CCTV and Context take the full right column when open, so the hysteresis
tests use a content-sized panel, and a test covers the full-height rule.
Co-authored-by: Manjunath Paparaju <[email protected]>
Phones and Cockpit keep their own layouts and do not show the switch. The
rails' name panels and edge mode run toward the middle in a right-to-left
language, and the switch reads its side again after the layout moves it.
Co-authored-by: Manjunath Paparaju <[email protected]>
Hidden rails (clean view, recording), sideways or empty moves, dialogs and
the world timeline keep their wheel; coasting after a step is swallowed for
at most 800 ms; line-based wheels step a panel per notch; the strip follows
where each rail is in a right-to-left language; a held step is dropped when
the pointer leaves the strip; nothing runs after removal.
Co-authored-by: Manjunath Paparaju <[email protected]>
Nothing replaces them on this branch yet, so the dock keeps them. The credit
pill and the HUD spacing built on it are left for a change of their own.
Co-authored-by: Manjunath Paparaju <[email protected]>
PID's routes.txt types lines 51, 52, 53, 58, 59, 232 and 275 as
trolleybuses (route type 11); the rest of 50 to 59 stay buses, as before.
Voice now also knows trolleybuses, gondolas, cable cars, funiculars and
monorails.
The transit layer knew five modes, so a trolleybus drew as a bus, a
funicular as a train and a coach as a city bus. The GTFS basic route types
are now modes of their own (trolleybus, cable_tram, monorail, funicular,
aerial_lift), each with an icon, colour, detection tier, card word, voice
name, speed ceiling and fleet model.
A vehicle record may carry its route's GTFS route_type as routeType. It
settles the mode after the feed's fleet hint and before its route hint, and
picks the finer model (a coach for 200, a high-speed set for 101) when the
type is a kind of the vehicle's mode. Retained history classifies epochs
through transitModeFor too, so trails join the live vehicle.
Route hints from each operator's routes.txt: Prague's trolleybus lines
(51-59, 232, 275) and Petrin funicular, and the trams, trolleybuses and
metro lines of Bucharest, Cluj-Napoca and Turin. A feed may also name
route corrections that outrank the route type: Le Havre's routes.txt types
its funicular (8-FUNI) as a bus, so it is corrected to a funicular.
A virtual aid (AIS message 21's VirtualAtoN flag) is a mark broadcast
where nothing physical stands, so drawing it as a buoy showed a mark
that is not there. The vessel source record now carries `virtual`
(false unless a source says true; any source may set it), AISStream
sets it from VirtualAtoN, and the vessel layer keeps such an aid's
disc: vesselModelSpec returns null for it and the layer gathers no
model item for it. Its card, selected card and readout add VIRTUAL
after its kind.
A buoy's card, selected card and HUD readout showed "0.0KT · 135°" from
whatever its source reported. A fixed mark has neither, so they now show
its kind alone, whole ("CARDINAL MARK N", which the 14-character type cut
read as "CARDINAL MARK "), in the bone white of its disc.
isAidToNavigation now knows every kind an aid report names ("Light, with
sectors", "Fixed structure off shore" had no aid word), leaves out a
"Buoy Layer" as it does a tender, and matches each type text once: the
vessel layer asks per vessel on its rotation pass, which now skips the
aid discs it never turns.
Aid rows no longer cost ships their place: aisStreamRows lists every
vessel ahead of every aid, so the layer's 12,000-row request never drops
a ship for a buoy, and vessels_in_area (and the brief built on it)
counts ships only. An aid's name runs on into its NameExtension.
A buoy, beacon, RACON, light or IALA mark in the vessel feed was drawn
as a ship. A vessel record whose type names one, or whose MMSI is an
aid-to-navigation station's (99MIDxxxx), is now isAidToNavigation and
vesselModelSpec draws it as the fleet set's buoy of its mark (cardinal
N/E/S/W, port or starboard hand, safe water, isolated danger, special
mark; a red can when the type names no mark), at its built size and
flagged fixed so its model does not turn to a heading. Its icon is a
disc instead of a chevron.
