The argument in text
For marginals (0.10, 0.10) the both-fail probability is identified only to [0.0, 0.10]; both endpoints are attained by explicit joint distributions that satisfy the marginals; independence names 0.01, a point inside that set it was never entitled to.
A declared illustrative input (CC-001) and the kernel's bounds and witnesses (CC-001, CC-004), reproduced from a clean clone in CI. The one hundred items, which plate lets which through, and the order of the migration are illustrative constructions (the Same Scores seed); no measured system appears. Plates are a metaphor for block-on-any composition with full exposure; sequential routing where the second plate never sees what the first stopped is a different estimand and is not shown.
What this does not establish
- does not certify any stacked system as safe, and shows no measured system
- does not estimate or recover the true dependence between any two guards
- does not select a point inside the interval; the amber point is what one assumption would select
- does not depict sequential routing; a second guard that never sees what the first blocked is a different estimand
- the items and the migration order are illustrations of feasible worlds, not data
What would require a correction
A clean-clone execution of the bound kernel on marginals (0.10, 0.10) returns AND bounds other than [0.0, 0.10], or a witness that fails a marginal check at tolerance 1e-9 (CC-001 / CC-004, REJECT); or any rendered frame shows a plate readout other than 10/100 while the pile moves — checkable frame by frame.
Sources and render record
Film manifest · Render receipt · First site deployment record
Current render: 2026-09-15. The receipt records the rendered inputs and output hash. It checks provenance, not scientific validity.