Handbook
Maturity roadmap
See Maturity state for the current snapshot and Maturity hierarchy for rollups.
Updated
Completed waves (A–F + G)
| Wave | Focus | Outcome |
|---|---|---|
| A–E | General spine, reasoning reference, life clusters | Partial I1 promotion |
| F | Life equalization | Uniform I1_runtime floor — 297 allowlist, 39/39 eval pass |
| G | Validator infra + honest reskilling I2 + spine A1 | Contract validator dispatch; 10 I2 caps wired; 40 caps at A1 (reasoning + spine) |
Wave G deliverables (current)
registries/validator_index.yaml— 163 bundle validators indexed from manifestssrc/forge_intelligence/validators/— contract-level deterministic checks by validatorkind- Reskilling T1–T10 lmeta units —
validatesteps + enriched payloads audit_capability_maturity.py— fails when I2 caps lackvalidatestep- General spine autonomy —
general.question-framing,general.validation-verification,general.experimentationat A1_supervised
Planned waves
Wave H — case-led I2 expansion
Promote I2 only where a multi-node case exists with fan-in, freeze gate, and executable validators.
| Target case | Intelligences | Caps (order of magnitude) |
|---|---|---|
| Reskilling hardening (done in G) | 7 intel, T1–T10 | 10 |
| Wellbeing mini-case | mental-health, emotional-regulation, sleep, health-navigation | ~15–20 |
| Agency mini-case | household, personal-org, nutrition | ~15–20 |
Not planned: mass I2 for all 297 capabilities.
Wave I — selective life autonomy
Autonomy lags technical depth on life domains (safety/rights lexicographic priority).
| Tier | Intelligences | Target |
|---|---|---|
| Done (G) | general reasoning + spine | A1_supervised (40 caps) |
| Candidate | personal-organization, household-domestic | A1 bounded subset |
| Defer | mental-health, grief, sexual-reproductive, safety-first-response, civic-legal | A0 indefinitely |
Wave J — platform and extensions
| Item | Purpose |
|---|---|
capability_demand_router.yaml + implementation_bindings.yaml |
Demand-driven orchestration |
benchmarks/human_life/ |
Regression corpus for depth work |
| IRS/MO pack merge | Extensions beyond 297-cap v4.1 index |
Realistic ceiling (v4.1)
| Arena | Technical (12 months) | Autonomy (12 months) |
|---|---|---|
| general (8) | I2 on spine + reference packs; I3 pilot on reasoning | A1 spine; A2 on bounded coordinator flows |
| life (31) | I1 floor (done); I2 on 8–12 domains with cases | Mostly A0; A1 on 2–4 low-risk domains |
Maturity vs autonomy
Technical maturity and autonomy are orthogonal (VOCABULARY.md). In practice:
- Technical uplift scales by case graphs and validator wiring (expensive, domain-specific).
- Autonomy uplift scales by risk tier — do not promote life high-risk intelligences to A1 without domain validators and policy review.
Rule: Never promote I2 or A1 by registry sweep alone; always tie claims to executable evidence (validators, cases, audit gates).