Why the verdict comes
last.

The trust problem

A verdict is only persuasive if the reader already accepts the referee.

Opening with “unsupported, confidence high” repeats the structure that made the question contested in the first place: an institution asks the reader to accept its conclusion.

That shortcut works when the reader already grants the institution. It fails precisely where DYOR is intended to help: a claim arrives wrapped in distrust, and another summary cannot dissolve that distrust by sounding more certain.

The useful output is therefore not a better pronouncement. It is a path the reader can walk: origin, retelling, prediction, test, observation, contradiction, and only then conclusion. The reader should be able to locate the exact step where their own judgment diverges.

Make the reasoning inspectable enough that disagreement has an address.

Follow the claim

Start with provenance, because repetition can impersonate corroboration.

A claim seen on hundreds of sites may still descend from one original statement. Host count tells you how far it traveled, not how many independent observations support it.

The dossier reconstructs where the claim originated and what each retelling added or dropped. Sources collapse by their earliest identifiable origin, not by domain. Four hundred paraphrases of one blog post still count as one origin family.

This is more than citation hygiene. It changes confidence. Untraced provenance cannot silently become independence, so sources with unknown origins never raise the confidence level and cap it at moderate.

Decide before looking

Write down what the claim predicts before running the search.

A failed search means something only if the expected observation and the place it should appear were named in advance.

For every required condition, DYOR registers a probe: what should be findable if the explanation is true, and where a researcher should be able to find it. Only after registration can the probe resolve to found, not found, or unsearchable.

The distinction matters. Not found can become evidence of absence when the test was genuinely diagnostic. Unsearchable is not a negative result; it means the test could not be run. Collapsing those outcomes would reward a hypothesis for evidence nobody was able to look for.

Expose the burden

A large claim is often a stack of smaller conditions hiding inside one sentence.

The data existed. The people involved knew. Nobody spoke. The oversight process failed. The records disappeared. Every condition must hold, but the claim usually presents them as one seamless story.

DYOR enumerates those conditions instead of inventing a probability for the whole claim. Each condition gets its own subquestion and can receive its own probe.

This makes the weakest link visible. It also avoids false precision: a narrated prior moves the argument onto the number, while a named condition gives the reader something concrete to inspect.

No narrated certainty

Confidence is computed from a ledger, not typed into a paragraph.

A badge can hide the thing a reader most needs to know: whether ten citations represent ten independent origins or one story repeated ten times.

The confidence assessment stores its inputs: supporting passages, distinct origin families, untraced sources, and whether counter-evidence was found. The level is derived from those ingredients.

The public report should show the ingredients beside the level. The reader can then disagree with a source, an origin link, or an evidentiary bearing without having to reject a mysterious score.

The publication promise

A report should be fast to read and slow to exhaust.

The first pass gives the question, scope, current state of knowledge, key findings, and uncertainty. Every important claim then opens into the evidence and reasoning beneath it.

Progressive disclosure is not a way to hide complexity. It is how a public site respects both readers: the person who needs the shape of the answer and the person who wants to reconstruct it from the source ledger.

Finished reports will carry the full chain: bounded question, definitions, competing explanations, pre-registered probes, source provenance, verified excerpts, evidentiary bearings, contradictions, findings typed as observed or inferred, derived confidence, limitations, and open questions.

Readable in minutes. Auditable for hours.