[morgaesis]

Hosted confidence needs a separate scorer rather than generator self-assessment. This enables Luna as both the generator and a separate scorer call, teaches admission validation to price a model shared across roles without allowing duplicate fallback entries, and keeps the qualification manifest empty so release authority still requires the exact-source managed benchmark and attestation.

[postil-dev]

2 blocking findings open
1 finding posted inline · 1 finding in review details

• Correct the stale scorer-chain claim in pull request description: The description incorrectly says hosted reviews use generator self-confidence with an empty scorer chain. Update it to reflect that Luna is now enabled as both the generator and scorer, while qualification still requires the managed benchmark and attestation.

_{Review details}

[postil-dev]

2 blocking findings open
1 finding posted inline

1 suppressed

• Correct the stale scorer-chain claim at .postil/pr-description:2: deterministically non-actionable; severity error, confidence 0.98. The description incorrectly says hosted reviews use generator self-confidence with an empty scorer chain. Update it to reflect that Luna is now enabled as both the generator and scorer, while qualification still requires the managed benchmark and attestation.

Before the next push

Run postil review --staged.

_{Review details}