Methodology

What this page covers

ACW's research lines are recurring efforts that turn public records into a versioned dataset and a briefing. This page states the standard they follow. The post How I make this describes how an individual article is researched, written and checked.

Five promises

• Every figure traces to a public source that ACW retrieved on a stated date.
• Nothing is contacted, surveyed or interviewed in a research line. Corrections arrive inbound only.
• Every published number can be rebuilt from stored inputs by a script.
• A dataset states what it cannot say before it states what it says.
• A note the owner has seen but not confirmed against a public source never enters a public dataset.

Sources and what each can establish

• Official record. A regulator, statistics office or government dataset. It establishes a fact about the record itself: what was reported or counted. It is the strongest class.
• Filing. A company filing to a securities regulator. It establishes what the company told the regulator, in the form it filed. A filing is a statement under legal duty, not an audit of operations.
• Standard or paper. A standards body publication or a peer-reviewed paper. It establishes a definition, a requirement, or a result reported by the authors. A paper's result is reported by the authors, not shown, until it is replicated.
• Company statement. A press release, web page, presentation or product sheet. It establishes only that the company said it, and is worded "says" or "states". It never stands alone for a performance, scale or deployment fact.
• Secondary report. News or analyst reporting. It is a lead only. A fact stands once it is traced to a primary source, which is the one cited.

When two sources of different classes disagree, the higher class decides and the lower one is recorded as contradicted. When two sources of the same class disagree, the disagreement is shown, not resolved. A source that states no reuse terms is quoted by locator, not copied.

Observation, finding, interpretation

• Observation. One recorded fact with its source and retrieval date.
• Finding. A stated comparison of observations that holds under a method written before it was computed.
• Interpretation. ACW's reading of what a finding means for a decision, labelled as a reading.
• Hypothesis. A claim the evidence does not yet support, labelled as a hypothesis.
• Recommendation. What ACW suggests a reader consider, labelled as a recommendation.

An observation becomes a finding only when the comparison and its denominator were written down before the numbers were computed, the same method covers every period compared, the denominator is stated (with none, the finding is a count and is worded as a count), it can be reproduced from the stored inputs, a human relevance review approved it, and the caveats that bear on it sit next to it.

Some things never count as a finding: a ranking of organisations, a statement about intent, quality or capability that the source does not make, and anything resting on a company statement or a secondary report alone.

Review

The author reviews each version before it is published: Shawn Sehy writes and publishes ACW, and this page names no other reviewer. The review checks that the research bears on a decision a leader has to make, that the rows follow the rules as written, and that the classification holds. If a second reviewer takes part, this page will name them. A dataset that bears on no decision is not published, however easy it was to build.

Corrections

ACW corrects the piece, states the change, records the date, and withdraws a claim it cannot support rather than softening it. A wrong published value is corrected in a new version that shows the earlier value and the date, and the earlier version stays retrievable. Nothing is rewritten silently. To report an error, use the inquiry form; ACW checks it against the source and answers on the same page. ACW does not pre-clear findings with the organisations it names and does not seek comment from them: accuracy rests on the sources and the review.

Independence and sponsorship

Sponsored work is offered selectively and may be paid placement, but conclusions are never paid opinion. A sponsor or paying client may not choose a line, a source, a rule, a classification, a sample or a conclusion, and may not see a dataset before it is approved for the public. A relationship with an organisation that appears as a data subject is disclosed on the dataset page and in the briefing, and that organisation's rows are processed by the same published rule. A dataset is ACW's own research: it is never presented as client work or as validated by an organisation it names. See how ACW works with sponsors.

Use of AI tools

Primary sources are read by the author. Drafts are in the author's own words, with tools used only for cleanup and the ACW content engine as an assistant.

Who is responsible

ACW is written and published by Shawn Sehy. Read the research to see the method applied.