Files
Benjamin Admin 24499a25e9
CI / detect-changes (pull_request) Successful in 5s
CI / branch-name (pull_request) Successful in 1s
CI / guardrail-integrity (pull_request) Successful in 4s
CI / secret-scan (pull_request) Successful in 7s
CI / dep-audit (pull_request) Failing after 53s
CI / sbom-scan (pull_request) Failing after 51s
CI / build-sha-integrity (pull_request) Successful in 5s
CI / validate-canonical-controls (pull_request) Successful in 3s
CI / loc-budget (pull_request) Successful in 15s
CI / go-lint (pull_request) Successful in 53s
CI / python-lint (pull_request) Failing after 15s
CI / nodejs-lint (pull_request) Failing after 1m13s
CI / nodejs-build (pull_request) Successful in 3m8s
CI / test-go (pull_request) Successful in 58s
CI / iace-gt-coverage (pull_request) Successful in 15s
CI / test-python-backend (pull_request) Successful in 27s
CI / test-python-document-crawler (pull_request) Successful in 19s
CI / test-python-dsms-gateway (pull_request) Successful in 13s
fix(ai-sdk): make interpretation-intent override reliably win
PR #34's fixed +0.25 guidance gain was too small live: for "Was empfiehlt der
EDPB zum DSB?" the binding Art. 37 (1.381) still edged out the boosted EDPB
guidance (1.348), because the live authority score gives the binding article a
topic/domain bonus the (partly English) guidance chunk does not match.

Replace the fixed gain with a deterministic lift: a semantically competitive
guideline (raw semantic >= best_binding_semantic - 0.05) is lifted just ABOVE
the best binding FINAL score (ordered by semantic), so authority/topic/domain
bonuses can no longer edge it out. Obligation questions (no intent signal) are
untouched — binding stays Top-1; off-topic guidance stays demoted.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-06-24 11:21:59 +02:00
..