fix: preserve DeepSeek thinking events

feat(evals): paired lockstep runner, lossless subagent grading, deprecated-battery guard

feat(experiments): DeepSeek stage-1 compaction arms + prefix-preserving summarization

fix(chat): keep one thread on one provider, drop the unselectable Gemini

test(chat): cover effective_tool_names for the default DeepSeek model

fix(tests): join live text deltas without separators, test a selectable Anthropic model

fix(retrieval): store the BM25 index as gzipped JSON instead of pickle (#13)

Format DeepSeek default tests

Use direct DeepSeek as default model

Use OpenRouter DeepSeek with Gemini fallback

fix(pipeline): single-source runs rebuild all_sources_data.jsonl safely

fix(evals): never judge faithfulness on silently truncated evidence

fix(registry): retire sources from every registry dict, not just the first

fix(public-bundle): fail closed on non-allowlisted raw and wiki content

fix(pipeline): re-embed chunks whose content changed in place

fix(pipeline): stop incremental context runs gutting updated sources

fix(activity): repeated Gemini search queries start a fresh cycle

fix(graphrag): query-relevant community reports, rerank floor parity

fix(retrieval): budget skip, standard RRF, fail-fast collection load

fix(config): honor AI_TUTOR_KB_DIR for every KB path

fix(citations): keep labels of links with balanced brackets

perf(citations): cache the KB manifest indexes

fix(api): close request-size gaps in body and history caps

fix(api): emit stream error frame after cleanup parts

fix(citations): handle malformed links; count only consulted docs in activity (#12)

Public docs-only bundle: allowlist filtering, leak audit, kb.tar.gz cold start (#11)

wip: eval harness + public docs bundle; relocate eval docs into evals/ (#10)

fix(citations): drop code-block URLs from chips; default URL reading off

feat(prompts): add course content reproduction policy and related tests

build(graphrag): make graphrag an eval-only optional extra (keep prod image lean)

Merge experiment #2: GraphRAG vs classical RAG (opt-in retriever + scoped comparison)

feat(evals): single-turn latency finding (F27) + de-confound latency tooling

graphrag: validate retriever vs real index, fix bugs, scope by source

graphrag: scaffold true GraphRAG retriever arm (opt-in, default unchanged)

Add Part C eval variants, LLM judge, and harness fixes

Add evaluation harness for memory/context-management experiments

Enhance knowledge base and API functionality

Enhance API validation and error handling

Enhance model validation and tool call management

Enhance chat service and provider events handling

Enhance chat service and configuration management

Implement thread management and transcript handling in chat service

Renamed 'scripts' folder to 'app', removed mongo and instructor libraries

Refactor test cases for improved clarity and functionality

Update project structure

ruff format

Enhance error handling and logging in chat and retrieval processes

Remove SourcePanel component and related message source logic

Refactor update_kb_wiki to remove --changed-only option

add knowledge base wiki exploration tool