E.16:3 - Forces
| Force | Tension |
|---|---|
| Creativity vs Safety | Exploration autonomy vs hard constraints and override duties |
| Locality vs Comparability | A budget stays bound to its claim, working situation, scope, window, policy, and override rule. A scheduled action check uses the actual holders, assignments and authority. A Work reference appears only when performance independently qualifies under A.15.1. |
| Simplicity vs Auditability | Lightweight authoring vs ledger‑grade evidence |
| Autonomy vs SoD | Helpful self‑action vs separation‑of‑duties and human‑in‑the‑loop points |
E.16:3.1 - Bias-Annotation
- Gov. Bias toward enforceable oversight (hard gates, SoD, canonical override SpeechActs). Autonomous exploration is allowed only within a stated budget and time window.
- Arch. Bias toward gate‑and‑ledger structure (Green‑Gate + Work‑anchored
AutonomyLedger). Mitigation:telemetrySpecRefcan scope what is emitted when full deltas are unnecessary. - Onto/Epist. Bias toward typed, testable constraints (MM‑CHR tokens, explicit admissibility checks). Mitigation: the budget fields marked
?are optional, so low‑risk contexts can start with a minimal budget and tighten its limits over time. - Prag. Bias toward measurable quotas may under‑express “soft” autonomy goals. Mitigation: pair
decision_tokenswithrisk_bandsto capture non‑counting limits. - Did. Bias toward explicit mechanics increases authoring surface area. Mitigation: provide a default
AutonomyBudgetDecltemplate and minimal harness cases in F.15.