Rubric and evidence map
This page connects the project brief to evidence in the repository. It is a navigation aid, not a claim that every criterion automatically receives full marks. Live behaviour, user feedback and performance claims need dated evidence from the released build.
Course-wide requirements
| Requirement | Implementation and documentation | Evidence still worth attaching |
|---|---|---|
| Git-compliant version control | Gitea repositories, GitHub Flow, feature branches and pull requests. See Git methodology. | Pull request list, contribution graph and examples of review before merge. |
| Responsive and accessible interface | Responsive website layout, semantic form controls, 42 ARIA uses, live status regions and a shared visual language. See Accessibility and responsiveness. | Preserve mobile/wide screenshots, keyboard-only run, 200% zoom test and an automated scan from the submitted deployment. |
| CI/CD | Gitea Actions quality workflow plus Vercel and Cloudflare deployments. See CI pipeline and Deployment. | Green workflow and deployment links for the submitted commit. |
| Non-monolithic front and back ends | Browser portal, Godot simulator, Python SDK and Supabase are separate boundaries. See Architecture. | Architecture diagram linked to deployed components. |
| Hand-written API | Researcher-facing Python SDK over the team's headless Godot protocol. See Python SDK and Internal API. | Package install and end-to-end notebook demonstration. |
| Authentication and security | Supabase Auth supports sign-up, sign-in, password recovery, optional 2FA and self-service account deletion; Postgres RLS enforces project access. | Apply the account-deletion migration, confirm production redirect URLs and retain role-by-role security results. |
| External API | Open-Meteo environment import. See External API integration. | Live success and graceful-failure demonstrations with attribution. |
| Public documentation | Docusaurus site hosted on Cloudflare Pages. | Public URL and successful production build for this revision. |
Sprint 3 rubric
| Criterion | Current evidence |
|---|---|
| User feedback | User feedback records four participants, requested changes and incorporated outcomes. Another dated round against the final model workflow is still recommended. |
| Automated testing | Python backend and SDK tests, Godot headless tests, model/training tests, UI contract tests, Ruff, gdlint and an 85% Python coverage gate. See Automated testing. |
| Feature implementation | Project workflow, custom arenas, stimuli, treatments, recording, replay, headless SDK, training, saved models and live ONNX execution are integrated. See Current release. |
| API implementation | Python SDK 0.4.4 is the public interface; internal web and Godot bridges remain implementation details. |
| Performance | Training notes report roughly 1,000 to 1,200 simulation steps per second on a two-core Colab-like machine. The team should preserve the benchmark command, hardware and raw output. |
| Improvement | UI restructuring came from user/client feedback; headless model training and live execution came from the clarified platform direction. |
| Documentation | Feature, API, database, testing, deployment, user and methodology sections are separated in this site. |
| Project methodology | Iterative GitHub Flow and layered testing are documented. Attach sprint-board and pull-request evidence. |
Final submission rubric
| Category and weight | Documentation and evidence | Honest remaining check |
|---|---|---|
| Database: data 3% | Database explains research and test records. | Record how much production data is demonstration/test data. |
| Database: deployment 2% | Supabase is hosted; Deployment explains the nine ordered migrations. | Export the applied migration list from production and confirm the account-deletion migration. |
| Database: structure 5% | Relational schema, JSON experiment configuration, indexes, constraints, storage and RLS are documented. | Resolve and record any live-schema drift. |
| API: availability 3% | SDK can be installed externally and works in Colab; internal and external APIs are documented separately. | Demonstrate a fresh install and hosted save. |
| API: architecture 5% | Public SDK, internal simulator protocol and hosted database boundaries are explicit. | Freeze compatibility policy for later SDK versions. |
| API: deployment 2% | Wheel and versioned simulation pack form the distribution. | Attach published package and checksum evidence. |
| API: performance 5% | A repeatable concurrent load-smoke tool records first response, throughput, percentiles and failures locally and in CI. See Performance testing. | Keep the submitted CI JSON artifact and distinguish local smoke results from production capacity. |
| API: design 10% | Python objects represent projects, experiments, runs and models; private HTTP/database details are not the public interface. | Document package deprecation/versioning rules. |
| App: accessibility 5% | Semantic controls, text status, language, dialog relationships, keyboard intent and palette are documented. | Preserve the final keyboard, zoom, screen-reader and automated-scan evidence; no unsupported zero-violation claim is made. |
| App: aesthetics 5% | Inter font, dark navy palette and shared web/Godot components are documented in Visual language. | Capture final wide and narrow views. |
| App: user experience 5% | Researcher workflow, understandable errors, popup project creation and persistent authentication are documented. | Time representative first-user tasks on the final build. |
| App: deployment 2% | Vercel production URL is public and Supabase/Cloudflare are independently deployed. | Record final deployment IDs against the submitted commits. |
| App: performance 5% | Compressed Godot export, lazy simulator loading and headless paths reduce load and training cost. The team has a release measurement procedure. | Preserve cold load, simulator-ready, control-latency and headless-rate measurements with hardware and cache state. |
| App: features 10% | Current release and user guides map the implemented product. | Demonstrate the client-priority journey end to end. |
| App: responsiveness 5% | Layout rules cover desktop, focused simulation, 390 px and 360 px targets. | Attach final screenshots and verify no horizontal overflow at the supported widths. |
| App: structure 5% | Navigation is grouped into researcher guide, architecture, APIs, interface, testing and development process. | Conduct one final first-time-user navigation test. |
| Misc: Git methodology 5% | Git methodology explains branch and merge flow. | Attach branch/PR examples, including integration fixes. |
| Misc: integration 3% | Open-Meteo is reviewed before application and stored with the experiment. | Demonstrate unavailable-service handling. |
| Misc: testing 10% | Layered tests, an 85% Python line-coverage gate and explicit JavaScript/GDScript coverage limits are documented. | Publish the exact final report and workflow link. |
| Misc: tools 5% | Ruff and gdlint are enforced in CI; Trello, Gitea pull requests, Actions and issue tracking are described in Work and bug tracking. | Export or screenshot the final tracker and issue history. |
Evidence rule
Use links, command output, screenshots and exported reports from the exact submitted revision. A source file that contains a feature is implementation evidence; it is not by itself proof that the feature is deployed, usable or secure.
AI Attribution: This mapping was prepared with OpenAI Codex assistance from the supplied COMS3011A project brief and the merged repositories. The team should replace general statements with its final dated evidence before submission.