# Deployment Run

An original, local FDE decision game within the [field guide](../index.html#game). The three fictional customer engagements are Northstar Services (dispatch), Cedar Supply (invoice matching), and Harbor Engineering (internal knowledge). Each has five decisions: discovery, integration, evaluation, incident recovery, and handoff.

## Learning model

The game applies the reviewed [field workbook](field-workbook.md) and [paired readings](reading-companion.md): observe the workflow, bound authority, test final outcomes, reconcile uncertain operations, preserve source governance, and demonstrate customer ownership. It does not reproduce the Benmore PDF or Agent Factory text. Names, messages, data, timing, and outcomes are original fictional examples. The [source review](material-review.md) describes the coverage and limits of the underlying research.

The player inspects evidence and sees time costs before committing. Feedback appears after commitment. Integration choices affect the later incident; omitted failure tests make recovery take longer. A controlled manual fallback or smaller supported scope can win. No action connects to a customer system, sends a message, or calls a model.

## Delivery gates and scoring

Each attempt starts with 50 trust, 50 reliability, and 14 simulated days. There is no real-time timer. A delivered mission must finish all five decisions with:

- Customer trust of at least 60.
- System reliability of at least 70.
- No days beyond the agreed delivery window.
- Verified recovery of the incident.
- An operating owner and backup who rehearsed the handoff.

Meeting all gates earns one star. Two stars require trust at least 70 and reliability at least 80; three require trust at least 80 and reliability at least 85. An incomplete delivery earns no stars and identifies the gates to revisit. XP is 40 plus 100 per star, one fifth of the combined trust and reliability rounded, and two per unused simulated day. Only a mission's best result contributes to campaign XP. Higher star tiers always have higher XP than lower tiers in these missions. Repeating a result cannot accumulate points.

These are transparent teaching rules, not forecasts of production performance, professional certification, or hiring readiness. The game rewards judgment within a deliberately bounded scenario. The linked project exercises develop the actual engineering evidence.

## State and accessibility

Attempts, completed decisions, best results, and the selected mission use the existing browser-local guide storage. All missions are available immediately; switching preserves each attempt. A reload restores the exact decision or consequence screen. Restarting an active attempt requires an inline confirmation and keeps the best result. The guide's existing clear-progress control also clears game progress. Blocked storage leaves a usable session and displays a notice.

Choices use native labeled radio buttons and a form. Missions and actions are keyboard-operable buttons; consequences move focus to their heading and announce the result. Status uses labels and numbers in addition to color. The game respects reduced motion, includes system/light/dark themes, and adapts to narrow viewports. Reading links preserve the game attempt; the shared top bar returns to the game.

## Implementation checks

Run `node --test scripts/game.test.mjs` to check all 729 complete decision paths, required delivery gates, delayed consequences, saved-state recovery, repeated-action protection, scoring monotonicity, and learning destinations. Build the complete static site with `node scripts/build.mjs`. No external service is required to play.