Computing Technologies student building software, web systems and a playable game.

The difficult part is keeping campaign state consistent while moving between one authored area and three maps assembled by a shared runtime builder. The Trace Receiver is deliberately used by several systems: it reveals recordings, creates enemy shield windows and carries the central story mechanic.

The project also includes region-aware lighting, a persistent day/night cycle, generated normal maps, adaptive procedural audio and an automated GitHub Actions path that imports the project, runs the campaign smoke test and exports the browser build.

The campaign is playable, but pacing and balance still need external playtesting. Combat action frames have less variation than the walk cycle, controller support is unfinished, and the browser build still depends on WebGL 2 and local browser storage.

A concept graph records prerequisites, every scored prompt is checked against a vendored FLELex vocabulary source, and deterministic tests keep completion separate from actual evidence of learning. The release checks cover linting, types, unit tests, production build, Playwright and dependency audit.

The wider product includes listening and speaking practice, mistake-driven spaced review, streak and garden rewards, private friend-code co-op challenges and deterministic roleplay with a guarded tutor fallback.

Production account sync stays unavailable until email delivery is verified and the privacy, consent, retention and youth-safeguard checks are complete. The public demo is intentionally the safer browser-only route.

The recommendation rules are deterministic and tested. They prioritise due review, remediation, Guided SQL, ready checkpoints, portfolio milestones and then the next genuinely taught topic. A real in-browser SQLite engine runs the fifteen-stage SQL route and the free SQL lab.

The app also has checkpoint gating, mistake recovery, a Guidebook derived from evidence, migration paths for older progress data and responsive checks across fixed viewports plus a continuous 320-1920 pixel sweep.

Nineteen mapped topics still need full lessons. Python and one API exercise remain clearly labelled simulations, and progress stays in one browser unless it is exported manually.

The radio stream and now-playing source are HTTP while the public site is HTTPS. Browsers block that combination, so two Cloudflare Workers re-serve the feeds. Netlify Functions and Blobs support session authentication, CSRF protection, owner editing, play history, listener feedback and first-party analytics.

The site depends on the station stream and metadata source staying available. A build check now recomputes every inline-script CSP hash so a design edit cannot silently break saved theme, contrast or text-size preferences again.

A static frontend talks to Netlify Functions, with users and conversations stored in Netlify Blobs. The retrieval layer splits context by audience before it reaches the model. Private source material is not committed or included in the static deployment, and there is no route that reads the protected pack back out.

This remains an experiment rather than a finished messaging product. The project is deliberately explicit about AI use, tells visitors not to submit sensitive information and keeps the privacy boundary in retrieval code rather than relying on a prompt.

The main work was setting boundaries clearly. Prices are visible, the site says what is and is not included, and the enquiry flow asks for a public link rather than credentials. The design is intentionally plain, direct and usable on a phone.

It is a focused service site rather than a large application. Its value is the clarity of the offer, the mobile contact path and the refusal to collect information the job does not need.