v1.75.0

New Features

• d61e7a7 - gooddata-eval: extend detail.latency_breakdown to 5 remaining kinds (commit by @Tomkess)
• 79a3d1d - gooddata-eval: register knowledge_question evaluator (commit by @Tomkess)
• 6b9d690 - gooddata-eval: add Langfuse OTLP/experiment foundations and LANGFUSE_BASE_URL (commit by @tychtjan)
• 89d8a89 - gooddata-eval: read Langfuse traces via v2 observations and move the client to core/langfuse (commit by @tychtjan)
• b12210b - gooddata-eval: link agentic runs as Langfuse experiments via OTLP root spans (commit by @tychtjan)
• b47cf1a - gooddata-eval: pass question, conversation and output into experiment spans from every agentic kind (commit by @tychtjan)
• 6488fe0 - gooddata-eval: write single-shot results as Langfuse experiments over OTLP (commit by @tychtjan)
• 334faff - gooddata-eval: make the K verdict a gate, pass@K or pass^K (commit by @myhoai)
• 1c1ae20 - gooddata-api-client: add the CREATE_METRIC workspace permission (commit by @Vojtasii)
• fa82a6a - gooddata-eval: record the turns and steps each agentic run took (commit by @myhoai)

Bug Fixes

• 9aa8d86 - gooddata-eval: return a copy of activated_skills from TurnResult.detail() (commit by @Tomkess)
• 7f04d92 - address review findings on judge faults, run counts and trace linking (commit by @tychtjan)
• 3f9b3f8 - gooddata-eval: persist skill_routing credit across conversation turns (commit by @Tomkess)
• 26d4f52 - gooddata-eval: track set_skills as replacing, not accumulating (commit by @Tomkess)
• 2d64613 - gooddata-eval: keep only the final set_skills declaration per turn (commit by @Tomkess)
• 3e1d84d - gooddata-eval: count skills replaced within a turn toward coverage (commit by @Tomkess)
• 3bd2a29 - gooddata-eval: assert alert group-by attributes (commit by @tychtjan)
• 57aa25f - gooddata-eval: three false-negative sources in eval scoring (commit by @myhoai)
• f5703ce - gooddata-eval: accept a string-encoded rejectedSpans count in OTLP export responses (commit by @tychtjan)
• 18073f8 - gooddata-eval: fall back to the default score retry delay for negative or NaN Retry-After (commit by @tychtjan)
• a7ca20b - gooddata-eval: restore the dataset_run_items compat shim and harden the Langfuse score path (commit by @tychtjan)
• 76470ba - gooddata-eval: keep gate out of the positional signature and in the merged report (commit by @myhoai)
• 454286c - gooddata-eval: report the gate verdict without redefining pass@K (commit by @myhoai)

Performance Improvements

• 1f9babd - take Langfuse trace linking off the eval item critical path (commit by @tychtjan)
• 8125426 - cut gooddata_sdk import cost by importing only what the SDK uses (commit by @hkad98)

Refactors

• 0aa2b77 - gooddata-eval: derive TurnResult.detail() from the model, not a key list (commit by @Tomkess)
• 55e22cd - gooddata-api-client: split LDM object permissions per family (commit by @jeskepetr)
• dc79848 - gooddata-eval: remove the legacy Langfuse ingestion and dataset-run code paths (commit by @tychtjan)
• 99935f9 - gooddata-eval: count turns and steps as int, not float (commit by @myhoai)

Tests

• 316c88a - gooddata-eval: use the real set_skills argument key in conversation tests (commit by @Tomkess)
• bbe1187 - gooddata-eval: add an in-process fake Langfuse server for wire-level tests (commit by @tychtjan)
• 7d8155e - gooddata-eval: cover the Langfuse v4 request choreography against a fake server (commit by @tychtjan)

Chores

• 279dc54 - repo: replace AIDA MCP with tracked AGENTS.md and CLAUDE.md (commit by @tychtjan)