Corrections and material changes

What happens after a report

We acknowledge a correction request, preserve the challenged version, review the cited evidence, and distinguish factual errors from reasonable disagreements in interpretation. If a correction is warranted, the page receives a dated note describing what changed and why.

Correction levels

• Minor: typography, formatting, or a link repair that does not change meaning.
• Material: a fact, calculation, label, chart, method description, or interpretation that could alter understanding.
• Withdrawal: evidence or privacy problems so serious that continued publication would mislead or harm.

September 23, 2026 — site focus and reporting template

• Templates focused: replaced generic ad and email examples with editable questionnaires, a decision brief and a results note. Each questionnaire includes its answer choices and can be loaded into the wording checker without sending text to a server.
• Recommendations narrowed: we removed the broad software directory and its affiliate link. The two website-build checks were also moved to our internal record because they did not help readers design or review a survey.
• Reporting example: the reporting checklist now uses fill-in fields instead of invented field dates and a fictional participant count. This makes the boundary between a template and an actual study record clearer.
• Methods page: it now explains the software checks and teaching examples actually available on the site. We removed the lengthy description of future fieldwork. No recruitment has opened.
• Guide revisions: five guides have shorter, task-specific revision checks in place of repeated advice about using AI. The calculations, cited methods and existing section links remain intact. Update dates identify the revised editions.

September 23, 2026 — review dates, file labels and documented tool limits

• Review-date scope: About and editorial standards previously described the “current” library as approved on September 9, although later edits had been released. That approval now explicitly refers to the September 9 edition. Later update dates do not imply a new human or independent expert review.
• File categories: the catalog introduced “two” fictional decision files but displayed a third, unrelated scale template beside them. The template now sits with the templates; scoring examples and actual software observations have separate labels.
• Checker limitations: the new twelve-case test note records outputs and missed patterns in rules version 1.0. The wording rules have not changed in this release. A lack of alerts still does not establish that a question is unbiased.
• Worked examples: the scale guide now includes reproducible scoring counts, and the concept guide includes a decision walkthrough. Both are clearly labeled constructed examples, not consumer results.

September 9, 2026 — calculation, template, and review disclosures

These corrections address errors in the earlier version of the library. They do not revise empirical findings: MTB has not published collected consumer results.

• Comparison calculator: the earlier normal-approximation calculation could display a zero-width interval for extreme counts, including zero outcomes in both groups. The message-comparison tool now withholds its interval, p-value, and significance conclusion when either group has fewer than ten outcomes or non-outcomes. Observed rates and differences remain visible. Withheld inference does not establish equality or zero uncertainty.
• Claim-register CSV: preset claim categories appeared in the implied-meaning column. The corrected claim register places them in claim_type and leaves the implied-meaning field for the reviewer. If you saved the earlier template, check your column mappings before continuing to use it.
• Question-order worksheet: the question-order audit makes consent before research exposure explicit. This clarifies the required sequence; it does not open recruitment or provide a participant-consent process.
• Editorial claims: earlier About, editorial-policy, and footer wording implied completed human review before sign-off. The correction disclosed AI assistance and distinguished editorial accountability from completed review. The owner subsequently approved the September 9 edition; the September 23 entry above clarifies the limits of that approval.

If an earlier calculator result or worksheet influenced a decision, revisit it using the corrected version and the stated limitations. Report remaining errors through the editorial contact.