This release asks a simple but important question. If many different histories of speciation and extinction can produce the same signal in a family tree, what can one tree really tell us once ordinary sampling uncertainty is included? The answer is not a single best history. It is a set of histories and, sometimes, a decision that remains valid across that whole set. The paper studies one exact fixed-stem reconstructed tree under a homogeneous time-varying birth-death model, conditional on the stem lineage surviving to leave sampled descendants. In that model, the number of tips has an exact geometric distribution. Given the number of tips, the unordered internal node ages have a known transformation through the pulled diversification scale, called F. Those two facts make it possible to build a finite-sample confidence set for the whole F trajectory. Exact inversion handles the tip count. A Dvoretzky--Kiefer--Wolfowitz--Massart band handles the node ages. Bonferroni propagation combines the two pieces. The candidate then carries that uncertainty into an affine representation of all compatible speciation and extinction histories. Instead of deciding from one plug-in curve, it asks whether a turnover cap is impossible for every signal in the confidence set, possible for every signal, or unresolved because the answer changes across the set. In the frozen 22-tip synthetic example, a cap of 0.70 is certified incompatible under the declared normalized constraint. A different cap that looks compatible on the plug-in curve becomes unresolved after signal uncertainty is included. This is the intended scientific lesson: a confident-looking answer from one estimated signal can disappear when the signal's finite-sample uncertainty is propagated honestly. The paper also reports a separate preregistered comparison with CRABS. That benchmark is deliberately not the main result. Across the 240 primary cases that returned a cloud, every cloud missed at least one sharp endpoint beyond the frozen tolerance, and 60 additional primary cases were structurally censored. Yet exact certification changed only 532 of 3,840 clustered decision statuses, 13.854 percent, below the preregistered 20 percent utility threshold. The H4 gate therefore failed. The release does not claim that the affine method is a must-have, an essential complement to CRABS, or broadly superior. It preserves both parts of the result: finite random clouds are not extremum certificates in the registered benchmark, and the measured decision impact was smaller than the prespecified target. The package contains a 23-page paper, accessible Markdown, exact code, 21 deterministic tests in ordinary and optimized Python, 720 implementation-diverse endpoint checks, a 160,000-replicate branching simulation, five detected semantic negative controls, a 20,000-replicate fixed-stem sanity check, and a sealed 1,100-cell comparison ledger with 1,900 manifest entries. Public GitHub Actions replay also passes, including a pinned Linux-container job. These are producer-side checks. They do not substitute for an unaffiliated process-theory review, an independently authored implementation, proof-assistant formalization, biological validation, or journal peer review. The confidence statement is conditional on exact node ages, the fixed-stem and stem-survival conventions, and the homogeneous time-varying model. It does not cover topology estimation, node dating, smoothing, lineage or trait dependence, model selection, misspecification, fossil preservation, or uncertainty in external biological constraints. Sharpness is only over the declared conditional model class. The scholarly creator is Anonymous; Ian Pitchford is the repository maintainer and Evidence Press publisher. This synthetic-voice briefing was released on 21 August 2026. It is a communication aid, not additional scientific evidence.