From 53a2b2db0b9d012044013dc98dcfd61df7fca341 Mon Sep 17 00:00:00 2001 From: John Stachurski Date: Tue, 4 Aug 2026 06:31:47 +1000 Subject: [PATCH 1/3] Migrate japan_population_by_age, and stop the front page going stale MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The last dataset committed into lecture-python-intro by #790 moves here, and the dashboard narrative that described it as newly arrived is brought up to date. The lecture repo held an on-demand e-Stat export: 20 KB of merged headers and Japanese column labels, read with sheet_name='第１表', skiprows=10 and usecols=[14], and carrying no record of where it came from. Rather than migrate it in that state, the same table was located at its publisher, the Statistics Bureau of Japan, where it has a stable URL, and rebuilt from there. The committed builder flattens the printed two-block layout into one row per age and keeps both the total and Japanese-national counts. The japanese_population column equals, value for value, the column prob_dist reads today, so the repoint cannot change lecture output. What changes is that the data now has a source, a citation, a licence and a builder. The validate stage earned its place while writing this: per-age figures sum to 123,807 thousand against a published total of 123,802, because every figure including the total is rounded to the nearest thousand independently. The check now allows that slack explicitly rather than either failing or being dropped. Dashboard: what_changed() quoted a hard-coded "4 datasets migrated" beside a tile computing the live figure. It now takes the count from migration state, so a landing wave cannot leave a stale number on the front page, and its closing card no longer flags these two datasets as newly arrived under the old conventions -- both have since moved, with their builders recovered. Part of QuantEcon/meta#336 Co-Authored-By: Claude Opus 5 (1M context) --- CATALOG.md | 3 +- lectures/japan_population_by_age.csv | 102 +++++++++++++++++++++++ lectures/japan_population_by_age.csv.yml | 89 ++++++++++++++++++++ migration.yml | 14 ++++ scripts/audit_annotations.yml | 5 -- scripts/japan_population_by_age.py | 102 +++++++++++++++++++++++ scripts/render_audit.py | 37 ++++---- 7 files changed, 332 insertions(+), 20 deletions(-) create mode 100644 lectures/japan_population_by_age.csv create mode 100644 lectures/japan_population_by_age.csv.yml create mode 100644 scripts/japan_population_by_age.py diff --git a/CATALOG.md b/CATALOG.md index 5bb399a..7ebd110 100644 --- a/CATALOG.md +++ b/CATALOG.md @@ -6,7 +6,7 @@ The migrated-dataset registry, **auto-generated** from the sidecar manifests (`lectures/*.yml`). Do not edit by hand — run `python scripts/build_catalog.py`. A dataset appears here once it has a manifest; files not yet migrated are tracked in [PLAN.md](PLAN.md) Phase 9. -**9 datasets migrated** · 2.4 MB total · 8 permitted / 1 restricted redistribution +**10 datasets migrated** · 2.4 MB total · 9 permitted / 1 restricted redistribution | Dataset | Class | Source | Licence | Redist. | Integrity | Builder | Size | Used by | | --- | --- | --- | --- | --- | --- | --- | --- | --- | @@ -16,6 +16,7 @@ The migrated-dataset registry, **auto-generated** from the sidecar manifests (`l | [**epl_match_goals.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/epl_match_goals.csv)
English Premier League — full-time scores, 2015-16 to 2024-25 | constructed | [openfootball / football.json](https://github.com/openfootball/football.json) | Public domain | ✅ permitted | ✅ verified | ✅ committed | 203.2 KB | [lecture-python-intro · fitting_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/fitting_distributions.md) | | [**japan_deaths_by_age.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/japan_deaths_by_age.csv)
