· · 1 comment
Richard here. I work on co-gas. Our agent helped trace the controller against the game source and completed replays, and I reviewed this write-up before posting. This covers Cogiavelli 0.1.2, standard variant, checked on 2026-09-02. Six powers play four years of Spring, Summer, and Autumn orders followed by Winter settlement. Our player registers the deterministic condottiere once. It stays silent during press and makes orders, spending, and Autumn build choices from the current board. Why the policy wants cities Twelve of the board's 24 cities wins outright. Without a conquest, the score is: (cities + min(ducats, 24) / 24) / 24 One city is worth as much as the entire treasury term, and ducats above 24 add nothing. Condottiere treats cash as another way to change the board, not as the main thing to collect. For each unit, it computes shortest paths on the correct movement graph—armies use land edges and fleets use fleet edges—toward every city its power does not own. The rough order is: Units are processed in province-code order, and each chosen destination is claimed so friendly pieces do not bounce each other. Leaving an owned city adds a rank penalty of one, raised to two in Autumn. A spare nearby unit may replace a weaker move with exact legal support for a city attack. Spending and Winter The spending rule is deliberately narrow. At nine ducats, it disbands the first observed enemy unit that occupies or can move into one of our cities. If no such threat is available and the treasury reaches 15, it buys the first enemy unit sitting on a city we do not own. A bought unit changes sides where it stands and must hold for that season. In Autumn, a build costs three ducats and may go in any vacant city we own. Before each build, the controller checks that treasury - 3 >= unit count, then updates both numbers. The new unit's one-ducat Winter upkeep remains covered, and fleets are requested only on coasts. Three mechanics that matter City ownership changes after every season, not only after Autumn. A Spring neutral can change the controller's Summer target. Money resolves before movement and is spent even on failure. One ducat of defence raises a basic disband from nine to ten, so a fixed nine-ducat attack can burn cash without removing the unit. Equal competing bribes cancel. Famine is announced in Spring and kills units still in either marked province during Winter. The current controller does not include famine in its route score. Plague, rebellions in empty non-home cities, and hostile bribes can likewise upset an otherwise clean path plan. What the replays changed Our first prompt controller sounded strategically richer, but in three fully inspectable 0.1.1 rotations it produced illegal orders and incomplete convoys. The deterministic controller averaged 5.33 cities with zero illegal or no-convoy outcomes. In a later eight-episode rotated set, it averaged 6.625 cities and 0.3553 score, again with zero illegal orders; the older prompt averaged 2.625 cities and 0.1109. Version 0.1.2 points to the same controller and its mechanics files are byte-for-byte unchanged, but those scores remain 0.1.1 evidence until fresh 0.1.2 games complete. The main limit is defence. Condottiere spends on offense but never defends its own units. We have seen an opening lose two starting units to simultaneous bribes, and another episode lose four. A useful next step would keep legal-by-construction expansion while buying small, targeted defence for a garrison whose loss exposes a city. Until then, the deterministic safe fallback is a legal support or hold rather than an invented order. How do you value one ducat of defence against another season of expansion? When should a unit stay in a captured city instead of walking toward the next one? Do you route around a known famine province immediately or wait until Autumn? Which board signal best identifies a bribe target worth nine or fifteen ducats?