A '2025-26-revised' release read as an unknown year went last, behind the
'2025-26' release, which then owned the year and dropped every revised row.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GIAS publishes its CSVs in Windows-1252 and sends no charset. The tap read
resp.text, so requests guessed the codec, and the encoding="latin-1" passed
to read_csv did nothing on already-decoded text. On 3 Oct 2026 the guess was
windows-1250, and "St Thomas à Becket" (138950, 149557) was stored as
"St Thomas ŕ Becket". A different guess on another day would garble other
accented names.
Both streams now decode the downloaded bytes themselves (gias_csv.py). A byte
Windows-1252 leaves undefined becomes U+FFFD with a logged warning instead of
failing the load, so one odd name cannot stop the daily refresh. None of the
nine extracts checked (1 Jul to 3 Oct 2026) contains such a byte.
Checked by running the tap on the real 3 Oct extract with .text forced to
windows-1250: all 52,586 rows decode, with no "ŕ" and no replacement
characters.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fact_ks4_destinations and fact_ks5_destinations join dim_school, so the
daily build's stg_gias_establishments+ and the monthly Ofsted build's
dim_school+ both selected them. They also read stg_ees_ks4/ks5_destinations,
which only the manually triggered EES DAG builds. Where that DAG hasn't run
since the destinations models landed, dbt_build fails with "relation
staging.stg_ees_ks4_destinations does not exist", and sync_typesense and
invalidate_cache never run. Production's register data has been stuck at
about 25 Aug 2026.
Both builds now exclude the descendants of the two EES staging models, as
the daily build already does for the KS2/KS4 lineage models. The EES DAG
still rebuilds the marts when their data changes.
test_dag_selectors reads the model graph from the SQL (CI has no dbt) and
checks that every scheduled build only reads models it or the daily build
builds. It failed for the daily and monthly Ofsted builds before this change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The old sync created a collection, imported batches without reading a
single import response, and swapped the alias regardless. A partial
import published a half-empty index, and two overlapping DAG runs could
prune each other's collections.
Publication now checks every import response and the final document
count before upserting the alias, and holds a session-scoped advisory
lock across the read and the publish so concurrent runs serialise.
Cleanup keeps the previous collection as a rollback pointer and is
best-effort: an uncertain alias response must never delete what might
still be live.
Also parses the Typesense URL properly instead of splitting on colons,
which mangled any host carrying a scheme and a default port.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
tudor