, or policy): The runtime reads the governed result through approved tools and queries. It does not open the raw sources or run a calculation of its own.
  • Agentic AI Runtime to Reliability staff confirm the item (human review): The runtime presents a Proposed priority for a person to review. It is not treated as a finding until they confirm it.

The runtime stage is the same Agentic AI Runtime described in the shared reference architecture. In this reference design it describes an intended responsibility. It does not establish an executed or deployed Energy agent. Request a Strategy Session

Drilling Performance and Contract Intelligence

Maturity: Reference design

Which well-days or contract periods need review?

Source categories
Well operations, drilling time, contract terms, and cost categories.
Governed result
A review list of periods that meet an approved exception rule.
Agentic AI Runtime interpretation
Explains the declared period, contract terms, and rule fields, and proposes an exception. It invents no cause, rate, or comparison.
Human review
A person confirms the exception before it is treated as a finding.

See the overlapping-downtime worked example

Full reference design contract
User
A drilling-performance or contract analyst preparing a review, not an autonomous field system.
Decision
Which wells, rigs, or contract periods need a person to look before a performance or cost conversation.
Source-data categories
Example categories: well operations, drilling time breakdowns, contract terms, and cost categories. These are pattern names, not a named operator's system.
Reporting grain and time
A well or rig, by day or by contract period. The reporting period has to be declared before two periods are compared.
Governed calculations
Time classification, application of an approved contract rate, and an exception flag the business has already defined. The reference design does not publish a numeric threshold.
Expected outputs
A review list of periods that meet the approved exception rule, with the governing fields attached so a reviewer can see why the period was listed.
Interpretation
A business-facing agent may explain those governed fields in the Gemini Enterprise app. It may not invent a cause, a rate, or a comparison the rule did not produce.
Missing data
A missing time code, rate, or period stays missing and is labeled as missing. The model does not fill it in.
Human review
A person confirms the exception before it is treated as an operational finding or a contract claim.

Land and Lease Intelligence

Maturity: Reference design

Which obligations need attention?

Source categories
Lease records, obligation dates, ownership interests, and acreage status.
Governed result
An obligation queue with the governed status and the relevant dates.
Agentic AI Runtime interpretation
Restates the governed status, the parcel and obligation relationships, and both dates, and proposes a status change. It infers no ownership or satisfied obligation.
Human review
Land staff confirm a status change before it is the position of record.
Full reference design contract
User
A land or lease administrator tracking obligations and acreage positions.
Decision
Which leases, obligations, or acreage positions need attention before a date or a commitment passes.
Source-data categories
Example categories: lease records, obligation dates, ownership interests, and acreage status. No company's land file is used.
Reporting grain and time
A lease or tract. The obligation date and the as-of date are both stated. A status without an as-of date is not a current status.
Governed calculations
Date classification, an interest calculation the land team has approved, and an obligation status. Unpublished formulas are not implied.
Expected outputs
An obligation review queue that shows the governed status, the relevant date, and the fields the status was derived from.
Interpretation
The agent may restate the governed status and dates. It may not infer that an obligation was satisfied when the status does not say so.
Missing data
A missing date or interest is reported as missing. It is not estimated for the queue.
Human review
Land staff confirm any status change before it is recorded as the position of record.

Operational Health and Maintenance Intelligence

Maturity: Reference design

Which assets or work orders need review?

Source categories
An asset hierarchy, work orders, downtime categories, and maintenance history.
Governed result
A prioritized review list carrying the governed priority.
Agentic AI Runtime interpretation
Explains the governed priority and its fields, and proposes an item for review. It guesses no failure cause and completes no incomplete record.
Human review
Reliability staff confirm the item before a maintenance action is recommended.
Full reference design contract
User
A maintenance or reliability lead reviewing assets and work.
Decision
Which assets or work orders need review under the health and priority rules the maintenance function already owns.
Source-data categories
Example categories: an asset hierarchy, work orders, downtime categories, and maintenance history.
Reporting grain and time
An asset or a work order over a declared operating period. Comparisons use that same period.
Governed calculations
Downtime classification, a priority rule, and an overdue status. The reference design does not state a failure rate, an accuracy figure, or a savings estimate.
Expected outputs
A prioritized review list. Each row carries the governed priority and the fields that produced it.
Interpretation
The agent may describe the governed priority in plain language. It may not invent a failure cause or recommend a part.
Missing data
An incomplete work order or downtime record stays incomplete and is labeled. The list does not guess the missing category.
Human review
Reliability staff confirm the item before a maintenance action is recommended.

Proposed architecture

This is a proposed reference pattern. It is not a demonstrated Energy runtime, and it is not a claim that Alteryx, Google Cloud, or BigQuery endorses Orisdale. Gemini Enterprise app is the business-facing experience. The Agent Platform is the technical build and governance surface where that path applies. Orisdale's work in the pattern is decision design, semantic context, implementation, and validation.

  1. 01Enterprise operational sources
  2. 02Alteryx One governed preparation and business logic
  3. 03BigQuery governed analytical outputs
  4. 04Semantic Intelligence
  5. 05Gemini Enterprise Agent Platform, where a technical agent is built and governed
  6. 06Gemini Enterprise app for the business-facing experience
  7. 07Human review before an operational finding is treated as confirmed

Related pages: Google Cloud, Alteryx One, Architecture, and the Workflow Agent. Workflow authoring and this decision design are different pieces of work.

Questions buyers ask

What is the current scope?

Three documented designs: drilling performance and contract intelligence, land and lease intelligence, and operational health and maintenance intelligence. Retail and healthcare are not part of this page.

How mature is it?

Reference design. Each design states the user, decision, data categories, grain, calculations, outputs, interpretation limit, missing-data behavior, and human review. All three are proposed designs.

What data does it require?

The relevant operational categories named in each design, at a declared grain and time context. The page uses example categories. It does not include a company's records.

Where does a person review the work?

A person confirms an exception, a status change, or a maintenance priority before it is treated as an operational finding. The agent may explain governed fields. It may not invent a cause.

What is the next step?

Request a Strategy Session about one of these decisions. The session does not promise a result, a threshold, or a production date.

Discuss an Energy decision