Claude Cheat Keys for UN Indicators & Dashboards

Indicator Checks

Four checks, four metadata sections, one join rule, one 13.3.1 card. Everything the carousel promised, with the prompts in full.

7 min read4 sectionsSources opened 2026-10-21. Metadata editions are named at each quote.
Save this page.Every prompt below copies in one click. Dates are the editions quoted.Jump to the first prompt

1. Four checks before a figure goes in a draft

Worked example: indicator 6.1.1, Ghana. Metadata 6.1.1 is the 2024-09-27 edition.

# Check Where it lives What it says for 6.1.1
1 Definition Metadata §2.a Improved source, accessible on premises, available when needed and free from faecal and priority chemical contamination. All three at once
2 Unit Metadata §2.b Percent of population. The country value is the minimum of the three ratios (§4.c)
3 Series and year SDG Global Database SH_H2O_SAFE, Ghana: 40.2 percent for 2022, 42.9 percent for 2024. Both flagged E, estimated. Source: Joint Monitoring Programme (2025)
4 Custodian Metadata §0.g WHO and UNICEF. Compiled by the WHO/UNICEF Joint Monitoring Programme (§3.f)

"Available when needed" means the household reports sufficient water, or water at least 12 hours a day or 4 days a week (§2.a).

The recorded run

Asked on 2026-10-20 at 21:24 UTC, with no tools and no skill: "What is the latest SDG indicator 6.1.1 value for Ghana?"

The answer, verbatim in part: "the latest figure I recall for Ghana is about 38% ... for around 2022." It also said it could not confirm whether a newer update had been released.

Matched by year, the database holds 40.2 for 2022. The latest year in the series is 2024, at 42.9. The record of the run is kept in the engine files. The database values were pulled again on 2026-10-21 and the bytes had not changed.

Project instructions

Paste into Claude
Before you give me any SDG figure, run four checks.

1 Definition  quote metadata 2.a with the edition date
2 Unit        say what it is a share of, from 2.b
3 Series      series code, country, year and the Nature flag
4 Custodian   name the agency in 0.g

A figure with no year is not a figure. Leave it out.
Never answer from memory. No source open? NOT IN SOURCE.
Two years available? Give both. Do not call either one latest.
Last line: what I still have to check myself.

Why a ministry figure can differ from the database: JMP fits a regression over all data points, while some national estimates use the most recent data point, and population figures can differ too (6.1.1 metadata, section 6, Comparability).


2. Four metadata sections that hold the limits

Section Field code What it answers
§2.a Definition and concepts STAT_CONC_DEF What the number counts
§4.b Comment and limitations REC_USE_LIM The custodian's own caveats
§4.f Treatment of missing values IMPUTATION What a blank or a zero means
§4.g Regional aggregations REG_AGG Whether a regional figure exists at all

The same section numbers and field codes appear in the 3.1.1, 4.7.1 and 6.1.1 files.

What they say

Section File Text
§2.a 4.7.1 "It measures government intentions and not necessarily what is implemented in practice in schools and classrooms."
§4.b 4.7.1 The Survey is based on self-reporting by government officials. Countries are asked to send supporting documents
§4.f 4.7.1 Unknown and blank answers are treated as zeros. Over 50 percent of a question missing and the component score is not calculated, it is reported as not available
§4.g 4.7.1 "Regional aggregates are not calculated."
§4.g 6.1.1 Each element is a population-weighted average. The regional value is the minimum of the three elements, or of two if accessibility or availability is missing
§4.f 6.1.1 "Imputed country estimates are not published and only used for aggregation."
§4.g 3.1.1 "Regional aggregations are calculated by aggregating the national-level estimates." Countries are weighted by live births

The stamp is not the same in every file

File Template format version Last updated
4.7.1 1.1 2026-03-27, printed on all 17 pages
13.3.1 1.1 2026-03-27, the 4.7.1 file with its target and indicator lines changed
3.1.1 1.1 2025-11-04
6.1.1 1.0 2024-09-27

Project instructions

Paste into Claude
From the metadata PDF I paste, copy four sections word for word.

