# Thiawyn — the story behind the game

Speaker script, 13 September 2026. Approximately 15–20 minutes.

## 1. From forgotten code to Google Play.

Start with the title slide. This is the story of a specific collaboration, not a promise that every game can be built this quickly. Allow around 15–20 minutes. F enters fullscreen; N shows these notes. Audio starts only when you click a sound button.

## 2. An idea waiting since 2015.

The first Git commit preserves the prototype from 2015. These are its original image assets, not a reconstructed screenshot of the old game. The neighbouring Jumble and JustRoids projects provided examples of how similar projects had been modernised and completed.

## 3. Before the magic: make it work.

The original used compileSdk and targetSdk 21; the published version uses 36. Modernisation also covered pausing, resuming, and handling recreated Android activities. Git provided local version history; that does not mean we set up an external backup or a Git server.

## 4. Your ship fires itself. Positioning is up to you.

The core game idea came from Anders. Automatic fire removes the need for a fire button, but not the need for skill. The camera rotates to create the feeling of continuous flight; physics uses a separate world coordinate system. Strong tilts received an extra speed boost following hands-on feedback.

## 5. Save a world from its own defences.

Kestrel and Warden started as working names in the campaign proposal. The theme is liberation rather than conquest. The ending opens a shell instead of blowing up a planet. We show the ending later, so this presentation contains spoilers.

## 6. “I’m taking my phone and going to bed now.”

This is a translated excerpt from Anders’s instruction. It was not a single image prompt: the agent worked with project files, build tools, and tests. The overnight result was explicitly a proposal for review, not automatically a finished store release.

## 7. Six celestial bodies. Eighteen attack waves.

Three waves per world gave the game a clear rhythm and kept the first campaign manageable. Each chapter introduces new behaviour and advances the story. The eight final locks match the eight visible sections of the shell.

## 8. Copper, midnight blue, and turquoise light.

The first campaign used twelve generated bitmaps. The menu panorama and title logo came later: fourteen game illustrations in total, plus separate store artwork. Prompts and original files were preserved. Three enemy silhouettes serve several roles; colour, size, shields, and behaviour provide the remaining distinctions. The profile picture is a real photograph, not AI-generated.

## 9. The music was already mine. The effects were made in code.

Try playing the menu track followed by the planet-clear fanfare, rather than every example. The music is Anders’s own, released as Brainlords D.E. The AI did not compose it. Selection was based on album notes and technical checks, not subjective listening. Eleven deterministic synthetic effects were generated by a script. The three music examples here are 22-second excerpts with a fade-out.

## 10. “I could finish it without even changing direction.”

This is a shortened, translated excerpt from the feedback. The key point for colleagues: working code is not the same as a good game. The first automated campaign run used an invulnerable pilot, so it could not establish whether the difficulty was appropriate.

## 11. Enemies that predict where you are going.

For example, the balance probe at a constant +6 produced W5 across all 18 waves before the change. Afterwards it clears five waves and then loses. The evasive bot used real hull health but exact projectile data. This shows that a known trivial strategy is countered and the campaign remains possible for the bot, not that human difficulty is proven. Twelve Java tests also covered range, hits, shields, and eight final locks.

## 12. The player should feel that something has happened.

These details came from specific requests after installing the game on the phone. Another small but important issue: the text said three locks while the artwork showed eight. Both the story and final battle were changed to eight. The growing planet is a visual cue, not a change to the physics or weapon range.

## 13. “Wow! I have to start playing this!”

The translated quote describes the feeling Anders wanted. The menu took several iterations, not one successful generation. STAR CHART and MUSIC & CREDITS became THE SIGNAL PATH and CREW & FREQUENCIES so the buttons matched the headings on their destination screens.

## 14. Three kinds of evidence.

Final release verification had twelve passing Java tests and no lint errors, but 34 lint warnings. Earlier instrumented Android runs tested the UI and victory flow. These were not repeated after every small text change. Full TalkBack navigation and wider device coverage were known limitations. Explain what the tests actually establish.

## 15. Google Play was a project of its own.

This required both automation and human actions. Google’s documentation said the first upload must be manual, but this project’s upload worked through the API once the invitation was sent. Another API error involved a review parameter; we adjusted the request. No keys, account credentials, or private files are shown in this presentation.

## 16. “In review” does not mean live.

First open testing, then production. The production submission replaced the ongoing review. On 9 September, Anders wondered whether the process was stuck: the API still showed it under review. On the 10th, published production status and a public store page with an Install button were confirmed. These are historical checks, not a live status widget. We also learned to distinguish track status completed from the actual release lifecycle.

## 17. An old idea became a released game.

The ending shows a liberated world. Publication was verified on 10 September 2026. We do not report exact time savings or costs because we did not measure them. What we can show is the output and the Git history.

## 18. AI increased the pace. Judgement came from collaboration.

Avoid concluding that technical expertise is never needed. The AI also made assumptions that had to be corrected and could not replace the user’s experience. Tool access made a difference: the model could work in the project, build, test, and read actual status. Clear boundaries and traceability made the results possible to assess.

## 19. Which forgotten idea would you bring back?

Discussion questions: Which part would be most useful in our work? What tests would catch our own most common mistakes? Which decisions should always stay with people? What tools and access would be needed? The Sources button provides the supporting material. This retrospective presentation was created on 13 September 2026.