These operations describe the v1.0.0 processing baseline.
victus-processing is operated as a local CLI-driven batch pipeline. Operators
run explicit stages, inspect local artifacts, and rerun failed or incomplete
work from persisted state.
Primary operational responsibilities:
data/ artifacts unless cleanup is intentional.uv, local files, and.env.Create or verify local runtime layout:
uv run victus-processing data-layout create
Inspect CLI:
uv run victus-processing --help
Run the main local flow:
uv run victus-processing metadata-extraction explore --mode broad-nutrition
uv run victus-processing bibliography-export generate-bib
uv run victus-processing pdf-intake link --metadata-id meta:s2:example --pdf data/artifacts/intake/pdfs/example.pdf
uv run victus-processing pdf-processing run
uv run victus-processing evidence-extraction run
Run a single DOI metadata fetch:
uv run victus-processing metadata-extraction from-doi --doi 10.1000/demo
This writes directly to data/lake/paper_metadata.jsonl.
Run smoke validation:
uv run pytest tests/test_cli_smoke.py -q
More CLI detail: CLI operations.
PDF-processing detail: PDF processing operations.
Runtime defaults live in config/*.yaml. Relative paths resolve from the
repository root. Optional root config.yaml can override domain config.
Secrets are read from environment variables. .env is only a local fallback;
prefer Infisical injection for real runs.
Common variables:
SEMANTIC_SCHOLAR_API_KEYLITELLM_PROXY_API_BASELITELLM_PROXY_API_KEY or legacy LITELLM_KEYLITELLM_METADATA_SELECTION_MODELLANGFUSE_PUBLIC_KEYLANGFUSE_SECRET_KEYLANGFUSE_HOSTPROMPT_LABELPROMPTS_LOCAL_DIRDEFAULT_LLM_MODELInfisical helper:
uv run victus-infisical-env export --env dev --path / --output .env
uv run victus-infisical-env run --env dev --path / -- victus-processing --help
Prompt Management:
src/prompts/ are fallback prompts.max_tokens; explicit stage config or remoteConfiguration contracts and stable path expectations live in
Contracts.
There is no central logging or metrics service in this repository.
Operational inspection sources:
paper_processing_state for per-paper progress;structured_blocks, paper_classifications, experiment_maps, andcanonical_evidence for processed outputs;data/reports/exports/paper_processing_state.csv for quick spreadsheetdata/ artifacts during recovery unless cleanup is explicit.SEMANTIC_SCHOLAR_API_KEY: Semantic Scholar may run with stricterdata/artifacts/pdfs/.uv run victus-processing --help and the smokeOperations covers:
Operations does not cover: