Claude Cheat Keys for UN Reporting & Briefing

Check that every statistic in a progress report states its evidence

Before a draft progress report is cleared, each statistic should state its data year and indicator, and its figures should agree across sections. I asked Claude to run this check on a stored copy of the report, with no web search. Here is what it found and how the skill works.

3 min read4 sectionsSources opened 2026-10-21.
Save this page.Every prompt below copies in one click. Dates are the editions quoted.Jump to the first prompt

The work moment

You are clearing a draft progress report. Every statistic needs a data year and an indicator, and the figures must agree from one section to the next.

The report itself gives its sourcing only in general terms. It was prepared "using input from more than 50 international and regional organizations." [p 1]

It sets one frame for its data years: "The remainder of the present section draws on the most recently available data from 2022 to 2025 to provide an overview of each Goal, with a focus on selected targets for which new data are available." [p 6]

For detail it points elsewhere: "The indicator framework, the statistical annex to the present report and the Global Sustainable Development Goal Indicators Database are available at https://unstats.un.org/sdgs ." [p 6]

What the test showed

Claude ran with the skill and no web search. It said: "I used only the stored copy you provided and did no web search."

On sourcing, it wrote: "None of the 352 claims names the agency or dataset it comes from."

It checked the arithmetic. The report says "In mathematics, proficiency fell from 41% to 36% (down 6 points), with similar declines across income grou ps." [p 10] Claude wrote: "41 − 36 = 5 points, not 6."

It compared sections. The report says "Nearly half (49%) are making only marginal progress, while 15% have regressed below the 2015 baseline." [p 2] Later it says "By contrast, there has been only marginal progress (32 %) or no progress (17%) on 49% of targets." [p 4] Claude listed the two together as a place where the report's own wording does not agree.

It did the same for poverty. The report says "Globally, the nowcasted estimate projects that extreme poverty will be 10.0% in 2026, down only 3 percentage points from 2015." [p 6] Claude noted that this "implies about 13.0%", against "falling from nearly 14% in 2015" on [p 4].

It also flagged the copy itself: "The stored copy is cut off." It "ends mid-word at [p 21] "Accelerati"."

The steps of the skill

  1. Attach the report and paste the prompt. Ask for no web search.
  2. Claude lists each statistic, trend and comparison claim. The skill asks it to pause so you can confirm that list before the check.
  3. Each claim is checked for four things:
    • Year: a calendar year or a dated period. "Phrases like "past decade", "now", "recent years" or "post-pandemic" don't count."
    • Measure: a number, or a clearly defined indicator.
    • Baseline: for trend or comparison claims, a starting point with its value, or the size of the change.
    • Evidence: for claims that one thing caused another, evidence for the link.
  4. Claims that miss a check go on the evidence gap list, each with its quote and page marker.
  5. Places where the report's own figures or wording disagree go on a separate list.
  6. Gaps that apply to the whole document, such as missing sources or a cut-off copy, are reported once at the top.

Copy-paste prompt

Paste into Claude
Please run an evidence gap tracking check on the 2026 Secretary-General SDG progress report attached below, using only this document and no web search. List each statistic, trend, or comparative claim and mark whether the report itself states the evidence behind it, such as the data year, source, or indicator. Quote the document word for word with the page marker for every quote you use, and list every claim marked unsupported as an evidence gap.

Sources