Four phases, asked in sequence, that separate innovation programs that scale from pilots that die. Applicable wherever large organizations struggle to turn policy into practice. Validated empirically in the defense AI sector, where the gap between governance and behavior is widest and the stakes are highest.
Each phase builds on the last. Skip one and the next one breaks. Most innovation programs start at adoption and wonder why nothing sticks.
Where is the system broken?
Before you can innovate, you have to understand what's in the way. Policy barriers, bureaucratic bottlenecks, misaligned funding, legacy constraints. Most transformation efforts skip this step and build solutions for problems they haven't actually diagnosed.
Where are doors opening?
Friction creates demand. Pathways are where that demand meets opportunity — new authorities, funding shifts, organizational realignments, market movements. The window is usually shorter than you think.
How do you move through the door without breaking what's on the other side?
Adoption without governance fails. But governance without behavioral follow-through also fails — and that's the less obvious trap. The question isn't whether you have a policy. It's whether the policy changes what people and systems actually do.
Did it work? How do you know?
Impact isn't "we deployed it." Impact is measurable change in the outcome you set out to affect. Without this, you can't tell the difference between a successful pilot and a vanity project — and you can't justify the next cycle.
Most innovation programs start at Execute. This methodology starts at Diagnose. That's the difference between a pilot that dies and a capability that scales.
The first application of this methodology became a published research program when we asked: does AI governance actually change organizational behavior? The answer — across thousands of cases and controlled experiments — is no. Not by itself. These findings validate the core premise across any domain where governance outpaces behavioral change.
3,533 cases across 49 countries. Governance frameworks correlate with more incidents, not fewer. The governance gap is the highest-scoring failure dimension.
79 experiments. 36,000+ agent messages. 112 findings. AI systems pass governance checks while group diversity collapses undetected. Visible monitoring suppresses drift 3–6×. Open source.
Pillars 1, 3, and 4 are grounded in published research. Pillar 2 — finding pathways — is applied weekly through The All Source Forge's signal intelligence and opportunity analysis. See all research →
The All Source Forge is the first product built on this methodology — evidence-based intelligence for people who build, govern, or deploy AI in defense. Weekly signal briefs. Deep dive opportunity analysis. Every signal analyzed through the four pillars.
AuspexAI is the operational network for AI safety research — volunteers donate compute, researchers propose experiments, the platform handles trust, replication, and signed audit logs. SENTINEL runs as the first tenant; the SDK is open to other research projects.