How mohammadmassaf/groundedforge works
Grounded Forge ingests user corpora, retrieves cited evidence, generates study or job-mode artifacts, and applies an independent cite-or-strike Critic before rendering results. Its README also describes evaluation of retrieval recall, grounding, and adversarial traps. The excerpts show ingestion dispatch, vector and keyword retrieval, and critique-loop behavior; several other subsystem links are documented in the README or indicated by the source index, so unsampled internals are represented at subsystem level.
Main entry points: main.py CLI commands · loop.py Critic-refiner loop · query.py Vector search · pipeline.py Ingest orchestration
Components 23
Corpus ingestion
- Corpus configuration
ingest/corpus_config.py - Ingest orchestration
ingest/pipeline.py - Source adapters
ingest/adapters - Chunk preparation
ingest/chunker.py - Chunk corpus
chunks
Evidence retrieval
- Vector index
retrieve/store.py - Vector search
retrieve/query.py - BM25 search
retrieve/keyword.py - Hybrid ranking
retrieve/hybrid.py - STAR evidence pools
retrieve/star_evidence.py
Artifact generation
- Artifact generator
generate/generator.py - Output schemas
generate/schema.py - Artifact renderer
generate/renderer.py
Verification and evaluation
- Critic-refiner loop
critic/loop.py - Independent Critic
critic/critic.py - Quantity guard
critic/quant.py - Run trace
critic/trace.py - Evaluation harness
eval/run_eval.py - Strike analysis
eval/strikes.py
Other components
- Corpus user
- CLI commands
main.py - Gradio interface
app.py - Groq model service
Connections 33
- Corpus user → CLI commands: runs commands
README.md - Corpus user → Gradio interface: uses interface
README.md - CLI commands → Ingest orchestration: ingests corpus
README.md - Ingest orchestration → Corpus configuration: loads sources
ingest/pipeline.py - Ingest orchestration → Source adapters: dispatches sources
ingest/pipeline.py - Source adapters → Chunk preparation: prepares chunks
ingest/adapters/files_adapter.py - Ingest orchestration → Chunk corpus: writes chunks
ingest/pipeline.py - CLI commands → Vector index: builds index
README.md - Vector index → Chunk corpus: reads chunks
retrieve/store.py - CLI commands → Vector search: queries corpus
main.py - Gradio interface → Vector search: queries corpus
app.py - Vector search → Vector index: searches vectors
retrieve/query.py - Hybrid ranking → Vector search: combines results
retrieve/hybrid.py - Hybrid ranking → BM25 search: combines results
retrieve/hybrid.py - STAR evidence pools → Hybrid ranking: gathers evidence
retrieve/star_evidence.py - CLI commands → Hybrid ranking: retrieves context
main.py - CLI commands → STAR evidence pools: gathers STAR evidence
main.py - CLI commands → Critic-refiner loop: runs artifact loop
main.py - Gradio interface → Critic-refiner loop: runs artifact loop
app.py - Critic-refiner loop → Artifact generator: generates candidates
critic/loop.py - Artifact generator → Output schemas: validates output
generate/schema.py - Critic-refiner loop → Independent Critic: checks claims
critic/loop.py - Critic-refiner loop → Quantity guard: checks quantities
critic/loop.py - Critic-refiner loop → Run trace: records verdicts
critic/loop.py - CLI commands → Artifact renderer: renders artifact
main.py - Critic-refiner loop → Artifact renderer: returns kept items
- Artifact generator → Groq model service: requests generation
README.md - Independent Critic → Groq model service: requests verdict
README.md - CLI commands → Evaluation harness: runs evaluation
main.py - Evaluation harness → Critic-refiner loop: measures grounding
eval/run_eval.py - Evaluation harness → Vector search: measures recall
eval/run_eval.py - CLI commands → Strike analysis: reports strikes
main.py - Evaluation harness → Strike analysis: analyzes strikes