[Mirror] Dataset loader provenance - CI-review mirror of PR #723 (do not merge) - #731

Closed

igerber wants to merge 17 commits into

mainfrom

mirror/pr723-dataset-provenance

Commits on Jul 27, 2026

fix: make legacy dataset provenance explicit

charlesshawmsix
authored and
igerber
committed
fix: preserve verified dataset downloads on cache write failure

charlesshawmsix
authored and
igerber
committed
fix: complete dataset loader provenance safeguards

shawcharles
authored and
igerber
committed
docs: castle treatment-coding registry note + dataset provenance doc refresh
igerber
committed
fix: widen download fallback to HTTPException; scope datasets.rst provenance claims
igerber
committed
fix: clean up partial cache files on write failure; align stale dataset prose
igerber
committed
fix: correct castle income label, float32 exactness claim, and LLM-guide provenance
igerber
committed
docs: reframe tutorial 09 divorce section as a simulation, not a replication
igerber
committed
fix: documented Card-Krueger workflow breaks on the canonical incomplete survey
igerber
committed
fix: qualify the fallback contract for verified-cache recovery; surface provenance in tutorial
igerber
committed
chore: map datasets.py to its REGISTRY section; correct stale castle-availability note
igerber
committed
docs: align the remaining fallback-contract surfaces with cache recovery
igerber
committed
fix: clear_cache() could not remove interrupted atomic-write scratch files
igerber
committed
fix: verified cache must survive every fresh-download verification failure
igerber
committed
docs: two internal contract statements still said a checksum mismatch raises
igerber
committed
fix: test_clear_cache_creates_directory wiped the developer's real dataset cache
igerber
committed
test: exercise the pinned production mpdta bytes end-to-end offline
igerber
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Mirror] Dataset loader provenance - CI-review mirror of PR #723 (do not merge) - #731

[Mirror] Dataset loader provenance - CI-review mirror of PR #723 (do not merge)#731
igerber wants to merge 17 commits into
mainfrom
mirror/pr723-dataset-provenance

Commits on Jul 27, 2026

fix: make legacy dataset provenance explicit

fix: preserve verified dataset downloads on cache write failure

fix: complete dataset loader provenance safeguards

docs: castle treatment-coding registry note + dataset provenance doc refresh

fix: widen download fallback to HTTPException; scope datasets.rst provenance claims

fix: clean up partial cache files on write failure; align stale dataset prose

fix: correct castle income label, float32 exactness claim, and LLM-guide provenance

docs: reframe tutorial 09 divorce section as a simulation, not a replication

fix: documented Card-Krueger workflow breaks on the canonical incomplete survey

fix: qualify the fallback contract for verified-cache recovery; surface provenance in tutorial

chore: map datasets.py to its REGISTRY section; correct stale castle-availability note

docs: align the remaining fallback-contract surfaces with cache recovery

fix: clear_cache() could not remove interrupted atomic-write scratch files

fix: verified cache must survive every fresh-download verification failure

docs: two internal contract statements still said a checksum mismatch raises

fix: test_clear_cache_creates_directory wiped the developer's real dataset cache

test: exercise the pinned production mpdta bytes end-to-end offline