@chrtco/sdk 0.10.40 → 0.10.42

package/CHANGELOG.md

@@ -13,6 +13,28 @@ All notable changes to `@chrtco/sdk` will be documented in this file.
   emits for the relief source. Exported for consumers that describe that
   source themselves, such as the tile-worker's TileJSON.
 
+### Fixed
+- **`offline.download()` left a second copy of every pack in a browser's HTTP
+  cache.** In a browser, an Electron renderer or Electron `net.fetch`, the
+  bundle GET was stored in the profile's disk cache as well as written through
+  the `fs` adapter, so each pack cost about twice its size on disk and pushed
+  tiles out of the cache. The signed URL is never fetched twice, so that copy
+  was never read. The bundle GET, including a ranged resume, now passes
+  `cache: "no-store"`. Node, Bun and Deno have no HTTP cache and behave as
+  before. On a runtime that rejects the field, such as Cloudflare Workers
+  before compatibility date 2024-11-11, it is left off. The API calls around
+  the download are unchanged.
+- **The `CHRT_LAND` and `CHRT_RELIEF` sources carried no attribution.**
+  `generateStyle({ land: true })` emitted a `CHRT_LAND` source with no
+  `attribution`, so a map using MapLibre's attribution control showed no
+  credit for CHRTland — which is built from OpenStreetMap, whose ODbL requires
+  `© OpenStreetMap contributors` wherever the map is shown. The source now
+  carries `LAND_ATTRIBUTION`, and `CHRT_RELIEF` names its source, USGS 3DEP
+  (`RELIEF_ATTRIBUTION`; public domain, a courtesy). `CHRT_BASE` already
+  carried `BASE_ATTRIBUTION` and is unchanged; the chart source's credit is
+  still the tile-worker's to stamp. `RasterDemSource` gains the optional
+  `attribution` field `PmtilesVectorSource` already had.
+
 ## [0.10.40]
 
 The first publish since 0.10.39 (2026-09-11, from `ada7e73e`). The publish

package/dist/atlas/browser.cjs

@@ -161,7 +161,12 @@ var S52_DUSK = {
   CHMGF: "#905090",
   CHWHT: "#90a0a0",
   CHCOR: "#a06030",
-  SNDG1: "#606060",
+  // Deeper-than-safety soundings. Was #606060, which measured 2.39:1 on this
+  // palette's DEPMD (#7ba6b8). DUSK water is LIGHT, so the only way to 3:1 is
+  // darker: #505050 = 3.07:1 on DEPMD, 4.09:1 on DEPDW (spec: #292e2e). The
+  // shoaler SNDG2 stays the louder numeral through its UIBCK halo cut-out.
+  // Guarded with NIGHT's by `sounding-legibility.test.ts`; see S52_NIGHT.SNDG1.
+  SNDG1: "#505050",
   SNDG2: "#b0b0b0",
   DEPSC: "#606060",
   DEPCN: "#2E2E2E",
@@ -232,7 +237,24 @@ var S52_NIGHT = {
   CHMGF: "#482848",
   CHWHT: "#484840",
   CHCOR: "#683818",
-  SNDG1: "#383028",
+  // ⚠ DELIBERATE DEVIATION from the published S-52 NIGHT value (#1f2223),
+  // for the reason DEPCN's below gives, on the deeper-than-safety soundings.
+  //
+  // Reported 2026-09-30 at night on CHRTdesktop (Raspberry Pi 4): "we mute
+  // out the ones that are 'deep' but I need to still be able to see them."
+  // The previous value, #383028, measured 1.37:1 on DEPDW and 1.08:1 on
+  // DEPMD: the muted soundings had disappeared entirely.
+  //
+  // #897350 is SNDG2 (#c8aa78) scaled to 43% of its linear light, so it keeps
+  // SNDG2's hue (the NIGHT table's warm linework family, kind to a
+  // dark-adapted eye and to red night). Against this palette's water it
+  // measures 3.10:1 on DEPMD and 3.91:1 on DEPDW, which clears the 3:1 floor.
+  // It is 0.49x SNDG2's contrast on any of them, so the eye still goes to
+  // the shoal soundings first. On the lighter DEPMS/DEPVS water the UIBCK
+  // halo carries it: fill against halo is 4.37:1.
+  // The window is narrow (luminance 0.175-0.187 meets both conditions), and
+  // `sounding-legibility.test.ts` pins both ends for every symbol family.
+  SNDG1: "#897350",
   SNDG2: "#c8aa78",
   DEPSC: "#383028",
   // ⚠ DELIBERATE DEVIATION from the published S-52 NIGHT value (#1f2223),