Remove dangerous git checkout $LAST_TAG from deployer prompt: it left the shared working copy in detached HEAD (breaking the next git pull) and did not roll back prod at all. Rollback now goes through the deploy hook (ssh ... bash ${HOOK} --rollback), which restores the app container to the previously running image. Narrow tools to Bash (git, curl) since the deployer no longer invokes docker directly.
Add .task*.md to .gitignore and remove already-tracked task files from
the index. These are orchestrator runtime artifacts (B-3) and should not
be committed.
- Tag v0.0.2 cut from main b5ba7b2 (PR #16 merged).
- enduro_russia pipeline run: ok, 5 new + 36 updated, 0 errors (39 tracks in DB).
- wikiloc: 403 from WAF on first request, graceful stop (config-complete, scrape-blocked).
- Public URL returns 502 due to pre-existing nginx config bug
(sites-enabled pointed to :5558, app listens on :5556). Patched the
config file in place; awaits operator-side `systemctl reload nginx`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
F-01 (P1): _buildGpsFiltersUI больше не хардкодит список источников —
подтягивает source_id из /api/gps-tracks/health.tracks_by_source
(ADR-013 §3 Решение D, опция D2). Маппинг source_id → label вынесен
в JS-константу GPS_SOURCE_LABELS. Активация четвёртого источника
теперь не требует изменений в этом коде.
F-02 (P1): attribution фиксируется в момент addSource, а не мутацией
src.attribution после. MapLibre AttributionControl не реагирует на
прямое присвоение — потому до этого фикса AC-15 проваливался бы в
UI-тестах. Теперь onPublicTracksCheckbox / restorePublicTracksState
сначала await _fetchGpsHealth() → _buildGpsAttributionString(),
потом _ensureGpsSources(map, attribution).
Добавлен кэш + in-flight Promise (window.gpsTracksLayer._healthCache /
_healthFetchPromise) — переоткрытие sheet'а фильтров не плодит
дублирующих сетевых запросов.
Все 24 node-теста gps_tracks.test.js зелёные.
Refs: ET-009
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace raw single-row fetch with aggregation over all pipeline_runs
rows sharing the latest started_at. Returns structured object with
regions[], sources_ok[], sources_error[], tracks_added instead of
a raw DB row with region_id/source_id strings.
Returns null when no runs exist (empty DB).
Update test_i40_health_endpoint: add db_with_pipeline_runs fixture
(two rows, same started_at, two regions) and assert the full
aggregated shape including concrete values.
Refs: ET-008
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
12-review.md (REQUEST_CHANGES, attempt 2/3) flagged 6 must-fix items
in the analysis/architecture artefacts plus matching bugs that had
already leaked into the committed implementation. This patch lands
both: documents corrected, code aligned with corrected specs, tests
updated.
P1-1: TRZ/ADR/Data/Risks referenced fictional layer ids
(`trails-grade1..5-halo-satellite`, `paths-bridleway-halo-satellite`).
Actual style*.json has only `trails-track-halo-satellite` and
`trails-path-bridleway-halo-satellite`; grade differentiation lives
inside one `match` expression on `tracktype` within `trails-track`.
Docs rewritten to operate on real ids.
P1-2: POI labels contrast was broken — spec changed only halo-color
to black, leaving `text-color: #333333` (light theme baseline)
unreadable over the new black halo. Code+docs now switch BOTH
`text-color` (-> `#ffffff` on satellite) AND halo together, with
per-theme baselines (`#333333` light / `#e0e0e0` dark) restored on
return to Schematic.
P1-3: BRD §5 hillshade risk said «hillshade auto-disabled on
satellite», contradicting TRZ/ADR/AC. BRD wording aligned: hillshade
keeps working over satellite; visual check is AC-04.
P1-4: background-color had four divergent sources (`#1a1a1a`,
`#2a2a2a`, `#1a1a2e`, `#f0ede6`), incl. an inverted-theme typo and a
baseline `#1a1a1a` that didn't match the actual `style-dark.json:28`
value `#1a1a2e`. Settled on ADR-004's single-constant model: `#2a2a2a`
on satellite for both themes; on Schematic restore per-theme baselines
`#f0ede6` (light) / `#1a1a2e` (dark). `_applyBackgroundForSatellite`
fixed accordingly.
P1-5: app.js already had `layerState.basemap` and `toggleLayer
('basemap')` (legacy «Базовая карта» switch). Neither TRZ nor ADR
specified the interaction. Added save&restore contract: on entering
Satellite save `layerState.basemap` to `_savedBasemapState` and
force-hide `osm-base`; on returning to Schematic restore osm-base
visibility from the saved value. CSS hook `body.satellite-active
#btn-basemap { display:none }` keeps the user from trying to enable
a hybrid mode (out of scope, BRD §3). TRZ §5.6, ADR-004 §8.
P1-6: `restoreTrailsState()` and `onTrailsCheckbox()` only managed
visibility of `trails-track` / `trails-path-bridleway`, leaving
their halo-underlay siblings as «phantom» halos when the user
unchecked grunты/тропы under Satellite. Introduced
`_applyTrailHaloVisibility(map, base)` reading checkbox state from
DOM; called from `onTrailsCheckbox`, `restoreTrailsState`, and both
branches of `applyBaseLayer`. Rule: halo visible ⇔ (base ===
satellite) AND (checkbox ON). TRZ §5.7, ADR-004 §9.
Docs bumped: BRD v2, TRZ v2, AC v2, Data v2, Risks v2; ADR-004
получает «Ревизии»-секцию (status remains accepted — only editorial
fixes, no decision change).
Tests:
- tests/unit/base_layer.test.js: rewritten 2 background-color
assertions (#1a1a1a expectation removed), added 6 new tests for
P1-2 / P1-4 (POI text-color per-theme baselines, single satellite
bg #2a2a2a, baseline restore on Schematic).
- All 33 JS unit tests + 22 pytest static checks green.
- Full pytest suite: 76 passed (excluding pre-existing
shapely-import skipped collection in tests/unit/test_health.py).
Refs: ET-007
Review: docs/work-items/ET-007/12-review.md (P1-1..P1-6)
ADR: docs/work-items/ET-007/06-adr/ADR-004-satellite-base-layer.md (rev. 2026-05-31)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
