Organisations adopt practices for all sorts of reasons. Because everyone else uses them. Because a vendor packaged them well. Because they were installed years ago by someone who has since left. Whether the practice works is often the last question asked, and sometimes it is never asked at all.
Proofweight exists to ask it. This page explains how.
How a question gets chosen
Not every interesting question is worth an investigation. A question earns one here when three things are true.
A decision hangs on it. Someone has to choose whether to keep a practice, change it, or stop it, and the answer would change what they do on Monday.
Organisations are acting on it now. This is not a survey of the literature for its own sake. The questions are the ones being decided in real companies this year.
The confidence around it exceeds the evidence behind it. This is the important one. Proofweight is most useful where a claim is repeated with more certainty than anyone has earned.
A topic trending on LinkedIn is not a reason to cover it. It is sometimes a reason to examine why it is trending.
The test
Most essays work through the same seven steps. The headings vary; the reasoning does not.
1. The claim. What exactly is being asserted, stated at its strongest rather than its weakest. A claim worth testing deserves a fair hearing first.
2. Why it is attractive. What problem does it appear to solve? Practices survive because they do something for someone, and understanding what usually explains more than dismissing it.
3. The evidence. What kinds of evidence exist, what they establish, and how directly they bear on the question actually being asked.
4. The counterweight. What contradicts, limits, or qualifies the dominant claim. What alternative explanations fit the same facts.
5. The organisational reality. How incentives, power, implementation constraints and unintended consequences change the result. A practice that works in a study can fail in a company for reasons that have nothing to do with the study.
6. The decision. What a leader or practitioner should consider doing, and the conditions under which that advice would change.
7. The confidence level. How certain you can reasonably be. Stated plainly, including when the answer is "not very".
What counts as evidence
Evidence is not one thing, and treating it as one thing is how bad decisions get made confidently. Proofweight distinguishes between:
- systematic reviews and meta-analyses, which aggregate many studies;
- replicated field studies, which have held up in more than one setting;
- single studies, which are a starting point and not a conclusion;
- company cases, which describe what happened somewhere without establishing why;
- practitioner experience, which knows how things behave in practice and proves nothing on its own.
Sources produced by vendors, advocacy organisations and consultancies are identified as such. That does not make them wrong. It makes them interested, and you are entitled to know.
A company case is not an experiment unless it was designed as one. "We removed ratings and engagement went up" tells you two things happened. It does not tell you one caused the other.
When the evidence disagrees
Sometimes credible work points in different directions. When that happens, Proofweight says so rather than picking the side that makes a better headline.
"Contested" is a real answer. It is often the most useful one, because it tells you the decision is yours to make on judgement rather than on findings, and that anyone selling you certainty is selling something.
The evidence maturity labels
Every essay carries one. They describe how mature the evidence is. They are not quality scores, and they are not truth scores.
| Label | What it means |
|---|---|
| Observed | Reported or documented, but causality has not been established |
| Early evidence | Some credible evidence exists, but it is narrow or not yet replicated |
| Replicated | The finding has appeared across multiple credible studies or settings |
| Synthesized | Systematic reviews or a comparable body of evidence allow a broader conclusion |
| Contested | Credible evidence or interpretation points in materially different directions |
| Not yet established | Available evidence is insufficient for a responsible conclusion |
A "Replicated" label does not mean a practice will work in your organisation. It means the finding has survived being looked at more than once.
Where practitioner experience fits
I have spent twenty years on the buying side of the practices examined here. I built some of them. That experience is useful for a specific purpose: knowing how a practice behaves once it meets budgets, calendars, managers who are behind on their own work, and incentives nobody designed on purpose.
It is not proof. When I draw on it, I mark it as experience, and I do not ask it to carry weight that only evidence can carry.
Corrections and updates
Every substantial essay shows its publication date and a separately stated evidence-check date. An ordinary formatting edit does not imply a new evidence review.
An essay gets updated when new evidence changes the practical conclusion, and the change is noted rather than quietly made. It does not get updated merely because someone else published something on the subject.
Essays on fast-moving subjects end with a section on what would change the conclusion. That section is the honest version of a confidence level: not a number, but a description of the finding that would make this piece wrong.
If you think something here is wrong, say so. Corrections that come from readers get the same treatment as any other evidence, which is to say they get taken seriously and checked.
AI and editorial accountability
Proofweight is produced by Aditya Rao with substantial AI assistance.
For the first edition, Aditya selected the seven-question investigation, supplied the editorial brief and prior writing, and authorized AI-assisted drafting and implementation. AI assisted research discovery, source inspection, synthesis, drafting and production.
This edition should not be read as a claim that Aditya personally read every cited paper or completed a separate line-by-line review of each generated draft. Source notes identify what was accessible and how it was used. A publisher abstract is not described as a full-text review.
Research findings, organizational accounts, practitioner observations and Proofweight proposals are distinguished in the articles. Illustrative scenarios are labeled and are not presented as Aditya's personal experience. Practical worksheets are original discussion aids, not validated assessments.
Aditya is the named publisher and the contact for editorial questions. A substantive correction should identify the claim, the evidence and how the conclusion changes. The publication's purpose includes revising work when better information becomes available.
AI assistance does not turn a plausible argument into verified evidence. Readers should be able to inspect the reasoning and its limits.
Independence
Nothing is sold on this site.
There is no advertising, no sponsored content, no affiliate arrangement, and no vendor who has paid to be covered or to avoid being covered. No organisation has been given approval over a conclusion.
Where I have a relevant interest in a subject, whether a past employer, a client relationship, or work of my own that an essay bears on, it is disclosed in that essay.
The reason is straightforward. The only thing a publication like this has is whether you believe its conclusions were reached honestly. That is worth more than any revenue it could earn by compromising it.
Getting it wrong
This method is designed to be wrong in public rather than confident in private. If an essay overstates its evidence, misses a study, or reaches a conclusion the sources do not support, that is a failure of the method and worth knowing about.
Write to aditya@proofweight.com.
Reusing the worksheets
The seven worksheets are original Proofweight discussion aids. You may use and adapt them inside your organization with attribution to Proofweight. They are not validated assessments or automatic recommendations about individuals.