A refinery LP that shows its work

Blending indices

Index blending for RVP, viscosity, freeze, pour and flash point; Ethyl RT-70 interaction blending for octane; cetane by the ASTM D976 density interaction or, as an option, ASTM D4737 Procedure A, which adds each blend's T10, T50 and T90; smoke point by a reciprocal index; weight-basis sulphur through ratio specs.

Marginal values with provenance

Every binding limit shows what one more unit is worth and the LP row, column and model input behind it. Price and limit ranging, margin sweeps and a tornado chart show how far each answer holds.

Crude assays

42 public ExxonMobil assays are bundled, with attribution. BP, Equinor and TotalEnergies assay files can be imported, and so can a manual CSV. Estimate a full assay from API gravity and sulphur (plus TBP points if you have them), or define a crude as a blend of assays. Changing the cut scheme re-cuts every crude, and swing cuts let the LP choose where a cut point falls.

Unit templates

Add process units from a template library and connect them on the flowsheet. Template numbers are marked as estimates from typical published values until you replace them with plant or licensor data.

Multi-period and multi-site

Periods with their own prices, turnarounds and tank inventory; several sites with transfers, crude routes and shared markets, all in one LP.

Integer decisions

Units that can be switched off with a minimum rate and fixed cost, one mode at a time, build-or-not decisions and crude bought in whole cargoes, solved with the gap to the best plan reported. When the plan without the integer rules already obeys them, it is found first; otherwise the independent parts of the search run side by side on several cores, with the same answer as one by one. Prove optimum and the background bound cover integer decisions too.

A schedule from the plan

Blend and crude schedules for the days ahead from a solved plan, in hourly slots: batches that meet every spec, several tanks per component and product with certification holds, and tank qualities tracked slot by slot. Crude and blend are reconciled by a bounded feedback loop; crude tanks can hold a mix of crudes and unit hold-up delays a crude switch downstream. Ships go berth by berth within draft and tide windows, with demurrage; pipelines track line fill and interfaces; tanks transfer between each other. Closing stocks can be published as the next period’s opening stocks, as a reviewed change you can undo.

Print-ready reports from the solve

A catalogue of seventeen reports, from the monthly operating plan and annual budget to crude valuation, turnaround impact, plan against actual and the model book, each catalogue cover drawn from the report’s own signature chart. Cover, contents, running header and footer, appendices with the provenance of every number; print to PDF, a self-contained HTML file or Excel, and bind monthly packs. Crude differentials against a marker crude, the budget down to EBITDA and cash by site with escalation (depreciation and tax indicative, not statutory accounts), and business-interruption loss with a deductible and ramp-down and start-up days.

Bring your refinery

A guided set-up from your planning model’s documents: PDF, Excel and CSV tables, and scanned PDF pages read by OCR on your machine (in the installed or hosted app, not the single-file build), with unsure cells marked for checking. A column mapper turns tables into capacities, yields, specs and prices, then every number is compared side by side with your current tool’s answer for a month. No PIMS file is read.

Energy transition

Carbon price and free allocation in the objective, bio co-processing, HEFA/SAF, electrolyser and blue hydrogen, CCS, renewable-fuel mandates and credits, and Scope 1, 2 and 3 with carbon intensity per product.

Scenarios and stochastic plans

Two-stage stochastic plans with the value of the stochastic solution and of perfect information, min-max-regret robust plans, and cargo valuation bands across scenarios.

Calibration and back-casting

Import historian exports (PI, IP.21 or any CSV/XLSX), reconcile plant and lab data, fit base yields and delta-base vectors from plant test runs or a simulator case study with confidence intervals, flag drift, and explain the gap between plan and actual.

Investment economics

Justified capex for any capacity or spec, capacity value curves and a ranked investment screen. Project cases are valued against the base case at a price deck by year, with each scenario on its own deck if you give it one and units coming into service in a later year, and with NPV, IRR and payback through an exchange file for inferTIC cost estimates.

Versions, merge and review

Named versions in the model file, a model diff between any two versions or files, three-way merge and a propose, review and approve workflow, all offline. An optional team server on your own infrastructure adds OpenID Connect sign-in that renews itself, roles, check-out, a switch that sends every change through review, named versions shared through the server, and an append-only change and audit log.

BI export and APIs

Plan results as a documented star schema in Parquet, CSV or JSON, written offline, with Power BI and Excel Power Query scripts and a ready-made Power BI report template (written as text, so it needs a first open and save in Power BI Desktop). The team server serves the same tables over REST with an OpenAPI document.

AI that proposes, never decides

An optional assistant helps build and edit the model. It proposes changes as validated patches shown as a diff; nothing applies until you accept. It never writes the LP, the solver input or result numbers. Use a local model, your own API key, or none.

Offline, no server

Runs in the browser, as an installable offline web app, as a single offline HTML file, or as a desktop app. Core features need no server; your model file holds its full change history. On the desktop, native solves run side by side, one per core. Light and dark themes, on this site and in the app.

• Prove optimum gives a bound on how far a plan is from the best possible; on large models it may not close within its time limit.
• Scheduling reconciles crude and blend with a bounded feedback loop, not a joint optimum; its quality tracking is a repair, not a proof of optimality; line fill is timed after the solve and units keep the plan’s yields through the week.

• Products are on a volume basis; weight properties go through ratio specs.
• Ethyl octane and D4737 cetane blending are not yet in Prove optimum’s search.
• A side-by-side comparison against a refiner's production model is planned with our first design partner.