docs(#603): population-statement survey — all 111 waves, verbatim

[ligon]

Assembles the per-country population-statement sweep into one reviewable evidence document:
slurm_logs/POPULATION_STATEMENTS_2026-07-21.org.

It records, for every wave the library ships, what that survey's own documentation says its sample represents — verbatim, with exclusions called out separately, and with the gaps left as gaps.

Headline counts

finding	waves
total waves swept (38 country directories)	111
a documented statement of some kind (source_type != not-found)	104
no statement of any kind (source_type = not-found)	7
plus: records NOT FOUND for a universe, fallback coverage claim only	3
…so, waves where no document names the population	10
statement lives in a file this repo ships	63
statement lives only in catalog metadata or a catalog-hosted PDF	41
extractor confidence high / medium / low	84 / 25 / 2

The standing rule it obeys

• Quotes are verbatim — sic spellings ("was been designed", "Bihar(Southern", "abovefor"), source typography ("Education :", "Agriculture:holders") and the sources' own internal contradictions are preserved. Every PDF-extraction repair (rejoined split words, collapsed justification spaces, a decoded glyph-substituted numeral, reconstructed bidi Arabic) is declared in that wave's notes.
• Gaps are recorded as gaps — every NOT FOUND carries the full search record, so the claim is falsifiable rather than permanent-by-default.
• Nothing is inferred from the data — no .dta was opened. Where inheriting a sibling wave's universe would have been defensible (Mali / Niger / Senegal 2021-22 EHCVM-2 panels), it was declined, and the entry says so.

Structure

1. What this is, how it was gathered, and the rule above.
2. Summary table — country, wave, source_type, confidence, and a short universe tag marked throughout as an editorial reading, not a quote.
3. The verbatim record per wave — population statement, translation clearly separated where present, exclusions in their own block for every wave, source file + locator, and the extractor's evidence notes.
4. The gaps, unsoftened, with what was searched for each.
5. What the evidence suggests — patterns observed, disagreements flagged and not resolved.

What the evidence shows

• 41 of 111 waves' population record does not exist inside the library. It lives on microdata.worldbank.org (and, for two Ghana panel waves, catalog.ihsn.org), and three waves already have a <universe> element that is present and empty. A record built by linking would track a moving target.
• A field literally labelled "Universe" exists for only 47 of 111 waves. For the other 64 the strongest text is coverage prose, sampling-procedure prose, or a report methodology chapter. 65 waves assert national scope in some form — but only 37 actually name the population.
• The single most repeated sentence in the corpus is boilerplate. "The survey covered all de jure households excluding prisons, hospitals, military barracks, and school dormitories." appears verbatim in 15 waves — and not only in the WAEMU/EHCVM countries but also Ethiopia 2018-19 / 2021-22 and Nigeria 2015-16 / 2018-19 / 2023-24. It attests a WB NADA metadata template, not a statistical office's claim about that survey.
• 19 waves subtract a named region by design (Tirana, Kosovo and Metohija, Likoma, Kidal, Tigray, Arlit, rural Borno, FATA, Ayacucho/Apurimac/Huancavelica, eleven occupied raions, nine zones of Afar and Somali…). Reasons split into administrative, cost/logistics, security, and frame-construction — and the reason matters. Malawi excludes Likoma through IHS3 and includes it from IHS4: a real universe break mid-panel.
• Exactly one wave declares an urban exclusion (Ethiopia ESS1, "rural and small town areas"). Three more are rural in fact but never say so (China 1995-97, India 1997-98, EthiopiaRHS) — the sweep declined to write it for them.
• Specialization mostly appears inside waves, not as whole waves. Only Liberia 2018-19 (forest-proximate EAs) is a specialized wave. But Serbia and Montenegro 2002/2003 each carry two samples with two universes (MOP recipients; Roma settlements), Azerbaijan 1995 has three sub-populations, Iraq 2012 administers three modules to a one-third subsample, Malawi 2016-17/2019-20 read a national cross-section and a panel into one library wave, and GLSS3 enumerates per-module sub-universes explicitly.
• "Declared universe" and "what the weights represent" come apart in real documents — Ethiopia ESPS-5 (weights represent the population excluding Tigray), Nigeria 2018-19 ("representative of areas of Nigeria that were accessible during 2018/19"), Liberia (stratification error changed the represented population after fieldwork), Tanzania 2019-20 (national universe, Extended-Panel cohort).
• Vocabulary consistency: partly — the exclusion axis far more than the inclusion axis. The unit of the universe is unstable even within one country's series (Panama shifts from dwellings to resident population between waves); several universes are circular; and three surveys' documentation denies a well-defined universe outright (China: "not a rigorous random sample drawn from a well-defined population"; EthiopiaRHS: "these data are not nationally representative"). The library currently has no way to express that.

Deliberately not done

No schema, no YAML key, no decision about whether Feature() should refuse to pool waves with different universes. §5.7 lists 17 unresolved disagreements, including one where my own editorial tag contradicts the wave's own UNIVERSE field (Mali 2018-19) — flagged, not adjudicated. Per #601's note, a proposal that encodes a judgement about what the library will pool on a user's behalf is a direction decision for Ethan.

Refs #603, #601.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Su5JKX3wKChyfMAdXrdCTr