-
Notifications
You must be signed in to change notification settings - Fork 0
Conformance
GhostFrame edited this page May 25, 2026
·
2 revisions
Conformance testing provides a quality gate for persona upgrades. A persona pack can declare a minimum test score; newer versions must meet or exceed that score to install.
In pack.toml:
[conformance_baseline]
score = 0.92
bundle_hash = "sha256:..."-
score-- Minimum acceptable test score (0.0 to 1.0). -
bundle_hash-- SHA-256 hash of the test bundle that produced the baseline.
A test bundle is a set of test cases that exercise the persona's behavioral properties:
- Input: A task description or scenario.
- Expected behavior: What the persona should do (or not do) in response.
- Timeout: Maximum time for evaluation.
The frameshift-conformance crate provides the schema, a runner trait, and scoring logic.
When upgrading a persona, the regression gate:
- Runs the test bundle against the new version.
- Computes a score.
- Compares against the declared baseline.
- Blocks the upgrade if the score drops below the floor.
This prevents persona regressions -- a newer version that lost behavioral properties won't silently replace a working one.
frameshift verify <persona>Runs conformance checks against an installed persona and reports the score.
Frameshift -- Same model. Different frame.
Getting started
Using Frameshift
Personas
Systems