Handbook
Glossary
Operator-facing vocabulary. Technical depth: system handbook.
Updated
| Term | Definition |
|---|---|
| Ambiguity | When deterministic stages cannot decide confidently; FI records the assessment instead of expanding prompts blindly. |
| Break-in | A bounded LLM step with a closed set of allowed actions — used only when cheaper rungs are insufficient. |
| Concept dual-wiki | Paired semantic (what an intelligence means) and execution (how it runs) pages per intelligence identity. |
| Documentation dual-wiki | Operator handbook (user concepts) paired with system handbook (maintainer depth). |
| Domain pack | A specific-intelligence bundle for one problem family (e.g. Human-Life, MathGenesis). |
| Freeze gate | Check that a session's human report still matches its machine records — prevents narrative drift. |
| General intelligence | Catalog and reasoning ladder that runs before domain-specific packs. |
| Handbook pair | Product Markdown in fi published as HTML in fiw. |
| Intelligence session | One governed run from problem class through pack execution to auditable evidence. |
| Maturity (I1/I2) | Honest readiness labels — I1 means contract-level runtime evidence, not domain expert validation. |
| Problem class | A named category of reasoning work that selects pack policy and ladder. |
| Reasoning economy | Spend deterministic and cheap steps before scarce LLM and human attention. |
| Reasoning ladder | Ordered rungs: deterministic → cheap model → break-in → human escalate. |
| Session dual-wiki | Machine JSON records (source of truth) plus a generated human-readable session report. |
| Specific intelligence | Per-domain packs with policies, flows, and fixtures for one problem family. |
| Token budget | Caps on LLM spend per session stage — recorded in the session ledger. |