Evidence, Limits, and Non-Claims

Most decision failures are not failures of intelligence. They are failures of bookkeeping: nobody recorded how strong the support actually was, so a thin conclusion travelled as far as a solid one.

Evidence discipline is the habit of attaching the strength of support to the claim itself, permanently, so the two cannot be separated in transit.

A documented capability is something the framework definition states directly: for example, that release logic classifies output into five categories.

An inference is a reasonable step beyond that statement: for example, that applying contradiction testing reduces premature conclusions. Plausible, consistent with analytic practice, and still an inference.

The distinction matters because inferences accumulate. Three inferences chained together produce a claim nobody would have accepted if stated outright.

That a modelled scenario shows what will happen. It shows what would follow if the stated assumptions hold.

That structured analysis produced a correct answer. It produced a recorded one.

That the framework covers a decision type simply because the vocabulary fits.

That a component named in the framework is present in a given implementation.

That governance language implies enforcement. Release classes advise; they do not act.

Ambiguity behaves differently by field. In research it is normal and publishable. In operations it is a live risk. In regulated settings it may be a compliance event.

The framework's response is the same everywhere: name the ambiguity, record who it affects, and let it change the release class rather than the wording.

When support is insufficient, the default is to withhold. In practice that looks like a shorter output, not a hedged one: the supported portion is stated, the unsupported portion is returned as unresolved, and the requester is told what would be needed to go further.

The uncomfortable part is that this sometimes means delivering nothing when something was expected. That is the intended behaviour, not a shortfall.

Escalate when the decision exceeds declared authority, when a material contradiction cannot be resolved with available evidence, or when the risk class is safety- or rights-critical and support is partial.

Escalation is not a failure of the analysis. It is the analysis working: it identified that the decision belongs somewhere else.

This site does not present outcome studies, benchmarks, or deployment evidence. Whether adopting the framework improves decisions in a measurable way is unresolved here.

Component availability is also unresolved in the general case. Verify against your own implementation documentation before relying on any specific layer.