Structural transitions admitted
Across three provider/model families under one frozen external qualification framework.
32 cases × 3 providers × 2 Fresh passes.The strongest current evidence is structural: across three tested provider/model families and two Fresh passes, AI action paths varied while the external qualification boundary remained substantially more stable.

Across three provider/model families under one frozen external qualification framework.
32 cases × 3 providers × 2 Fresh passes.Structural validity plus required semantic-coordinate coverage.
The single residual remained visible rather than being repaired or hidden.The external admission boundary agreed across both Fresh passes.
Exact action agreement was lower: 87/96.Qualified outcomes remained highly stable even when model action paths differed.
Internal same-program evidence; not population-level reliability.GPT-5.6 Sol, Claude Sonnet 5 and Gemini 3.8 Flash produced the 192 Fresh candidate slots. The same external qualification semantics admitted all 192 structural transitions.
The tested result supports maintaining qualified judgment outside any single AI model so it can be inspected, inherited and requalified by another system.
Companion Terrarium tests preserved protected state and authority boundaries while judgment lineage continued.
≈ 0.0060% of median provider latency.
This is a measurement of the deterministic transition-qualification core. It is not a claim that total deployment, evidence gathering, organizational governance, integration or review overhead is 0.0060%.
The repeated supporting evidence preserved the same architecture-native work pattern in both Fresh runs.
Version 1.3 makes the corrected prospective Fresh product qualification the primary evidence and keeps earlier mechanism, recovery and durability results as supporting evidence rather than pooling them into one global score.