Purpose and contribution
Mosaic Together explored a small multiplayer idea: players rotate geometric tiles until they complete a shared picture. The development brief prioritized room-code joining, synchronized screens, replayability and browser access without player accounts or installation.
Michael Oakes supplied the brief, suggested the cooperative mosaic concept and evaluated the result. ChatGPT Work performed the documented implementation and testing work. The project demonstrates AI-assisted development and technical review; it is not presented as Michael’s sole-authored software.
Keeping a shared board consistent
The main engineering problem was handling overlapping actions. Two players might turn different tiles simultaneously, select the same tile, retry a request or send a delayed move after a new round begins.
The recovered source contains several complementary controls:
- Conditional database updates: a change is saved only if the stored room revision still matches the revision read. Conflicting writes reload state and retry within a bounded limit.
- Expected tile orientation: a move based on an outdated orientation receives a conflict rather than silently turning the tile again.
- Round identifiers: delayed moves from an earlier puzzle are rejected.
- Action identifiers: repeated actions are recognized within the latest 256 retained receipts per room.
These mechanisms are verified in the recovered implementation. They support a bounded consistency design, not an unlimited “exactly once” guarantee.
The interface uses short polling rather than continuous streaming. Its documented normal polling interval is approximately one second, excluding network and processing time; that is an implementation choice, not a measured latency promise.
What the retained evidence shows
The recovered repository exactly matches deployed version 3, commit a320279c2c97a52067424d0ec7299f95b4f12608.
Its October 26 verification report identifies version 1, commit c84382460a98a794fd8e4d20422d080e3d8bb064, as the tested application release. Comparing those revisions shows changes to nine documentation and evidence files. Application code and both test suites are unchanged.
| Evidence level | Supported finding |
|---|---|
| Verified implementation | Source contains revision checks, tile-conflict handling, stale-round rejection and bounded duplicate-action protection. |
| Verified test artifacts | The repository contains 13 rule-test definitions and an HTTP suite with 13 check groups. |
| Historical reported execution | The producer’s report records passing rule tests, local and public HTTP checks, two-browser synchronization, replay and refresh recovery. Raw execution logs were not recovered. |
| Preserved visual evidence | A retained image shows two named players, two total turns and two completed tiles. It is consistent with the report but does not independently prove synchronization between browsers. |
The report describes a development-origin failure involving crypto.randomUUID, followed by an implementation change. It also distinguishes source-reviewed recovery fixes from exercised browser behavior: no artificial production outage was injected to validate those recovery paths.
Technical progress did not settle product quality
The prototype was deployed, but Michael subsequently rejected Mosaic and directed work toward a separate maze project.
That decision remains part of the outcome. The available evidence supports an implemented prototype and a documented engineering process; it does not establish that the game met the owner’s expectations for an engaging portfolio project.
The maze is a separate artifact. Its implementation and tests are not evidence for Mosaic.
Sources and scope
This study draws on the original development brief, the recovered Sites repository, the game-rules and room-persistence modules, both retained test suites, the architecture and October 26 verification documents, the preserved gameplay image and Michael’s later rejection. Source inspection and revision comparison were completed November 6, 2026. The underlying private conversations and local review files are not public links.
The review did not recover the original downloadable source ZIP, the release receipt, raw historical test logs or browser traces. It did not rerun the tests.
No independent human playtest, physical-device matrix, load test, security audit, accessibility certification or contest-submission confirmation is established. The study makes no claims about adoption, reliability, time savings or commercial results.