All Wells Shut. Reservoir Still Busy? We Need to Talk.
CLARISSA · · Road to ATCE
The wells are shut. The production schedule has not started. The reservoir model is supposed to be resting. Yet the pressures are moving and the fluids seem to have arranged their own activities.
Before anyone admires the forecast, that deserves a conversation.
One useful part of CLARISSA’s workflow is a test whose ideal result is gloriously uneventful. In the architecture described by our final ATCE paper, the model runs a synthetic settling period before scheduled activity. For a starting state intended to represent equilibrium, the question is simple: does it stay there?
A quiet starting point earns its keep
Section 5.1 of the paper describes this as the third deterministic validation gate. All wells are shut during the settling period. The stated acceptance criteria are flat pressures, static phase volumes, and no flows between regions. The workflow also checks initialized in-place volumes against the initialization report.
Why spend simulator time asking for nothing to happen? Because an input file can be readable by the simulator while its initial state still needs attention. If an intended equilibrium state drifts during this test, the simulation itself has exposed a mismatch between that intention and the initialized model.
That gives the engineer something concrete to investigate before adding the effects of production or injection. A moving curve is much easier to discuss when the model has not yet been asked to do anything.
The parser has a different job
The CLARISSA architecture uses several checks because each answers a different question. The first gate uses the simulator’s input stack to parse the deck. The second reconciles cumulative volumes from source production data with the emitted schedule. Equilibration follows; analytical plausibility checks come after execution.
These checks complement one another. A parser checks whether the input can be interpreted. It cannot, merely by accepting the file, demonstrate that the requested starting state remains stable when the simulator advances it.
The strength here is the sequence: turn an engineering expectation into an executable check, then inspect the result. A fluent explanation is useful. A test tied to the intended physical state gives that explanation something to stand on.
Doing nothing is a test, not a universal law
The qualification matters: this check concerns a model intended to start in equilibrium under the test conditions. It is not a claim that every real reservoir with shut wells must be static. A deliberately transient starting state needs acceptance criteria appropriate to that problem.
Nor does passing this gate establish that a forecast is right. The paper explicitly distinguishes a well-formed, self-consistent model from one fit for a particular decision. Its demonstrations use OPM Flow at reference-problem and single-asset scale; the engineer remains responsible for fitness assessment.
Capable frontier models can already tackle simulation workflows. CLARISSA’s engineering checks remain useful regardless of how persuasive the underlying model becomes. Evaluation through RIGOR is another part of making results examinable; the paper does not establish superiority over Opus 5.5.
A useful pause on the Road to ATCE
Our October 3 roadmap sets aside October 10–11 for consolidation, bug repair, and planning. That is a planned work period, not evidence that every task has been completed. For anyone following the ATCE presentation, equilibration is a concrete example of the engineering discipline behind the broader story.
A useful question to bring to a demonstration is: what did the model do before the operating schedule began, and what evidence shows that the starting state was acceptable? It is a small question with a very worthwhile answer.
That fits our vision for subsurface AI: make it easier to ask engineering questions and easier to examine the answers.
Sometimes the most reassuring thing a reservoir model can do is absolutely nothing. We should give it the chance.
Explore CLARISSA and request a demonstration →
Sources: final manuscript SPE-234136-MS, A Conversational User Interface For Autonomous Reservoir Simulation Deck Generation And Execution, Sections 5.1 and 7; Road to ATCE plan dated October 3, 2026. The validation workflow is described by the paper; this article does not report an independent replication.