AISStream now subscribes to AidsToNavigationReport by default, and the
server names each report's aid type code (0 to 31) as its kind instead
of reading it as a ship type.
readsb rows (adsb.lol and a local dump1090 receiver) often send no track for
an aircraft on the ground, only its true heading. The track then read as
missing and the aircraft was drawn facing north, so one taxiing south looked
like it was moving backwards. Each normalizer now falls back to true_heading,
never the magnetic heading.
Every aircraft model (Flights, Military Flights, Local ADS-B) mixed 94% of a
flat class colour over its mesh, so airliners drew as white silhouettes and
military ones as amber. Models now use HIGHLIGHT: white leaves the mesh's own
colours as built, and the tracked and emergency colours tint the mesh without
hiding its paint. The IR boost still draws them flat white.
The car, van, fire truck and pushback tug were stacked boxes. The car is now an
operations SUV (rounded body, glass house with a sloped windscreen, roof light
bar, grille and lamps), the van a high-roof panel van, the fire truck an airport
crash tender with roof and bumper turrets on three axles, and the pushback a low
chamfered tug with a glazed cab. All stand on the ground at their built size.
Their trail anchors are measured again.
Feeds carry many airport ground vehicles (type ZVEH) and ground stations
(GRND) that were drawn as the generic plane, some at the 10 km airborne
default.
- types.json gains aircraft_callsign rules: a ZVEH callsign starting FIRE is
a fire truck, PUSH/TUG a pushback tug, DEICER/SQUID/SQB a van, else a car.
GRND draws a new 11.8 m radio mast (other/radio_mast.glb).
- resolveAircraft and aircraftModelSpec take the callsign; a matched rule
names the model outright, ahead of the type's class model.
- The layer spec cache keys such types by the rule matched, so it holds at
most one entry per rule.
- These models (catalog category `other`) draw at their built size, as such
models already do, and a
contact with no altitude whose model is one now stands on the ground; the
feed's own type is kept for such contacts, as no airframe lookup knows it.
- Local ADS-B passes the callsign too.
The session snapshot recorded a user's switch only after the CCTV panel had
already rendered the layer off, so a first switch-off by the user inside
Contacts, or after a partly failed restore, still kept the panel open.
Context controls now hold the layers _clearLayersOutsideContextMode is
switching off while its transitions run; suspendsLayer reads that set, and
the snapshot is no longer consulted.
- A frame with no speedMps gets an exact anchor: an inertial one ran on at
zero speed and closed at 0.75 m/s, so the camera fell kilometres behind a
moving train. The docs example now passes speedMps and anchor: 'exact',
a name for its HUD, and isEntryAllowed for a train bound to its line.
- Transit's snapshot height is the floor where playback draws the vehicle,
not the newest report's floor uphill, and null while that floor is cold,
so a view waits for the surface instead of standing on the ellipsoid.
- Rail's direction follows playback once it has placed the train: a report
that runs back (a reversal, or GPS jitter) no longer turns the view
while the drawn train still runs forward.
- A frame's id is trimmed as the candidate's is.
- A double-click on the selected vessel no longer enters an application's
subject that took entry after it.
- The docs import from gods-eye-view/..., the package's name.
An inline panel that lost its WebGL context showed the panel runtime's own
notice and, over it, an alert dialog with no buttons that it could not act
on. With allowReload false a lost context now clears the notice and shows
none, as the docs already said. keepPanelRendering no longer logs render
errors itself: the render failure notice logs each one. Name the Cesium
private fields finishRaise and reportThrown rely on, and the version they
were checked against.
Turning a layer on does not leave Contacts, so a user could turn CCTV back
on inside the mode and off again; the layer was still in the session
snapshot and the panel stayed open on an empty frame. The snapshot now
records every layer the user switched during the session, and
suspendsLayer is true only for a layer isolation switched off that the
user has not touched since.
Most live vessel feeds send no length, and the resolver then fell through
every length rule to the range's last model drawn as built, so a cruise
ship or a container ship showed as the smallest model of its range.
types.json gives each passenger, cargo, tanker, fishing and service range
a typical_len (cargo 300 m, tanker 250 m, passenger 140 m, fishing 40 m,
...); a vessel with no usable length is chosen and drawn at it.
