QuantEcon · generated data audit · Python lecture family

Lecture data — audit & migration dashboard

Every dataset referenced by lecture source across the 8 synced Python-family repos, regenerated from each repo's main by a script — not maintained by hand. Companion to the migration of lecture data into QuantEcon/data-lectures.

● audit ran 2026-08-07
41distinct static data files referenced by lectures
18migrated — every consumer reads data-lectures
22lectures fetching live third-party API data
26committed data files no lecture references
0live URLs into the retired legacy repo (was 8)
5GitHub URL spellings in use (was 6)

Migration progress

migrated — every consuming lecture reads the central copy (18) copied here, lectures not yet switched (0) not migrated (23, of which 6 queued for the next waves)

Denominator: the 41 distinct static files lectures read today. The full per-series manifest and the milestone plan are on the migration tracker.

Migration tracker → Per-dataset lifecycle — what landed, what's repointed, which PRs did it, what's queued next. Full audit → Every data reference across the 8 repos: hosting patterns, registries, orphans, provenance, live-API analysis. Dataset catalog ↗ The migrated-only registry generated from the manifests — class, license, integrity, consumers.

Where lecture data comes from today

data-lectures
33 refs
own-repo URL
13 refs
local path
4 refs
sibling repo URL
3 refs
external data repo
12 refs
%%file embedded
11 refs
live API
22 lectures

Counts are lecture→file references (a file read by two lectures counts twice); the live-API row counts lectures. Full taxonomy on the audit page.

What has moved since the 2026-07-15 snapshot

The audit was first taken by hand on 2026-07-15. 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.

18 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). 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 lecture-python.rst and archived; the P2 repoints plus the ols maketable repoint (lecture-python.myst#966) and the retirement sweeps (lecture-python-programming#576, lecture-python.myst#968) removed every live reference.

Branch-pinned refs: 1 → 0.

The critical flag on SCF_plus_mini_no_weights.csv — pinned to a feature branch of high_dim_data — was fixed by lecture-python-intro#793; it now reads main.

lecture-wasm joined the audit — and changed one fact.

The WASM mirror carries its own copies of intro's data files (all unread — its lectures fetch 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.

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 (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 (QuantEcon.manual#108) lands.