How neurostuff/pondie works
Pondie turns neuroimaging papers into validated, queryable records, maps source wording to shared values, and selects records and coordinates for meta-analysis. The documented extraction journey moves from paper inputs through nine ordered stages to record assembly and validation; optional repair is separate. Normalization and query selection are downstream capabilities. The graph follows README-described contracts and stages; source excerpts are partial, so implementation-level calls not explicitly evidenced are omitted.
Main entry points: stages.py Extraction stages · builder.py Record assembly · normalization Field normalization · driver.py Extraction runner
Components 19
Paper extraction
- Extraction runner
pondie/extraction/driver.py - Extraction stages
pondie/extraction/stages.py - Corpus preparation
pondie/extraction/corpus - Prompt rendering
pondie/extraction/prompt - Evidence retrieval
pondie/extraction/evidence - Optional repair
pondie/extraction/repair
Record contracts
- Record assembly
pondie/extraction/record/builder.py - Deterministic fixes
pondie/extraction/record/fix - Record validation
pondie/extraction/record/validate.py
Normalization
- Field normalization
pondie/normalization
Meta-analysis selection
- Pooling selection
pondie/query/engine.py
Shared foundations
- Schema contract
pondie/schema/reader.py - Record formats
pondie/formats/values.py - Shared vocabularies
pondie/vocabularies - Data locations
pondie/paths.py
Other components
- Researcher
- Paper corpus
- Model gateway
- Meta-analysis result
Connections 21
- Researcher → Extraction runner: runs extraction
README.md - Paper corpus → Extraction runner: provides papers
README.md - Extraction runner → Extraction stages: runs stages
README.md - Extraction stages → Prompt rendering: requests model work
README.md - Extraction stages → Model gateway: sends model calls
README.md - Extraction stages → Evidence retrieval: gathers evidence
README.md - Corpus preparation → Paper corpus: prepares inputs
README.md - Extraction stages → Record assembly: builds record
README.md - Record assembly → Deterministic fixes: applies repairs
README.md - Record assembly → Record validation: checks rules
README.md - Record validation → Schema contract: checks conformance
README.md - Record assembly → Record formats: uses value wrappers
README.md - Extraction stages → Optional repair: optionally repairs
README.md - Researcher → Field normalization: normalizes records
README.md - Field normalization → Shared vocabularies: maps shared terms
README.md - Researcher → Pooling selection: selects records
README.md - Pooling selection → Record formats: reads record values
README.md - Pooling selection → Meta-analysis result: returns pooled subset
README.md - Field normalization → Record formats: reads record wording
README.md - Extraction runner → Data locations: uses run locations
- Field normalization → Data locations: uses data locations
pondie/normalization/_records.py