NPT Is Not a Contract Charge: Turn Downtime Records into a Reviewable Exposure Assessment

Non-productive-time records can support a commercial review, but recorded hours are not automatically billable hours, lost revenue, or contractual liability. A useful assessment first resolves the operational intervals, then applies an agreed interpretation of the relevant contract, and finally presents the evidence and unresolved questions to the accountable reviewer.

One event can produce several records

Consider a drilling operation with equipment and weather records covering part of the same morning. The operational report may need both descriptions. Maintenance, operations, and commercial teams may each need a different view of the incident. None of that means every recorded hour should be added into a single financial total.

The commercial question is more specific: which intervals are relevant to a particular agreement, what conditions apply, and what can currently be established from the evidence? The answer can differ from the operational classification even when both are valid for their own purpose.

An AI-generated narrative becomes risky when it jumps from "four recorded hours across two events" to "four hours of exposure." That sentence has skipped interval reconciliation, contract applicability, and the review of attribution. A professional workflow makes those steps visible rather than presenting a calculated-looking amount too early.

Establish the operational time boundary first

The initial contract for the analysis should define the asset, activity, reporting window, timezone, and event-identity policy. It should identify which system supplies the accepted start and end timestamps and how later corrections are handled. Open-ended events and contradictory timestamps need a defined exception route.

Events should only be combined where the analysis intends to describe the same resource or scope. Two rigs stopped for two hours do not create the same measure as one rig stopped for four hours. Asset-hours and elapsed time are different quantities. A group-level summary needs to preserve that distinction.

For the synthetic example below, both records concern the same asset and analysis window and use the same timezone. The timestamps are supplied. They are not, by themselves, validated operational events. Intervals are treated as start-inclusive and end-exclusive, so a boundary instant is not counted twice. That convention is an analysis rule for this example, not a contractual interpretation.

A three-hour interval with four recorded hours

A careful statement describes the supplied timestamps and says that operational meaning awaits review. A different statement—"equipment caused two chargeable hours"—requires a rule and supporting evidence that this example does not supply.

This example is deliberately numerical without being financial. It does not establish an operational interruption, an approved allocation, applicable commercial treatment, commercial exposure, lost revenue, a receivable, or an invoice. EX-AGREE-00 is fictional, and its applicability is not established.

Treat contract applicability as a separate decision

The relevant agreement may define covered activities, effective dates, rate treatments, exclusions, notice requirements, or evidence needed for a particular claim. Which provisions matter depends on the actual agreement and the situation. An implementation should not infer universal rules from the words "NPT" or "downtime."

The business needs a named contract owner and an approved interpretation that can be implemented as bounded logic. The review record should identify the agreement and version, the relevant provision reference, the applicable operational scope, and any unresolved interpretation.

Document retrieval can help locate a clause, but finding text is not the same as establishing that it governs the event. An amendment may change the treatment. A date may place the event outside the agreement's effective period. The source may be incomplete. Those conditions belong in the output, not in an invisible confidence adjustment.

Where interpretation requires specialist commercial or legal judgment, the system should route the question to that process. It should not make the model's preferred reading appear to be an approved business rule.

Preserve unresolved states in the exposure assessment

A proposed assessment can distinguish three useful states: the operational interval is established; the applicable treatment is established under an approved rule; or a required issue remains unresolved. These states are more informative than a single green or red result.

For example, supplied timestamps can be used to calculate a three-hour union while attribution of the overlapping hour remains unresolved. That does not justify allocating the hour to the first record, splitting it equally, or choosing whichever explanation appears most often in the notes. Any allocation needs its own agreed basis. Operations and commercial review stay separate. This example records no decision and no investigation request.

If a bounded scenario estimate is useful, show its assumptions explicitly and keep it distinct from confirmed treatment. An assumption-based estimate should not be labeled a receivable, liability, or invoice instruction. The commercial reviewer may accept a treatment, request further evidence, or keep the matter open.

Give AI an evidence assembly task

The interpretation layer can help assemble a briefing from approved outputs: the affected intervals, relevant records, rule version, proposed treatment, and missing information. It can explain why two overlapping records produce three elapsed hours and identify which allocation question prevents a stronger conclusion.

It should not add an unsupported cause, choose a contract version because its wording is convenient, or convert an uncertain classification into a financial claim. If the records disagree, the briefing should make the disagreement easy to inspect.

A useful review item might read: "The interval is reconciled. Treatment of 09:00–10:00 is unresolved because both event categories are recorded and no approved allocation decision is linked." That tells the reviewer what is known and what action would move the case forward. It is more useful than an impressive-looking exposure total whose assumptions cannot be found.

Implement around traceable records

The proposed architecture retains the source event identifiers and their revisions, normalizes time under the agreed convention, and calculates intervals at the defined resource grain. A separate rules representation records approved applicability and treatment. The review output joins those results without merging their authority.

The customer's data platform can hold the source and governed outputs according to its architecture. Preparation or business-logic tooling can implement the agreed transformations where appropriate. An AI runtime then reads permitted results and evidence. There is no need to imply that every component is a different system or that all computation moves through one mandatory vendor sequence.

The record carried into review should be reproducible: source versions, analysis cutoff, rules version, result, unresolved issues, and the person's disposition. If an event is corrected later, preserve the earlier assessment and show which conclusions changed.

Test cases that challenge the easy answer

Start with the overlap example, then test adjacent intervals, duplicate records, nested intervals, a missing end time, and records from different assets. Confirm that timestamp normalization does not manufacture a duration or hide an inconsistency.

Contract tests should include an amendment, an event outside the effective window, a missing provision reference, and a treatment requiring evidence that is absent. Narrative tests should challenge the words "caused," "chargeable," and "confirmed." Each must have an explicit basis when it appears.

Acceptance should also cover a revised source record. A new timestamp must trigger re-evaluation of affected outputs, while unaffected records remain traceable. A screenshot of a correct total is not enough to establish that change handling works.

Make disagreement an actionable output

A commercial reviewer may accept the operational interval while disputing the proposed treatment. Record those outcomes separately, with the evidence requested and the owner responsible for the next step. Do not force a single accepted/rejected status to represent both questions. If a later document resolves the disagreement, the revised assessment should identify what changed and which earlier assumption is no longer needed. That makes the review record useful to the next person who handles the case, rather than leaving them to infer the history from a polished final summary.

How Orisdale would scope the first engagement

Orisdale's proposed starting point is one operational review question, a bounded event population, and a contract owner willing to define the relevant treatment. The initial work would examine source quality, agree the interval and applicability rules, and identify the decisions that must remain with the business.

Potential deliverables include an interval contract, a mapped evidence register, a versioned treatment specification, synthetic and customer-approved acceptance cases, and a review view with explicit unresolved states. Whether the work proceeds to implementation depends on the usable evidence and agreed scope, not on the availability of a conversational interface.

The Energy blueprint describes related decision patterns, and the shared architecture shows where review stays with a person. For a practical discussion, bring one difficult event pair and the agreement that governs it to Orisdale. The first useful outcome is a defensible boundary between operational facts, approved treatment, and the judgment still required—not an automatically generated charge.