- a vehicle drawn as a model keeps its selected card and stays in the detection overlay
- transit and rail models drop the 1.5 m floor lift their icons carry
- vessel models are clipped below the waterline
- models switch to an unlit white look under NVG/FLIR and back
- unknown transit modes keep their icon
- aircraft model specs are cached per set; no key-array copies per frame
Vessels, Transit and Rail turn nearby icons into fleet models at the
aircraft layers' distances. A vessel is drawn at its AIS length and beam
(static dimensions now reach vessel rows), standing at the sea surface; a
transit vehicle or train stands on its road or rail, raised by its type's
elevation, and is painted in its mode or category colour where the model
marks it.
Flights, Military Flights and Local ADS-B draw each aircraft with the fleet
model its ICAO type, class or emitter category selects, scaled to the type's
length, built from the up-front loaded bytes. The eight third-party aircraft
models and their attribution are removed; every shipped model is now the
project's own.
121 GLBs under public/models/fleet (about 2 MB, MIT like the code) with
catalog.json and types.json. src/data/fleetModels.js picks a model by ICAO
type, aircraft class, emitter category, AIS ship type and size, GTFS route
type, transit mode or rail category, and sizes it life size; the set loads
once, up front. Trail anchors are measured from the meshes.
startPageLocale, followPageLocale and switchLocale start, follow and switch
the page's language as the standalone application and the globe picker do,
for an application that offers the language elsewhere and passes
localePicker: false. The standalone application uses them.
Selecting a vessel asks Open Vessel Photos (a free, keyless API) for its
photo by MMSI, once per MMSI. The CONTACTS view shows it under the vessel's
name and MMSI with the photographer, the photo's open licence and a link to
the vessel's page; only images on Open Vessel Photos' own host are shown, and
an answer about another MMSI is no photo. The vessel source exposes the lookup
as getEnrichment({kind: 'photo'}) and fetchOpenVesselPhoto for applications
with their own vessel source.
The activation probe raycast only the photoreal tiles. On a globe map the ray
missed, or ran on through the Earth and met a primitive thousands of
kilometres away, and the monitor plane stood at the camera's full range:
about a kilometre wide and hundreds of metres up. A hit at or beyond the range
now counts as none, and on a globe map the plane also clamps short of the
globe's terrain along the axis, or flat ground at the mount's ground while
that terrain is still loading, whichever is nearer.
The preview settled when the image had loaded, so the panel's <img> took the
src while it was still decoding and then faded in over 180 ms: every camera
click lingered on an empty box. The preview now settles once the frame is
decoded (a rejected decode still shows the loaded frame, and a decode that
never settles shows it after one second), and the fade is gone.
The points-of-interest keys (Q/W/E/R/T while a city is expanded) are a
second bare-key document listener, and it did not look at modifiers
either: Alt+R or a delivered Ctrl+R picked the fourth POI and flew the
camera there. It now leaves Cmd, Ctrl and Alt chords to the browser like
the application shortcuts; Shift still counts as the key. The docs and
the changelog entry say so.
A failed request for the view in between left its error set, and the
early return for the already drawn view never cleared it, so the drawn
features showed with an error state.
The browser shared one partition-index build per entry but ran it under
the first view's signal. A camera move cancels the old view before the
next one starts, so the next view was handed the same build, already
cancelled, and drew nothing (the view layer takes an AbortError for its
own cancellation) until the camera moved again. EPC hit this on its
first view after any quick move. The build now belongs to no view: each
view stops waiting on its own signal, and the build runs on for the next
view and every view after it.
When Remove switched a layer off and then its destroy failed, the
manager published no visibility change, so the selection still listed
the entry as on: the link shared it as on and a reload restored it on
while its row read off. A failed Remove now drops the entry from what is
on and remembered whenever its layer ended up off.
A share's entries joined the library after the stored ones, and the
selection kept its first 64, so a link opened into a full library never
showed what it shared and the next hash write dropped its lib=. Now the
oldest entries the share leaves off make room for it, and their layers
are removed with them.
Browse counted an entry as added only while its layer was registered,
but an entry this instance cannot serve (switched off by the operator,
or its server unreachable) stayed in the selection and held one of the
64 places. A library full of them looked empty and refused every Add.
Such entries now show as added, so they can be removed one by one or
with Remove all.
