Implement Semantic Consolidator V0

- add deterministic canonicalization support for extraction items
- add facts-only semantic consolidation using local Ollama
- preserve source evidence and validate complete fact coverage
- add conservative merge rules and non-LLM tests
- record the first validated real-life consolidation benchmark
- document current scope, limitations and next evaluation step
This commit is contained in:
2026-07-31 11:25:46 +02:00
parent 6e34334506
commit 90aa34d5d0
20 changed files with 5517 additions and 93 deletions
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,24 @@
# Semantic Consolidator V0 Report
- Scope: facts only
- Model: `qwen3.5:9B`
- LLM call count: 1
- Runtime: 390.119 seconds
- Fact item count: 33
- Prompt characters: 15393
- Estimated prompt tokens: 3849
- Prompt eval count: 4295
- Eval count: 2344
- Merged fact groups: 1
- Source facts involved in merges: 2
- Singleton fact groups: 31
- Validation: passed
- Output path: `samples/benchmarks/semantic_consolidator_v0/consolidated_extractions.json`
## Actual Merges
### Der Leiter F&E führt die Projektliste auf dem zweiwöchentlichen Schnittstellen-Stand-Up.
- Source fact IDs: fact_0025, fact_0031
- Merge reason: Semantic equivalence: Both items state that the Head of R&D leads the project list at the bi-weekly interface stand-up meeting. The difference in wording ('auf den' vs 'auf dem') and minor elaboration on defining project types does not change the core factual proposition regarding who performs which action when.
- Review decision: accepted as correct.