Analysis application tools
This directory owns application materials, decoding, evidence and analysis rules without adding Core dependencies or private nodes.
domain.ts: request/schema validation, units, deduplication, conflicts, reference checks, comparisons and serialization.adapters.ts: snapshots, text/HTTP reading, normalization, an external SQLite Provider and local report tools.
Reuse file-ingestion RegisteredTools for PDF, CSV/XLSX and image OCR, and retrieval/web.ts for bounded HTTP/LinkeDOM parsing. retrieval/memory.ts explicitly ingests internal knowledge through the Memory Worker; analysis graphs read it through MEMORY.SEARCH. storage initializes Redis Context and database Memory.
AnalysisAdapters(directory, request, python) exposes tools, providers, and close(). The external database uses documents(id, tenant, body) and queries only an approved document ID plus bound tenant. Internal Memory validates key, kind and tenant. A trusted application configures sources, authority, periods and page origins; the model cannot change them.
Register analysis_read, analysis_normalize, analysis_report, analysis_publish and reused decode_pdf, read_spreadsheet, ocr_image through createInteractionWorker({ tools }); register the external Provider through createRetrievalWorker({ providers }). Execute through Runtime and close the adapter afterward.
HTTP uses approved origins, no redirects, a 20-second deadline and a 4 MiB response cap. Production URLs use HTTPS; explicitly configured loopback HTTP supports the fixture publisher. Local inputs must be regular files up to 4 MiB. Normalized sources allow at most 30 blocks of at most 1200 characters each. Only trusted controllers should modify sources, Memory and task directories.