akvo / eswatini-droughtmap-hub / #207

89%

push

coveralls-python

Feature/222 review copy and iks aggregation fixes (#223)

* [#222] Fix mislabelled units on the individual review page

The Weather Stations column headlined the MET station's monthly rainfall
total as "SPI · 52.3". There is no station SPI anywhere in the hub —
confidence.py derives one at read time for the agreement score and never
stores it — so the number was rainfall in mm the whole time. Its subtitle
also claimed "1 per inkhundla"; the MET network is 8 stations per region.

The headline was read from weather.data regardless of resolution, so an
Inkhundla with no station in its own region rendered a nearest-fallback
station's rainfall directly above "No data available — no weather station
in this Inkhundla's region". That is the exact number the per-region rule
(D-9) exists to suppress, shown in the largest type on the panel. It is
now gated on `region` alongside the block below it.

Same unit error one column over: the CDI-E 12-month chart was captioned
"mm / month" over a 0-1 percentile-rank composite.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* [#222] Compute IKS trend and heatmap from real submission dates

Three synthetic remnants sat in the compute path rather than behind a
mock flag, so they did not show up when the prototype fallbacks were
removed on 2026-07-17:

- The week axis was the hardcoded list "May 01".."Jul 24". Jan-Apr
  submissions were clamped into the first column and Aug-Dec into the
  last, so a chart drawn in September described a winter quarter that
  had not happened. Replaced by utils.rolling_weeks(): a trailing 13
  ISO weeks ending this week, with out-of-window submissions dropped
  rather than clamped.
- The heatmap never bucketed by week at all. An identical .count() ran
  once per cell, which cost 20x13 queries and wrote the same total into
  every week column. Now two queries, counting distinct kobo_id per
  (Inkhundla, week) — one report selecting five indicators is one
  submission, whi... (continued)

1845 of 2342 branches covered (78.78%)

Branch coverage included in aggregate %.

8540 of 9340 relevant lines covered (91.43%)

0.91 hits per line