The live refresh had three faults. Its data age counted from when a
fetch finished, so with the manager ticking every cacheSeconds the next
tick found the data just short of due and data refreshed every other
tick. It switched the layer off and on to redraw, which removed the old
features before the new ones were built, closed an open feature card and,
if the rebuild failed, left the layer on but empty. And the tick the
manager runs right after enable refetched after a failed first load, so
a second failure switched the layer off instead of showing UNAVAILABLE.
The local GeoJSON layer gains replaceFeatures(): it builds the new set
while the old one stays drawn, removes the old source only once the new
one is in the scene, keeps a selected feature that is still in the new
set (same id, or same place and label), and on a failed rebuild puts the
old features, records and selection back and rejects. The library's
refresh uses it, counts age from the request, allows a little slack on
the tick, and does nothing until a first load has succeeded.
The core layer lifecycle imported the whole Data Library manifest
contract only to check the lib- prefix, so the layer-lifecycle package
owned a feature module. The namespace rule (prefix, entry id pattern,
the two id helpers) now lives in src/data/libraryLayerIds.js, which the
lifecycle owns; the manifest contract imports and re-exports it, so
library callers are unchanged.
Imagery outlines fetched Natural Earth's 1:50m countries (about 3 MB)
from the natural-earth-vector repository's master branch on GitHub at
run time: a moving, rate-limited source whose failures silently removed
the outlines. Core already ships the pinned 1:10m countries pack, so
the outlines now read that, once, the first time an outline needs it.
The share encoder wrote only registered entries. Until availability
loaded nothing was registered, so any hash write in the meantime (a
camera move during start-up) deleted lib= from the address bar, and an
entry this instance could not serve (its proxy unreachable) was dropped
for good: copying the link or reloading lost those entries. The URL now
carries what the page opened with until the library starts, and every
entry that is on afterwards, served here or not.
Every restore wrote the selection it applied to browser storage, so
opening a link that switched no library entry on stored every saved
entry as off, and the next bare visit restored them off. Core layer
state never writes a recipient's preferences from a share. The library
now keeps what is remembered as on apart from what a share switches on
for the session: a share's entries still join the library, but only a
local restore and the viewer's own choices change what is remembered.
Remove dropped an entry from the selection and reported success even
when the manager could not disable or destroy its layer, leaving a
registered, possibly visible layer that Browse, share links and saved
state no longer knew. Remove now keeps the entry and reports failure,
so its row and Remove all still reach it and a later Remove can retry.
Add registered and switched on an entry before the selection was cut to
its first 64, so a 65th was dropped from the selection but stayed on:
Browse showed it as not added, share links left it out and Remove all
missed it. Start-up registered a share's entries the same way past the
cap. Add now refuses an entry a full library cannot keep, and start-up
registers only what the selection keeps.
Paging stopped once maxFeatures were read before it noted that the
upstream had more, so a page holding exactly the cap with
exceededTransferLimit set read as the whole dataset. FEMA USA
Structures (2,000 features, one page) never showed PARTIAL in a dense
view. Stopping at the cap or the last page with more upstream now marks
the load truncated.
The browser's partition index kept whether it was complete only to
schedule a rebuild and handed the partitions on without it, so a view
counted unread partitions from an index that might be missing some and
could read as complete, even empty, over an area it had never indexed.
A view over an incomplete index now reads PARTIAL until a rebuild
completes.
Only a failed fetch() marked a footer read as worth retrying. A body cut
off after its headers arrived (a reset, the proxy's 20-second timeout)
was taken for an unparseable file: the partition was left out and the
index still read as complete, so the server cached it without that
partition for the entry's cacheSeconds (24 hours for EPC). Such a read
is now retried like any fetch failure; an oversized body is still final.
Moving from view A to B started B's request; moving back to A before it
finished returned early because A was already drawn, so B's request ran
on and its features replaced A's while the camera was at A. Returning to
the drawn view now cancels the request in between.
A share restore skipped every entry whose settled state already matched
the share, but a layer still switching on for an earlier share reads as
off until it settles. Opening a second share in that window left the
first share's layer to come on anyway. The restore now compares with the
manager's effective visibility (the newest request), so the newer share
supersedes the earlier one.
