The standard
What a claim has to survive before we publish it.
This is a set of requirements, not a workflow. Every one of them is checkable from the finished piece — if a synthesis does not meet them, you can tell by reading it.
Five requirements
- 01
Traced
Back to the primary source, not to whoever repeated it most recently.
- 02
Accessed
Full text preferred. A claim never reaches past what could actually be verified.
- 03
Weighed
Against what would have to be true for the claim to be wrong.
- 04
Classified
One of four verdicts. No fifth option, no hedging between them.
- 05
Translated
Into what the evidence changes in practice — including when that is nothing.
Traced
A claim is traced back to the evidence it rests on, not to who repeated it.
Most performance advice is a conclusion that has travelled a long way from its source. It gets quoted, compressed, quoted again, and by the fourth repetition the qualifier has fallen off. The first requirement is to walk it back to what it actually rests on — a trial, a review, a guideline — and to record which of those it is, because what a source is determines what it can support.
Where a claim traces to a source that does not say what the claim says, that becomes the finding. Several pieces in this archive exist for exactly that reason.
Accessed
Full text is preferred, and the claim never reaches past what could be verified.
Abstracts overstate. Effect sizes shrink between the abstract and the results table, control conditions turn out weaker than described, and the population is narrower than the conclusion implies. So full text is what a synthesis is built on wherever it can be obtained.
Sometimes it cannot be. When only an abstract or a citation record is available, the claim is explicitly constrained by what that actually establishes — and a record that carries no more than a title and a journal cannot substantively support a claim at all. What was read is part of what is known.
Weighed
Disagreement is looked for, not averaged out.
Every question is worked from both directions: what supports the claim, and what would have to be true for it to be wrong. The second search is the one that changes verdicts, and it is the one most easily skipped, because the first already produced an answer.
Where credible sources disagree, the disagreement carries through to publication rather than being resolved by picking a side. A confident answer to a contested question is a failure, not a service.
Classified
Four verdicts, because “it works” and “we don’t know yet” are different answers.
Every synthesis carries exactly one of these, visible on the card before you open it. There is no fifth option and no hedging between them.
- Supported
- The claim is adequately supported by relevant evidence within its stated boundaries, with no unresolved contradiction strong enough to overturn the conclusion.
- Contested
- Credible primary sources disagree. The disagreement is real and unresolved, not an artefact of one weak study.
- Uncertain
- The evidence is too thin, too small, or too indirect to support a conclusion in either direction. Absence of evidence, not evidence of absence.
- Unsupported
- The claim is widely repeated but the primary sources do not support it, or the sources cited do not say what they are said to say.
Translated
Into what the evidence changes in practice — including when that is nothing.
Every synthesis says what the evidence changes in practice, and the recommendation never reaches further than the evidence does. Where the honest reading is that nothing should change yet, that is the practical implication, stated as plainly as any other.
It also ends in limitations. Not as a disclaimer at the bottom, but as part of the answer: knowing where a finding stops applying is most of knowing how to use it.
On tools
Searching, retrieval and drafting all use software, the way every research process has for thirty years. Nothing in the list above is satisfied by software, and no tool decides a verdict — agreement between two search passes is a process signal, not evidence. Nothing publishes that cannot be defended line by line against its sources.