docs: add MGTP programme positioning v0.1

docs/MGTP_PROGRAMME_POSITIONING.md

@@ -0,0 +1,42 @@
+# MGTP Programme Positioning (v0.1)
+
+## Core invariant
+Governance must bind at runtime, not at policy.
+
+## Scope boundary
+MGTP governs execution transitions only: tool calls, service hops, state mutations, and irreversible commits.
+MGTP does not attempt global semantic safety.
+
+## Non-goals
+- No alignment claims
+- No general AI safety claims
+- No marketing language
+- No behavioural modelling
+
+## Lanes
+
+### Lane A — Builds (Executable proof)
+- Transition registry
+- Deterministic evaluate_transition
+- DecisionRecord artefacts
+- Execution-boundary conformance tests
+
+### Lane B — Articles (Translation layer)
+- Runtime vs policy enforcement
+- Halt rights vs monitoring
+- Authority leakage at transitions
+
+### Lane C — Academic papers (Formal model)
+- State space definition
+- Authority lattice
+- Boundary completeness theorem sketch
+- Mapping to EU AI Act Articles 9, 12, 14
+
+## Calibre standard (acceptance criteria)
+1. Deterministic (byte-identical replay)
+2. No primitive mutation without explicit version bump
+3. Registry version tied to ENGINEERING_STATUS.yaml
+4. Transitions enumerated
+5. Refusal propagation test present
+6. No hidden state
+7. CI matrix proof across Python versions