Editorial

How these pieces are made

Last updated .

Scriborium reads the research behind one writing question at a time. These are the rules each piece is held to, so you can tell what the studies found from what I'm arguing.

Reading the source, not the summary

I read the study itself, including the methods and limitations, before I cite it for a finding. When I could only read an abstract, or could not open the source at all, the piece says so in the source list, and I cite it only for what a named secondary source says about it. A vendor page is cited for what the vendor claims. What a study found is cited to the study.

Grading the evidence

Each piece sets out the main studies: the design, which students were studied and at what grade, and what was found. I look at whether students were randomly assigned, how large the sample was, how writing quality was scored and by whom, how long afterward it was measured, and, where the paper reports it, who paid for the work. Where a study gives an effect size or an interval, I give it, and I do not round a modest or uncertain result into a confident one. Meta-analyses are read for what they pooled, since combining unlike studies can hide the differences that matter. Preprints are labelled as not peer reviewed.

Words I will not stretch

I use causal verbs, such as improves, raises and causes, only when the design can support them. Otherwise it is was associated with or was higher in. Better scores on a surface feature, such as spelling or grammar, are not better arguments, and a piece says which one a study measured. Findings from university or second-language learners are not carried over to younger students without saying so.

Student work and examples

This site does not publish real student writing. Where a piece needs an example of a draft or of feedback, it is invented for the purpose and labelled as invented.

How I search

Each piece includes a short method note: where I searched, with which terms, on what date, and what I left out and why. The search starts from the claim's origin and the claimant's own references, uses open research indexes such as ERIC, Crossref, OpenAlex and Europe PMC, and follows citations where it can. If a study that matters was missed, that is a correction.

Citation checks

Before a piece publishes, a script opens every linked source and checks that it loads. It resolves every DOI on Crossref and compares the title and year with what I cited. It checks each paper's retraction status on OpenAlex. It also searches the source text for every passage I quote, word for word. The results are printed at the foot of the piece with the date. These checks show that a source exists and says what is quoted. They do not show the source is right.

Claims, ledger and independent check

Each factual sentence is tied to a passage in a source the script opened, and a model that did not write the draft checks every claim against its passage alone. AI reviewers with different briefs then read the draft. A piece that fails any check does not publish; its date moves. These are model judgements, and a pass is not proof. A sample of published pieces is audited afterward.

Corrections

If something is wrong, write to [email protected]. A correction goes at the foot of the piece with its date and what changed, and is logged on the corrections page. I don't edit a mistake out quietly.

Independence and disclosure

Scriborium is my own project. Nothing on it is sponsored, and it is not client work. I build websites and web strategy for a living at thisismyurl.com. Where a piece touches something I have built, sold, used or been paid by, the byline note says so.