Japan — deaths by single year of age, 2023 | constructed | [United Nations, Department of Economic and Social Affairs, Population Division — World Population Prospects 2024](https://population.un.org/wpp/downloads) | CC BY 3.0 IGO | ✅ permitted | ✅ verified | ✅ committed | 1.7 KB | [lecture-python-intro · observed_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/observed_distributions.md)
[lecture-python-intro · fitting_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/fitting_distributions.md) | | [**japan_earthquakes.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/japan_earthquakes.csv)
Japan region — earthquakes of magnitude 5 and above, 2000-2024 | constructed | [Advanced National Seismic System (ANSS) Comprehensive Earthquake Catalog (ComCat), US Geological Survey](https://earthquake.usgs.gov/earthquakes/search/) | US Government work — public domain | ✅ permitted | ✅ verified | ✅ committed | 172.8 KB | [lecture-python-intro · fitting_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/fitting_distributions.md) | +| [**japan_population_by_age.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/japan_population_by_age.csv)
Japan — population by single year of age, 2024 | constructed | [Population Estimates, Statistics Bureau of Japan, Ministry of Internal Affairs and Communications](https://www.stat.go.jp/english/data/jinsui/index.html) | Japan Statistics Bureau terms of use | ✅ permitted | ✅ verified | ✅ committed | 1.3 KB | [lecture-python-intro · prob_dist.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/prob_dist.md) | | [**lingcod_msy_recovery.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/lingcod_msy_recovery.csv)
Pacific Coast lingcod — biomass and fishing pressure relative to MSY | constructed | [RAM Legacy Stock Assessment Database](https://www.ramlegacy.org/) | CC BY 4.0 | ✅ permitted | ⚠️ unverifiable | ⚠️ unrecovered | 2.3 KB | [lecture-python-intro · msy_fishery.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/msy_fishery.md) | | [**realwage.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/realwage.csv)
OECD real minimum wages — 32 countries, 2006–2016 | constructed | [OECD — Real minimum wages (RMW)](https://stats.oecd.org/Index.aspx?DataSetCode=RMW) | CC BY 4.0 | ✅ permitted | ⚠️ unverifiable | ⚠️ unrecovered | 118.7 KB | [lecture-python-programming · pandas_panel.md](https://github.com/QuantEcon/lecture-python-programming/blob/main/lectures/pandas_panel.md)
[lecture-python.myst · pandas_panel.md](https://github.com/QuantEcon/lecture-python.myst/blob/main/lectures/pandas_panel.md) | | [**us_adult_heights.csv**](https://github.com/QuantEcon/data-lectures/raw/main/lectures/us_adult_heights.csv)
United States — adult standing height by sex, NHANES 2015-2018 | constructed | [National Health and Nutrition Examination Survey (NHANES), US Centers for Disease Control and Prevention, National Center for Health Statistics](https://www.cdc.gov/nchs/nhanes/index.htm) | US Government work — public domain | ✅ permitted | ✅ verified | ✅ committed | 123.1 KB | [lecture-python-intro · prob_dist.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/prob_dist.md)
[lecture-python-intro · observed_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/observed_distributions.md)
[lecture-python-intro · fitting_distributions.md](https://github.com/QuantEcon/lecture-python-intro/blob/main/lectures/fitting_distributions.md) | diff --git a/lectures/japan_population_by_age.csv b/lectures/japan_population_by_age.csv new file mode 100644 index 0000000..1bb379a --- /dev/null +++ b/lectures/japan_population_by_age.csv @@ -0,0 +1,102 @@ +age,total_population,japanese_population +0,716,695 +1,757,738 +2,797,780 +3,828,810 +4,836,818 +5,872,853 +6,915,897 +7,940,921 +8,980,961 +9,1004,987 +10,1003,985 +11,1028,1011 +12,1032,1015 +13,1057,1041 +14,1066,1051 +15,1071,1057 +16,1091,1076 +17,1085,1070 +18,1089,1062 +19,1121,1061 +20,1189,1102 +21,1207,1109 +22,1261,1148 +23,1297,1163 +24,1307,1169 +25,1295,1168 +26,1312,1184 +27,1304,1180 +28,1298,1181 +29,1309,1196 +30,1300,1196 +31,1268,1167 +32,1275,1185 +33,1265,1180 +34,1288,1201 +35,1308,1229 +36,1345,1274 +37,1378,1311 +38,1396,1335 +39,1445,1388 +40,1493,1438 +41,1512,1457 +42,1512,1456 +43,1524,1475 +44,1590,1543 +45,1629,1583 +46,1689,1646 +47,1732,1692 +48,1812,1773 +49,1885,1848 +50,1990,1954 +51,2023,1987 +52,1985,1949 +53,1926,1890 +54,1864,1828 +55,1833,1799 +56,1787,1753 +57,1778,1750 +58,1387,1359 +59,1705,1678 +60,1595,1568 +61,1551,1525 +62,1497,1475 +63,1463,1443 +64,1467,1448 +65,1487,1470 +66,1440,1424 +67,1393,1378 +68,1456,1441 +69,1492,1478 +70,1482,1470 +71,1560,1548 +72,1633,1622 +73,1710,1700 +74,1813,1804 +75,1962,1952 +76,1911,1902 +77,1784,1775 +78,1088,1081 +79,1141,1135 +80,1360,1354 +81,1279,1273 +82,1273,1268 +83,1191,1186 +84,1035,1031 +85,852,849 +86,856,853 +87,821,818 +88,758,755 +89,657,655 +90,558,556 +91,494,493 +92,419,418 +93,344,344 +94,273,273 +95,218,218 +96,171,171 +97,112,112 +98,81,81 +99,52,52 +100,87,87 diff --git a/lectures/japan_population_by_age.csv.yml b/lectures/japan_population_by_age.csv.yml new file mode 100644 index 0000000..789514e --- /dev/null +++ b/lectures/japan_population_by_age.csv.yml @@ -0,0 +1,89 @@ +# Manifest for japan_population_by_age.csv. + +filename: japan_population_by_age.csv +title: Japan — population by single year of age, 2024 +description: > + Population of Japan by single year of age from 0 to "100 and over", as of + 1 October 2024, for the total population and for Japanese nationals. Counts + are in thousands, as published. Used as a first concrete example of a + probability mass function. + +# Constructed: the upstream workbook is Table 1 of the Population Estimates, +# laid out for printing -- ages run down a left-hand block and continue in a +# second block to its right, under two levels of merged headers. Flattening +# that into one row per age is our processing. +class: constructed + +source: + name: > + Population Estimates, Statistics Bureau of Japan, Ministry of Internal + Affairs and Communications + series: > + Table 1 — Population by age (single years) and sex, and sex ratio; total + population and Japanese population, as of 1 October 2024 + url: https://www.stat.go.jp/english/data/jinsui/index.html + file_url: https://www.stat.go.jp/data/jinsui/2024np/zuhyou/05k2024-1.xlsx + citation: > + Statistics Bureau, Ministry of Internal Affairs and Communications, Japan. + Population Estimates, 1 October 2024. + +license: + name: Japan Statistics Bureau terms of use + url: https://www.stat.go.jp/english/index/copyright.html + redistribution: permitted + verified: 2026-08-04 + # The Statistics Bureau permits reproduction of its published statistics with + # the source indicated, which the citation above does. Its terms follow the + # Japanese government standard open-data licence, compatible with CC BY. + +retrieved: 2026-08-04 +maintainer: QuantEcon + +# --------------------------------------------------------------------------- +# Integrity +# --------------------------------------------------------------------------- +# Migration check (not a manifest field — a one-time repoint gate): the +# japanese_population column of this file equals, value for value, the column +# that lecture-python-intro's prob_dist.md previously read out of an e-Stat +# export committed at +# lectures/_static/lecture_specific/prob_dist/japan_population_by_age.xlsx, +# so the repoint cannot change lecture output. +# +# That export was an on-demand e-Stat download with no recorded provenance -- +# the `builder_status: unrecovered` case AGENTS.md describes. Rather than +# migrate it in that state, the same table was located at its publisher, the +# Statistics Bureau, where it has a stable URL, and rebuilt from there. The +# published figures are identical, so the lecture is unaffected and the data +# gains a source, a citation and a builder. + +integrity: + sha256: 5bf07638cb75fa99e9dffb7ed2e272982a938384d49c9e1c4065117d696b81d8 + upstream: + status: verified + date: 2026-08-04 + against: 05k2024-1.xlsx + note: > + Written by re-running the committed builder against the Statistics + Bureau workbook on the retrieval date. The 2024 estimates are a closed + release, so re-running should reproduce this hash. + +schema: + format: csv + columns: + - {name: age, dtype: int64, description: single year of age; 100 means "100 and over"} + - {name: total_population, dtype: int64, description: total population at this age, in thousands} + - {name: japanese_population, dtype: int64, description: Japanese-national population at this age, in thousands} + row_count_floor: 101 + date_range: {start: 2024, end: 2024} + known_nulls: {} + # Every figure is rounded to the nearest thousand independently, so the + # per-age counts sum to slightly more than the published totals: by 5 + # thousand for the total population and 3 thousand for the Japanese + # population. The builder allows for this and asserts the sums are within 60. + +consumers: + - repo: QuantEcon/lecture-python-intro + file: lectures/prob_dist.md + +builder: scripts/japan_population_by_age.py +builder_status: committed diff --git a/migration.yml b/migration.yml index 52d2b6e..725eb60 100644 --- a/migration.yml +++ b/migration.yml @@ -155,6 +155,20 @@ datasets: date: 2026-08-04 cutover: null + japan_population_by_age.csv: + pilot: P5 + status: landed + # Migrated out of lecture-python-intro, which held an e-Stat export with no + # recorded provenance. Rebuilt from the same table at its publisher, the + # Statistics Bureau, whose copy has a stable URL; the figures are identical. + # Flip to `repointed` when the lecture PR merges. + prior_pattern: local-path + landed: + pr: QuantEcon/data-lectures#XXX + date: 2026-08-04 + repoints: [] + cutover: null + # Planned waves that have not landed anything here yet. `datasets` names the # files as the audit sees them today, so the dashboard can join the two views. # `title` is the reader-facing milestone name (the dashboard is read by people diff --git a/scripts/audit_annotations.yml b/scripts/audit_annotations.yml index 73e4dfd..d436831 100644 --- a/scripts/audit_annotations.yml +++ b/scripts/audit_annotations.yml @@ -114,11 +114,6 @@ datasets: description: FRED-derived snapshot provenance: constructed-committed note: make_data.py + README beside the data - japan_population_by_age.xlsx: - description: Japan population by age group (prob_dist bimodality example) - provenance: author-assembled - flags: [new-since-2026-07-15] - note: added by lecture-python-intro#790; upstream source not yet recorded life-expectancy-vs-gdp-per-capita.csv: description: Our World in Data, life expectancy vs GDP per capita provenance: verbatim diff --git a/scripts/japan_population_by_age.py b/scripts/japan_population_by_age.py new file mode 100644 index 0000000..563e538 --- /dev/null +++ b/scripts/japan_population_by_age.py @@ -0,0 +1,102 @@ +""" +Builder for lectures/japan_population_by_age.csv. + +Extracts the population of Japan by single year of age from Table 1 of the +Population Estimates published by the Statistics Bureau of Japan, as of +1 October 2024. + +The upstream workbook is laid out for printing: ages run down a left-hand block +and continue in a second block to its right, under two levels of merged +headers. This builder flattens that into one row per age. + +Counts are published in thousands and are left in those units, since that is +what the source states. The final age, 100, means "100 and over". + +Stages: fetch -> pre-process -> validate -> write. +""" + +import io +import os +import re +import urllib.request + +import openpyxl +import pandas as pd + +CURRENT_FILE_DIR = os.path.dirname(os.path.abspath(__file__)) +REPO_ROOT = os.path.dirname(CURRENT_FILE_DIR) +PUBLISHED_DIR = os.path.join(REPO_ROOT, 'lectures') + +SOURCE_URL = 'https://www.stat.go.jp/data/jinsui/2024np/zuhyou/05k2024-1.xlsx' + +OUT_FILE = 'japan_population_by_age.csv' + +SHEET = '第1表' + +# (age label column, first data column) for the left and right blocks of the +# printed table. Data columns run: total both sexes, male, female, sex ratio, +# Japanese both sexes, male, female. +BLOCKS = ((1, 2), (10, 11)) +TOTAL_OFFSET = 0 # total population, both sexes +JAPANESE_OFFSET = 4 # Japanese-national population, both sexes + +# The "total" row of the source table, used to check that we read the columns +# we think we did. +PUBLISHED_TOTALS = {'total_population': 123802, 'japanese_population': 120296} + +# Every figure, including the total, is rounded to the nearest thousand +# independently, so the per-age figures need not sum exactly to the published +# total. 101 roundings of up to half a thousand each bound the gap at ~50; +# the tolerance below sits inside that and far below any column shift. +ROUNDING_TOLERANCE = 60 + + +def fetch(): + with urllib.request.urlopen(SOURCE_URL) as response: + return openpyxl.load_workbook(io.BytesIO(response.read()))[SHEET] + + +def pre_process(ws): + rows = [] + for age_col, data_col in BLOCKS: + for r in range(1, ws.max_row + 1): + label = ws.cell(r, age_col).value + if label is None: + continue + # age labels look like "0 歳", "1", "100 歳以上"; the totals row + # and any notes carry no leading digits + match = re.match(r'^(\d+)', str(label).replace(' ', '').replace('　', '')) + if not match: + continue + rows.append({ + 'age': int(match.group(1)), + 'total_population': ws.cell(r, data_col + TOTAL_OFFSET).value, + 'japanese_population': ws.cell(r, data_col + JAPANESE_OFFSET).value, + }) + return pd.DataFrame(rows).sort_values('age').reset_index(drop=True) + + +def validate(df): + """Refuse to write anything that is not the shape we expect.""" + assert list(df.columns) == ['age', 'total_population', 'japanese_population'] + assert df['age'].tolist() == list(range(101)), 'ages are not 0 to 100' + assert not df.isnull().values.any() + # Column check: our per-age figures must sum to the published totals, up to + # the rounding slack described above. This catches a shifted column, which + # is the way this parse would fail. + for column, published in PUBLISHED_TOTALS.items(): + gap = abs(df[column].sum() - published) + assert gap <= ROUNDING_TOLERANCE, \ + f'{column} sums to {df[column].sum()}, published total is {published}' + + +def run(): + df = pre_process(fetch()) + validate(df) + df.to_csv(os.path.join(PUBLISHED_DIR, OUT_FILE), index=False) + print(f'wrote {OUT_FILE}: {len(df)} rows, ' + f'{df["total_population"].sum():,} thousand people') + + +if __name__ == '__main__': + run() diff --git a/scripts/render_audit.py b/scripts/render_audit.py index 119f5b0..ee0c677 100644 --- a/scripts/render_audit.py +++ b/scripts/render_audit.py @@ -402,18 +402,24 @@ def pattern_barlist(audit: dict) -> str: """ -def what_changed() -> str: - # The narrative diff vs the hand-built 2026-07-15 artifact. Static prose: - # it describes a fixed historical comparison, not live state. +def what_changed(audit: dict) -> str: + # The narrative diff vs the hand-built 2026-07-15 artifact. The prose is + # static -- it describes a fixed historical comparison -- but any count it + # quotes is taken from live state, so a landing wave cannot leave a stale + # number on the front page. + recs = (audit["migration"] or {}).get("datasets") or {} + n_re = sum(1 for r in recs.values() if r.get("status") in ("repointed", "final")) return f""" -

