A letter is the start.
The evidence is the point.

First, define the claim.

A paper can contain a sound measurement and an unsupported explanation. We identify the exact proposition under review before assigning a score: what was observed, what was inferred, and how broadly the authors presented it. A failed interpretation does not automatically invalidate every measurement in the paper.

Our reviews assess the claim against the evidence available at the review date. When an original version differs from a correction or final publication, we identify the version and chronology. A grade is an editorial assessment—not a statistical probability, journal impact factor, misconduct finding, or measure of a person's worth.

Five criteria. One published calculation.

Each criterion is scored from 0 to 5. The total is Σ (criterion score ÷ 5 × weight), rounded to the nearest whole number. Missing evaluation is recorded as ungraded, not as a zero.

We give claim support the largest weight because a well-presented claim can still be wrong. Transparent corrections can earn credit without restoring an overturned claim's support score. These weights are an explicit editorial choice, not a scientifically validated universal measure.

What the 0–5 scores mean

• 5 — Strong: compelling relevant support or exemplary performance on the criterion, with no material unresolved problem identified in the sources reviewed.
• 4 — Substantial: strong evidence with a bounded limitation or a remaining uncertainty that does not overturn the central conclusion.
• 3 — Mixed: meaningful support with important unresolved limitations or conflicting evidence.
• 2 — Weak: a serious limitation undermines the criterion; the central inference needs considerably more support.
• 1 — Very weak: a major failure or a thin evidential basis leaves little support for the stated claim.
• 0 — Not supported on this criterion: the relevant claim is contradicted, or the evaluated requirement has no demonstrated support. The reason must identify which applies.

An absence of independent replication is not proof of error. We distinguish missing evidence from a failed replication and explain field-specific checks. Astronomy cannot rerun a past merger, for example, but independent detectors, waveform checks, and subsequent events can test the interpretation.

Reading A through F

Thresholds simplify communication. A score of 89 is not meaningfully different from 90 without understanding the reasons. Read the criterion scores and sources before comparing letters.

How researcher grades work

A researcher profile aggregates only papers listed on that profile. An influential claim central to a reviewed contribution has weight 2; a narrower finding has weight 1. We compute the weighted mean of evaluated paper scores, round once, and apply the same letter boundaries. Unevaluated reviews do not enter the average.

“Influential” identifies a claim's role in the reviewed work, not a claim that we have catalogued an entire career. Every profile shows its paper count and coverage. At launch, the index is deliberately small. A one-paper record is a one-paper record, even when the author is famous.

We attribute a result to the named authors or collaboration in its primary publication. Large-team entries are labeled as collaborations. Authorship does not establish equal responsibility for every measurement or analytical decision, and a shared paper grade cannot resolve individual contributions.

The index is ordered by reviewed-claim score, with names breaking ties. It is not a ranking of all scientists, and it does not estimate a career-wide mistake rate from a selected sample. We do not use reputation, citation counts, institutional prestige, or an allegation of bad motives as scoring inputs.

What counts as a documented mistake?

Examples include a demonstrated instrument-calibration error, an incorrect computation, a control that fails to distinguish the proposed mechanism, or a causal interpretation not justified by the design. The review must connect the problem to a primary source and explain its consequence for the specific claim.

A different scientific opinion is not automatically an error. Neither are a null result, a correction, a retraction, or an unsuccessful replication interchangeable with misconduct. Retraction and misconduct findings are attributed only when established by the relevant primary record.

Selection, review, and revision

1. Choose an influential or informative claim and identify its exact publication and version.
2. Read the primary paper and relevant subsequent research, corrections, and collaboration or institutional reports.
3. State the claim, supporting evidence, strongest objections, and unresolved scope limits.
4. Score all five criteria with written reasons and linked sources.
5. Publish the review date and keep a revision record when the analysis changes.

Reviews are editorial syntheses and may use software-assisted research and drafting. They are not represented as independent expert peer review. Source links let readers check the reasoning directly. There are no paid score upgrades, public popularity votes, or hidden reputation multipliers.

New evidence can improve or lower a grade. Authors and readers can submit corrections and responses through the same correction process. We examine the evidence rather than charge a fee for consideration.