Evidence and Uncertainty
Research decisions depend on knowing which part of an argument is secure and which part is still being tested. This chapter organizes those distinctions across fields. It complements the topic assessments, which own the detailed evidence; it does not create a competing list of discoveries or a ranking of research programs.
Find the uncertainty that matters
Section titled “Find the uncertainty that matters”| Reader question | Guide | What it helps you do |
|---|---|---|
| Which results should constrain a proposal? | Established anchors | Use a theorem or robust framework without exceeding its scope |
| What exactly is being conjectured? | Active conjectures | State a falsifiable or provable proposition with its quantifiers |
| How should an ambitious idea be assessed? | Speculative ideas | Separate a mechanism, model, prediction, and test |
| Why might an attractive inference fail? | Known pitfalls | Audit limits, selection, resources, and model ambiguity |
| Where is missing evidence a research opportunity? | Evidence gaps | Design a discriminating and feasible next result |
Use status labels for the vocabulary and evidence standards for the claim-specific contract. Stable foundational arguments are developed in Foundations.
A living assessment changes for a stated reason
Section titled “A living assessment changes for a stated reason”A useful revision identifies the old claim, the new evidence, the affected assumption, and the resulting change in scope. It need not produce a dramatic reversal. A result may become more reliable in one regime and less general in another. A new proof can close a mathematical gap while leaving the experimental identification problem untouched.
The annual archive preserves dated perspectives. Read a historical record as evidence about what was supported at its cutoff, and a topic page as a bounded assessment of its cited literature. A recent review date is not a guarantee of exhaustive coverage.
Disagreement can identify the next project
Section titled “Disagreement can identify the next project”When two papers appear to disagree, first align the system, observable, input ensemble, approximation, and limit. Different boundary conditions or resource constraints can make both conclusions valid. If the disagreement survives that alignment, identify a calculation or measurement on which the explanations differ.
For example, a finite-time memory plateau and an asymptotic thermalization scenario need not contradict one another. The many-body localization assessment explains why system size, time, disorder, and rare regions matter. A useful project can show that present data are insufficient to discriminate the scenarios, provided it states what additional evidence is needed.
Keep the conclusion proportionate
Section titled “Keep the conclusion proportionate”An unresolved inference is not evidence that every alternative is equally plausible. Weight alternatives by their explanatory scope, internal consistency, independent support, and the tests they survive. At the same time, do not promote a preferred interpretation to an established fact simply because its competitors are difficult.
The practical output is a revised project decision: continue, narrow, seek another observable, improve a baseline, or stop. Choosing a Research Direction makes those decisions explicit before major resources are committed.