Skip to content

MAP.md

Latest commit

 

History

History
89 lines (79 loc) · 5.59 KB

MAP.md

File metadata and controls

89 lines (79 loc) · 5.59 KB
Raw

Map

The initial view frames the Bhote Koshi river corridor. Current OpenStreetMap linework (four separate mapped ways) and Kodari, Tatopani, Bahrabise and Khadichaur labels sit above a faint district background. The translucent river band is a visual guide, not a measured 2016 flood extent. © OpenStreetMap contributors (ODbL); attribution is available on the map. The bundled 18,016-byte GeoJSON is an exact copy from codex/test-data; its provenance/archive references resolve on that branch, not to a bundled corpus here. No other corpus inputs were copied. Other rivers are OpenStreetMap waterway=river ways inside Nepal (ODbL, Overpass extract of 2026-09-27), simplified to about 400 m in features/map/nepal-rivers.geojson.

When the dashboard supplies place_counts, Map aggregates exact coordinates without jitter and draws a heat layer weighted by report count. High-zoom circles and the keyboard-accessible list open settlement totals and missing/safe/other source assertions. The list shows at most the 20 busiest coordinates. The caption separates located and unlocated reports; neither the glow nor the river band is an uncertainty or flood boundary. Counts are reports, not unique people or media locations. An empty count list preserves the Maya pins and evidence panel. The additive count backend/importer is owned by Miguel. It is integrated from 814a07d; the map consumes its fields without changing the importer or schema.

The corridor loads independently of boundaries and evidence, with its own five-second unavailable message and late recovery. HTML settlement labels need no external glyph service. get_bhotekoshi_corridor is the server endpoint; registered alongside get_boundaries in main.jac by Miguel.

People reports with mapped evidence offer Show on map. It opens Map, selects the first linked place in snapshot order, and centers its pin. Repeating the link centers it again; normal tab switches preserve the viewport and People search. Reports without valid mapped evidence have no link. This navigation follows the report's media location, not the person's current whereabouts.

Pins show the most urgent current report among people linked by a claim at that media location: Missing, Needs review, Reported safe, then No current status. Needs review includes conflicting reports, injury, or death. The evidence panel cites current person reports separately from the retained media-linked reports. A pin remains a reported media location; a status update does not relocate anyone. The bridge changes from Missing to Reported safe after the hospital update while retaining the police report. The seeded market has no linked person report and therefore shows No current status. Identity candidates are not silently combined.

Pins and the keyboard-accessible place list work while district boundaries load. A failed request or five-second delay displays a boundaries-unavailable message; a late success restores boundaries without changing selection or viewport. The map observes container resizing and keeps viewport/selection across tabs.

Map checks (27 September, corridor/count consumer): 66 Jac tests and the whole-program compiler gate pass, as does the production web client build. Nine real Maya browser checks pass (hospital update/one alert, retained police evidence, reload/reset, keyboard selection, pan/zoom/tab preservation, and failed/slow boundaries). Eight additional browser component checks use explicitly mocked count responses: exact-coordinate aggregation, 170 rows to four points, high-zoom circle selection, source-status sums, top-20 cap, empty fallback and corridor failure. Both suites pass at 375px with no document overflow or page errors. These count fixtures are not an importer or a corpus execution result. Four further checks pass against the real persisted corpus: idempotent import, 1,356 located / 2,244 unlocated reports at 170 records / four coordinates, correct settlement status totals, 375px rendering, and reset isolation. The test uses the shared incident-selector UI and unchanged backend responses. Counts were Tatopani 345, Kodari 339, Khadichaur 339 and Bahrabise 333. No corpus alerts were created. Local screenshots on Anshu's host: /private/tmp/trace-corridor-mobile.png and /private/tmp/trace-real-corpus-mobile.png. Hosted deployment and physical-phone rehearsal remain untested.

If a development preview retains a compiler overlay after a source correction, stop it and rebuild before restarting. A stale generated .jac/client/.jac-build-error.json may retain the old error; remove that marker only after a successful build confirms the correction. Do not reset shared data.

jac test features/map/test_map.jac
jac build --check_only
jac build --client web
# Separate terminal, in an isolated checkout with its own .jac/data:
jac start --dev --port 8092 --api_port 8093 main.jac
# With Playwright already available to Node (no application dependency added):
TRACE_MAP_URL=http://localhost:8092 node features/map/verify_map.cjs
TRACE_MAP_URL=http://localhost:8092 node features/map/verify_density.cjs
# Imports/reuses the real corpus in this isolated preview:
TRACE_MAP_URL=http://localhost:8092 node features/map/verify_corpus_map.cjs

The browser check resets the preview's demo. It accepts localhost only. Set TRACE_BROWSER_EXECUTABLE to an existing Chromium executable if needed; NODE_PATH can point to existing Playwright packages. Set TRACE_MAP_SCREENSHOT to capture the 375px result. Shared integration change: a defaulted LocationEvidenceView.person_id filled from the claim's About edge in the location loop; no changes to the investigation snapshot fields or seed are needed.