What it is for, and what it is not
It answers one question: given your assumptions about strikes, repair and state response, what happens to the supply of AI-95+. It is not a prediction, and it will not choose assumptions for you. Two analysts who disagree can load both views and see whether they disagree about today or about the future, which is usually the whole quarrel.
It never lets you choose which refinery is hit next. Past damage is entered plant by plant; future damage is drawn statistically across plants by reach and size. That line is deliberate.
1. Set your own view of today
Open Plants. Every plant is one row: the state the Lab assumes, which units are up or down, how long a return should take, how many times it has been hit, and its strike history. Pick one of the three reconstructions of today, edit any plant, and send the edited plants to the Lab. You can also load a whole reconstruction from a CSV using the template under "Plants and constants" on the Scenario deck.
Your reconstruction is then tested twice: against the pump data actually observed today, and against whether the strike assumptions you set could have produced the state you started from. The result is the one line at the top of the Scenario deck; click it to see the detail.
2. The hazard controls
Strikes a week and the share of them that damages a unit are separate, because analysts disagree far more about the second than the first. Every event behind the defaults is listed in the strike log. Ballistic strikes are a separate control with their own target set and start date.
3. Repair, parts and response
Repair is expressed in weeks, not multipliers. Two things can hold a repair up: no engineers free, or no part. Parts arrive at a countrywide production rate, not on a per-order timer, so many simultaneous losses queue behind each other. Response levers cover imports and the Moscow supply commitment. Event cards carry dated decisions: a pause in strikes with an end date, a declared countrywide rationing step, rationing in the capital.
There is no countrywide rationing dial. Each region's step is derived from its own coverage, because a single countrywide step has never existed and would never be announced.
4. Reading the results
Three blocks answer different questions and can move in opposite directions. Station availability is what a driver sees, and it improves when demand is destroyed as well as when supply recovers. Production is what the system makes, against the four floors of demand that cannot be compressed. Is this structural separates crude capacity, which churns, from the ability to make high octane, which does not come back.
The green zone on each slider is what has been observed. Leaving it is allowed and asks you for a written justification, which travels with the scenario. Constants marked unsourced are judgments, not measurements.
5. Sharing and registering
By default you are exploring: nothing is required, permalinks work, exports are marked exploratory. When you have found something worth defending, register it as a claim by giving it a name, a thesis, a falsifier and a check date. Only then can it be cited and scored. The tool can draft all four from what you changed.
Every export carries the passport: the parameters you moved, the justifications, the evidence mix, and whether the scenario reproduces today. A chart without its passport is not a result.