What moved since the {PREV_SNAPSHOT} snapshot

-

The audit was first taken by hand on {PREV_SNAPSHOT}. Regenerating from live -repo state two days later, the picture has already changed — which is why this dashboard is generated.

-
4 datasets migrated. -

The first two migration waves moved lingcod_msy_recovery.csv (was a +

What has moved since the {PREV_SNAPSHOT} snapshot

+

The audit was first taken by hand on {PREV_SNAPSHOT}. Everything below is +regenerated from live repo state, so this section follows the programme rather than freezing +a moment in it — which is why this dashboard is generated.

+
{n_re} datasets migrated. +

The first waves moved lingcod_msy_recovery.csv (was a Colab-breaking local path) and the pandas_panel trio realwage.csv / countries.csv / employ.csv (5 of their 6 reads pointed at the retired -legacy repo). Every consuming lecture now reads the central copy. Details on the +legacy repo). Later waves landed the data behind the probability lectures. Every consuming +lecture reads the central copy. Details on the migration tracker.

Legacy-repo URLs: 8 → 0.

The pre-MyST QuantEcon/lecture-python repo was renamed to @@ -428,11 +434,14 @@ def what_changed() -> str: intro's copies by URL), and its short_path fetches intro's committed graph.txt over the network. That file was "shadowed dead weight" in the prior audit; it is now load-bearing.

-
2 new datasets appeared (lecture-python-intro#790). +
The 2 datasets that arrived under the old conventions have since moved.

us_adult_heights.csv and japan_population_by_age.xlsx entered -prob_dist as local-path reads — the pattern P1 just migrated away from. New data -keeps arriving under the old conventions until the styleguide -({issue_link("QuantEcon/QuantEcon.manual#108")}) lands.

+prob_dist (lecture-python-intro#790) as local-path reads — the pattern P1 had just +migrated away from. Both are now here, and in both cases the missing builder was recovered +rather than recorded as a gap: the heights extract is reproduced byte for byte from NHANES, +and the population table was rebuilt from its publisher, the Statistics Bureau, replacing an +e-Stat export that carried no provenance. New data will keep arriving under the old +conventions until the styleguide ({issue_link("QuantEcon/QuantEcon.manual#108")}) lands.

""" @@ -496,7 +505,7 @@ def render_index(audit: dict) -> str: {migration_meter(audit)} {cards} {pattern_barlist(audit)} -{what_changed()} +{what_changed(audit)} """ return page("QuantEcon lecture data — audit & migration dashboard", "overview", body, audit) From 56fba7f209e71cdd654fdda0a450c06ba17ae743 Mon Sep 17 00:00:00 2001 From: John Stachurski Date: Tue, 4 Aug 2026 06:32:38 +1000 Subject: [PATCH 2/3] migration.yml: record the landing PR number Co-Authored-By: Claude Opus 5 (1M context) --- migration.yml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/migration.yml b/migration.yml index 725eb60..6562adf 100644 --- a/migration.yml +++ b/migration.yml @@ -164,7 +164,7 @@ datasets: # Flip to `repointed` when the lecture PR merges. prior_pattern: local-path landed: - pr: QuantEcon/data-lectures#XXX + pr: QuantEcon/data-lectures#33 date: 2026-08-04 repoints: [] cutover: null From e19d0eda6770591cf0ecf6ae9c85394f0e0b1df6 Mon Sep 17 00:00:00 2001 From: John Stachurski Date: Tue, 4 Aug 2026 06:33:16 +1000 Subject: [PATCH 3/3] Keep the xlsx annotation until the lecture repoints The scan reads lecture repos' main, where prob_dist still fetches japan_population_by_age.xlsx, so removing its annotation in the same PR that lands the CSV replacement fails the strict build on an unannotated reference. The entry stays, now recording that the CSV supersedes it, and goes when the repoint removes the last read of the xlsx. Co-Authored-By: Claude Opus 5 (1M context) --- scripts/audit_annotations.yml | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/scripts/audit_annotations.yml b/scripts/audit_annotations.yml index d436831..f682c01 100644 --- a/scripts/audit_annotations.yml +++ b/scripts/audit_annotations.yml @@ -114,6 +114,15 @@ datasets: description: FRED-derived snapshot provenance: constructed-committed note: make_data.py + README beside the data + japan_population_by_age.xlsx: + description: Japan population by age (prob_dist PMF example) + provenance: author-assembled + note: > + an e-Stat export with no recorded provenance, added by + lecture-python-intro#790. Superseded by japan_population_by_age.csv, + which is rebuilt from the Statistics Bureau original and manifested + here; this entry goes when the lecture repoints and the xlsx leaves the + lecture repo life-expectancy-vs-gdp-per-capita.csv: description: Our World in Data, life expectancy vs GDP per capita provenance: verbatim