2.a  Definition and concepts
4.b  Comment and limitations
4.f  Treatment of missing values
4.g  Regional aggregations

Put the section number in front of each. Do not summarise.
A section with separate parts (Survey, Greening): keep both,
labelled. First line: the Last updated date and the template
format version printed on page 1.
A section that is empty or not applicable: show it as it is.

When the four sections do not answer it, SDG Data Contacts lists a focal point per indicator for questions on definitions, methods of computation and data (page last updated 22 September 2026).


3. Same 42 days, two different counts

Metadata 3.1.1, 2025-11-04 edition.

Term Window Cause
Maternal death While pregnant or within 42 days of termination of pregnancy Related to or aggravated by the pregnancy or its management. Not unintentional or incidental causes (§2.a)
Pregnancy-related death The same 42 days Irrespective of the cause of death (§2.a)

The custodian does not use pregnancy-related data as they are. Incidental or accidental deaths are 10 percent of pregnancy-related deaths (excluding HIV-related deaths) in sub-Saharan Africa and 15 percent in other low- and middle-income countries, and the estimates adjust for this (§4.e).

Three measures, three denominators (§4.b)

Measure Divided by
MMR 100,000 live births in the same period
MMRate Person-years lived by women of reproductive age
PM All deaths among women aged 15 to 49

The SDG API carries the unit code PER_100000_LIVE_BIRTHS on every MMR record. The rule for which deaths count is only in §2.a.

Project instructions

Paste into Claude
Before I put two maternal death figures in one series,
read metadata 3.1.1 section 2.a and tag each source:

  MATERNAL           cause related to or aggravated by pregnancy
  PREGNANCY-RELATED  any cause, same window
  UNCLEAR            the source does not say

Two different tags in one series: write JOIN BLOCKED, name rows.
Check the base. MMR is per 100,000 live births. Anything else,
say so. Do not convert one definition into the other.
Last line: the question for the WHO focal point.

The custodian for 3.1.1 is WHO, Department of Sexual and Reproductive Health and Research (§0.g).


4. A 13.3.1 card, built before the value

13.3.1 has a file of its own, Metadata-13-03-01, and it is the 4.7.1 file. Only the target and indicator lines differ, and its §0.a still reads Goal 4. Both are the 2026-03-27 edition, so every section below reads the same in either file.

Line What the metadata says Section
Codes 4.7.1, 12.8.1 and 13.3.1. All eight series carry the tag [4.7.1,12.8.1,13.3.1] §0.d, §0.f
Measures Government intentions, as reported by officials §2.a
Where Greening is captured under the Curricula component §4.a
Built from 13 keywords: 4 environment and sustainability, 5 climate change, 4 biodiversity, in grade 3, 6 and 9 science and social science curricula and national curriculum frameworks §3.a
Score One Greening Curriculum Index, environment and sustainability weighted 50 percent, climate change and biodiversity each weighted at 25 percent. The four Survey components are not combined §4.c
Limit "As it is based on counts of keywords, it does not capture how these keywords are used." §4.b
Partial Three of four document types: the value carries a Partial Data qualifier §4.f
Regional "Regional aggregates are not calculated." §4.g
Years Greening data refer to 2023-2024. Collection is expected to be annual §3.c, §5

Project instructions

Paste into Claude
Build an indicator card for 13.3.1 from the metadata I paste.
Metadata first. Do not read any value until the card is done.

1 Codes     the code and every code it shares, from 0.f
2 Measures  quote the 2.a sentence that limits the claim
3 Score     which score I cite, and its rule from 4.c
4 Limit     quote 4.b for that same score
5 Regional  quote 4.g. If none, write NO REGIONAL FIGURE
6 Edition   the Last updated date from page 1

An outcome sentence the card cannot support: OVERCLAIM, and why.

Sources