Handbook
Platform overview
Operators, integrators, and product teams who need governed problem-class reasoning — routing work through deterministic steps before spending LLM tokens or human attention.
Updated
Who is this for?
Operators, integrators, and product teams who need governed problem-class reasoning — routing work through deterministic steps before spending LLM tokens or human attention.
What problem does it solve?
Teams delegate more work to models and agents, but still need clear intent, bounded escalation, and evidence they can review. Forge Intelligence makes reasoning economical: pick the right ladder for a problem class, run the right domain pack, and record why tokens were or were not spent.
What can you do?
- Select a problem class and let general intelligence resolve the reasoning ladder and pack policy.
- Run domain packs (Human-Life, MathGenesis, PhysicsGenesis, and others) under explicit policies.
- Produce an auditable session with route trace, token ledger, and ambiguity outcomes.
- Review a human-readable session report derived from machine records — not hand-edited narrative.
- Attach session evidence to downstream consumers (ForgeRun, Lenses) without FI becoming the system of record.
What can you not do?
- Use Forge Intelligence as a general-purpose agent or chat runtime.
- Replace LCDL, Fleet, or Lenses — FI composes LCDL tasks and attaches evidence; it does not own execution or run approval.
- Treat the capability catalog as proof of domain expertise — most nodes are contract-level readiness, not validated expert behavior.
Limitations
Forge Intelligence is runtime reasoning economy, not Platform doc-hydration or claim registries. Maturity labels (I1/I2) describe contract evidence, not finished product features. See Boundaries and the system maturity pages for honest scope.
Trust model
| Boundary | What it means |
|---|---|
| Data | Session machine JSON is the source of truth; human reports are regenerated projections. |
| Execution | FI selects policy and records telemetry; LCDL executes governed tasks. |
| Human control | Escalation rungs and freeze gates preserve auditability before release. |
| Evidence | Route trace, token ledger, and ambiguity assessments attach to sessions. |