Forge Intelligence

Handbook

ADR-0001: Intelligence as reasoning economy

Accepted for v0 scaffold.

Updated

Status

Accepted for v0 scaffold.

Context

LCDL and lmeta already provide deterministic-first orchestration with opt-in LLM break-in. Consumers (KA, Cockpit, Dark Factory, Campaigns) each implement local routing ladders. Token spend and ambiguity handling are documented but not standardized on the ForgeRun spine.

Platform also uses "intelligence layer" for documentation hydration — a different concern.

Decision

Introduce forge-intelligence as a private product that owns:

  1. Problem-class catalog and pack policies (general + specific intelligence).
  2. Per-domain lmeta libraries and LCDL task references (not a forked runtime).
  3. Session dual-wiki (machine JSON + generated human report + freeze gate).
  4. v1 attach schemas: intelligence_policy, ambiguity_assessment, token_ledger, route_trace.

Schemas live in fi v1; Platform ADR and boundaries reference them. Promotion into forge-platform/schemas/ follows evidence from consumer integration.

Consequences

  • Reasoning economy becomes a named product with handbook pair (fi/fiw).
  • LCDL remains the governed reasoning executor.
  • Doc hydration naming must qualify "intelligence" as docs vs reasoning when cross-linking.
  • Consumers opt in via problem_class_id; no hard import requirement in v0.