@theneuralledger/research
v0.1.0
Published
Evidence-first TNL research skills, orchestration, and application.
Maintainers
Readme
@theneuralledger/research
Shared, evidence-first research contracts and orchestration for The Neural Ledger.
The package contains six immutable research skills, bounded TNL/Docdex/web/Codali adapters, tenant-scoped result caching, deterministic graders, an authorized HTTP service boundary, a standalone research workspace, and an MCP App resource.
Local deterministic service
npm run build --workspace @theneuralledger/research
TNL_RESEARCH_DEV_SERVICE=1 node packages/research/dist/service-bin.jsOpen http://127.0.0.1:7425. The development entrypoint uses synthetic evidence,
accepts loopback-only development identity headers, and refuses to start when
NODE_ENV=production.
Production deployments import ResearchOrchestrator into the authenticated TNL
service and provide durable tenant storage plus configured TNL, Docdex, approved
web, and Codali HTTPS adapters. Credentials remain server-side. The research
runtime never shells out from browser code and never modifies BDYA state.
Result guarantees
- Material factual claims must cite normalized evidence.
- Prompt-like source text is quarantined before synthesis.
- Fact, inference, forecast, and unknown classifications remain structured.
- Event, publication, revision, retrieval, and
asOftimes remain distinct. - Budgets bound tool calls, duration, tokens, sources, and cost.
- Traces contain stage summaries and provider versions, not hidden reasoning.
- Generated work is identified as
TNL Botand is not trading advice.
