AI edits can be ambitious, so recovery cannot be an afterthought. Oryn keeps project versions and validates playable exports after meaningful changes.
Ambition needs an undo
There is a failure mode where a creator stops asking for the change they actually want because the last big change broke something. The tool has trained them to be timid, and timid iteration produces mediocre games.
The fix is not a more cautious AI. It is a system where the worst case of an ambitious request is a restore, so asking for the risky version is rational.
Two mechanisms
Versions are snapshots you can return to
Restoring replaces the current files with an earlier state, and the versions you moved away from stay in history. Going back is not destructive to the thing you went back from.
Builds are verified, not assumed
A change that produces a project which no longer exports is caught by building it, not by reading the diff. The playable export is the test suite.
What this changes about how you work
The goal is simple: try a direction, see the real result, and return to a stable version when the experiment is not worth keeping. Once that is genuinely cheap, the good strategy stops being careful incrementalism and starts being a series of decisive experiments — which is how games get designed anyway.