Governance,
not persuasion.

The first version written from measurement of a running implementation. Stages produce, Gates permit: Stages 1–6 and Gates 7–10. Recognition inverts the premise default, so nothing arrives recognised. Revision 1 makes Gate 1 advisory, replaces sixteen phrase lists with a grammar layer, opens the probe route, and gives Gate 9 the power to send faulty prose back to the composer rather than replace it.

NMA-14 Rev.1 verifies the plan; nothing in it ever inspects the words the navigator reads. NMA-15 adds Gate 5 Prose Traceability and Gate 6 Posture Conformance, both deterministic and both run on the generated prose, plus provenance tiers and the one-decision-maker rule. Model output is never a source.

Corrects Routing Rule 2 (I:HIGH alone routes NMA-3), adds categorical I/A screening, Glass Box accountability mechanism, two-stage BOF, forced station engagement, Threshold Ambiguity mode, and seven scenario types. Structure before synthesis — including the premises themselves.

Adds Gate 2 Step 0 — mandatory premise verification before any claim classification. Auto-Govern mode, pre-committed BOF variant, and Constraint 13. Structure before synthesis — including the premises themselves.

Complete governance architecture unifying NMA-11A and NMA-12. Mandatory pipeline, execution modes, ICXATRS with overrides, seven-station committee, BOF arbitration, and S8 gatekeeper.

Seven governance levels, eight deliberative stations, six commercial clusters. The complete multilevel pipeline — execution-verified.

Minimum viable governance for high-stakes AI interactions. Pure behavioral constraint — the activation key any AI system can execute.

Structure before synthesis. Six-step output pipeline enforcing Flow Mode — eliminating fabricated certainty, narrative smoothing, and premature closure.

Source code documentation for the full governance engine: core pipeline (Gate 1–S8), four autonomous research loops (A/B/C/D), Inner House deliberation system, active learning module, memory architecture, NMA-6 premise verification, test battery (50 scenarios), and comparative experiment infrastructure.

Traces the full NMA lineage from a clinical observation through fourteen architecture versions. Covers Push Mode and RLHF miscalibration, the Navigational Mind Framework, what each gate in the pipeline actually does, the Glass Box accountability mechanism, and why the architecture will be replaced. An honest account of what worked, what failed, and why both matter equally.