Handbook
Charter
Forge Intelligence makes problem-class reasoning economical: maximize deterministic control flow and minimize ambiguous LLM spend, while still allowing bounded LLM acceleration where judgment is genuinely needed.
Updated
Mission
Forge Intelligence makes problem-class reasoning economical: maximize deterministic control flow and minimize ambiguous LLM spend, while still allowing bounded LLM acceleration where judgment is genuinely needed.
North-star promise
From problem class to auditable session: pick the right ladder, run the right pack, record why tokens were (or were not) spent, and attach evidence consumers can review.
Architectural stance
Forge Intelligence is not an agent and not a second LCDL runtime. It is a pack
library + policy contracts + session dual-wiki that composes forge-lcdl tasks and lmeta
flows under explicit problem-class policies. node_executor is a thin adapter only —
see ADR-0002.
Two intelligence modes
| Mode | Job |
|---|---|
| General | Catalog, reasoning ladder, ambiguity/token policy selection |
| Specific | Per-domain packs (packs/<domain>/) with policies, lmeta, fixtures |
Guardrail
A feature belongs in forge-intelligence only if it improves reasoning economy for a named problem class and produces attachable session telemetry (route trace, token ledger, ambiguity assessment).
Distinction from doc hydration
Platform's documentation "intelligence layer" (claim registries, hydration maturity) is docs governance. Forge Intelligence is runtime reasoning economy — different product, different contracts.