Introduces the RadarSource seam and the manifest it speaks, plus the demo
frames that make the app runnable with no network and no credentials.
MockRadarSource is a real adapter rather than test scaffolding, and that is
what makes the rest of milestone 3 testable: the timeline, prefetching, cache
eviction and every degraded path can be exercised offline because the demo
frames parse the same manifest document the Python worker will publish. Its
assets come from tool/generate_mock_frames.py — committed so the app runs from
a clone, generated by a script so they can be regenerated instead of
hand-edited. Twenty-four 512x512 frames total 108 KB, and the generator is
pure standard library so nobody needs Pillow to build the app.
The manifest parser rejects three things that would otherwise fail silently and
look plausible:
- A CRS other than EPSG:3857. The map overlays each PNG on a lat/lng quad,
which only lines up if the image is already in Web Mercator; anything else
renders visibly skewed with no error to explain why.
- A missing attribution. The frames are a derived product of CC BY-SA data, so
the credit has to travel with them rather than be remembered at render time.
- Legend stops that do not ascend, which would silently mislabel intensities.
The legend travels in the manifest rather than living as a constant here,
because the worker chose those colours when it rendered the PNGs and a local
copy could drift. RadarLegend.colorFor returns null below the lowest stop:
"no precipitation" has to be transparent, not the first colour of the ramp, or
a dry region renders as drizzle everywhere.
Every failure reaches the caller as a single RadarUnavailableException
regardless of cause, because the app's response is the same in all of them —
hold the last good frame and say how old it is — and branching on cause would
only invite divergence.
ArpaRadarSource is a stub whose every method throws. It is named by the region
config as unavailable and must fail loudly: quietly serving something else
would misreport where the data came from.
Verified: analyze clean, 95 tests passing, including a check that every frame
the manifest lists exists and is a real PNG, and that its bbox matches the
region config.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Sets up the Nuvolari monorepo layout (app/, backend/, docs/, tool/) with the
groundwork that does not depend on the Flutter toolchain: gitignore covering
Flutter, Python and every secret file shape; env.example.json as the template
for --dart-define-from-file; a verification script that skips stages whose
target does not exist yet so it is runnable from day one; and a CI workflow in
GitHub Actions syntax so it runs unchanged on Gitea or GitHub.
The documentation records facts verified against the live services rather than
restated from the brief. Two of them change the design:
- DPC VMI rasters are 1200x1400 Float32 on a 1 km grid in a custom projection
centred on Italy, not EPSG:4326 or EPSG:3857, and their GeoKeys are
internally inconsistent. Reprojection is mandatory and the source CRS must be
read from each file rather than hardcoded.
- The ARPA CAP feed carries six level values, not four: BIANCO for the
avalanche scale out of season and "-" for no data. Collapsing either into
VERDE would report "no alert" where the bulletin reports "not assessed".
Also documents why the frames we publish inherit CC BY-SA from the DPC source,
and why rain notifications subscribe to cell topics from the device so no user
location ever reaches a server.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Alby96andClaude Opus 5