How every number is produced

Company parameters

Traits are dimensionless scenario parameters bounded to [0,1], not empirical company scores. An initial reliability of 0.90 is a preference proxy, not a claim that 90% of real tasks succeed. Outcomes pricing is a fixed annualised bundle at the segment's assumed usage, so per-seat and per-task prices are only comparable after that normalisation.

Segments

Outside-option utility 0.20. Choice temperature 0.20. Prices may not fall below 50% of the role's original list price.

Decision catalog

Outcome-bundle conversion is allowed once per company and already applies to TaskPilot at the start, so that role cannot buy it. Research is the round's primary commitment, creating a real opportunity cost. Projects completing after Q4 remain in the ledger as committed work with no terminal benefit.

Resolution order

prepareRound(previousState, environment) performs the first two steps and freezes the result. Both sides then choose from that same frozen state. resolveRound(preparedState, playerAction, opponentAction) performs the rest and returns the next state plus an ordered event ledger.

1. Activate previously completed projects.
2. Reveal the scheduled public environment event, exactly once.
3. Accept both commitments simultaneously, judged against the start-of-decision information.
4. Deduct investment cash.
5. Apply immediate actions (price cut, research).
6. Recalculate traits and prices against their bounds.
7. Allocate customer transitions.
8. Recognise quarterly revenue and operating cash costs.
9. Mark negative ending cash as funding required.
10. Create the explanation records you see in the ledger.

Formulas and parameters

Each ledger line in the game links to one of these ids.

Environment presets

The quarter an event fires is deliberately not listed here. It becomes visible when the event happens, or a quarter early if you commission research.

The opponent

The P0 opponent is deliberately simple, and it is important that you know how simple. Each quarter it:

1. enumerates its legal actions in the catalog order above;
2. forecasts up to 2 quarters for each one, assuming you hold and no further public events occur;
3. picks the action with the highest forecast ending cash;
4. breaks ties on higher ending customer count, and then on catalog order.

The look-ahead is clipped to the quarters that remain. The opponent commits from the same information snapshot as you: it is handed a public snapshot with your research and the hidden event schedule stripped out, so it cannot read your unlocked action or see what is coming. It has no memory, no model of you, and no notion of long-term positioning. Beating it is not evidence that a strategy is good.

Rewind, branching and role reversal

Rewinding a quarter creates a new run from that pre-decision snapshot. Earlier decisions and the external event sequence are preserved, the original run is left untouched, and the new run records the original as its parent. The opponent policy is recomputed from the branched state rather than replaying its old moves.

Switching sides starts a fresh game from the original opening state and environment, with your complete recorded action sequence driving the other company. Those actions are replayed when legal. When one becomes illegal the script holds and states why; it is never silently upgraded to a stronger move. That session is labelled a fixed-opponent comparison. Outcomes are only comparable within the same scenario, and there is no universal strategy score.

Explanations

Every explanation in this build is generated from the engine's own event ledger by the rules on this page. There is no AI narration endpoint, no API key and no network call, so there is no narration failure that could interrupt play. Optional generated dialogue is a later phase; it would sit behind a server adapter, would never be able to write to game state, and the rule-based text here would remain the fallback.

Limitations

• Every company, price, cost, trait and event in this exercise is a fictional design assumption. None of it is a measurement of a real market or a prediction about a real company.
• Demand is fixed. There are 1,000 teams for all four quarters, so nobody can win by growing the market.
• Revenue is recognised as ending customers times annual price divided by four. Contracts, ramp-up, billing terms and partial-quarter recognition do not exist here.
• Customer choice is a softmax over four trait terms and one price term. Real buying behaviour has switching costs, procurement cycles, brand and channel effects that this model does not represent.
• Traits are single numbers on a [0,1] scale. A reliability of 0.90 is a preference weight, not a claim that 90% of tasks succeed.
• There is no financing, no hiring, no acquisitions, no product portfolio and no free-text strategy. One commitment per quarter, from a fixed catalog of eight.
• The opponent is a two-quarter greedy cash maximiser that assumes you hold. It is deliberately simple and is easy to out-think; beating it is not evidence of a good strategy.
• Four quarters is far too short for most of these investments to compound. Read the ordering of outcomes, not the magnitudes.
• No playtest data, usage statistics or user research informs any number here.