A dataset layer loaded its features once and kept them until it was
removed: the parsed set survives disable/enable, so NWS alerts, GDACS and
EONET showed what they held when first switched on for the whole session.
A library dataset now asks the manager for a refresh every cacheSeconds.
Once its data is that old it loads again, and swaps the new set in only
when it is ready, so the old features stay drawn while the request runs
and stay if it fails. An enable after cacheSeconds also loads afresh.
Disable and destroy cancel a refresh in flight.
Zoomed out, a hotspot's reports give way to its beacon, and the cards an application draws through pictureCards went with them; they came back only once the camera had stopped after zooming in. The cards now come from every report the close-up view would mark, at any height; the layer's own markers and beacons are unchanged.
The browser stamped every trip-updates answer as just read. While an
operator's feed fails, the rail proxy replays its last good copy for up to
ten minutes, and a cached copy can be a TTL old, so those delays never
reached RAIL_UPDATES_MAX_AGE_MS and a selected train's card showed them as
current. Delays now take their age from the snapshot's fetchedAt (capped
at now), as positions do in ingestion.js, so they stop applying 75 s after
the server last read the operator.
TaM publishes km/h where GTFS-Realtime asks m/s, and the Transit layer's
entry for the same feed already sets speedUnit: 'km/h'. The Rail entry did
not, so the proxy passed the readings through and the card multiplied them
by 3.6 again: a tram at 30 km/h read about 108 km/h. A registry test now
requires a Rail feed that Transit also reads to declare the same unit.
createNewsLayer takes a pictureCards(cards, { selectedId }) function; createApplicationCatalog passes it through newsOptions. With it, the layer paints no picture cards of its own and hands the function one card per pictured event on the map, most severe then newest first, with the selected report's id; it hands no cards when the layer turns off.
GDELT exports are read in parallel, so two reports with the same time
arrived in whichever order their exports finished, and the feed test that
pages through them failed now and then. Same-time reports now order by
most mentions, then by id.
A weather feed may also be a frame source. With an optional
getFrameImage({ product, time, size, bbox, signal }) the layers take
whole-extent images, detail windows and the global mosaic from it; with an
optional getFrameTile({ product, time, z, x, y, size, signal }) regional
tiles and their prefetch come from it, null being a transparent tile. Pixels
may be an ImageBitmap, ImageData, canvas or VideoFrame of exactly the
requested size; the layer draws them once and closes them. Cancellation
reaches the source, and a feed without these methods reads the image and
tile routes as before.
AIS Vessels are keyed by the id their live source gives, with the MMSI as an
attribute: an MMSI can be shared, spoofed or reused, so two vessels with one
MMSI are now two vessels. Looking a vessel up or selecting it by MMSI still
finds it.
A share link made while a vessel is selected opens that vessel. A UUID vessel
id travels as 16 bytes in a new thing field (13). A shared thing that is not
reporting now says so in a sentence per kind, with its id (ICAO, NORAD, MMSI
or vessel UUID) on a second line.
A live source may now declare a timeline: it can answer a snapshot at a
given time (source.timeline = { coverage(), stepMs, subscribe? }, read by
readSnapshotTimeline). While such a layer is on it follows the world clock
and registers the coverage. At the live edge it polls exactly as before;
away from it the layer asks for getSnapshot({ ...query, atMs }) at the step
at or before the clock time and shows that snapshot instead of polling.
src/time/snapshotTimeline.js owns the following: a seek, time going
backwards or leaving live replaces every contact with the snapshot at the
new time (play waits on it, for at most 8 s); playing on keeps one
snapshot ahead and gates play on it, applying those snapshots as polls so
contacts move smoothly; outside coverage the layer is empty; back at live
it is cleared and polls at once. One load at a time, and a load the
source drops for a newer one is not a failure.
While following, aircraft are drawn at the clock time with no render
delay, records and absences are judged at the clock, track requests carry
atMs, and vessels move between fixes along the great circle. The world
clock counts jumps (seeks and returns to live) in getState().
A source with no timeline never follows the clock, and every layer stays
live, unchanged.
The radio and push-to-talk helpers shared by both voice providers now pass
English catalog text, and the OpenAI controller words its detail when shown,
like the Gemini session. A token failure without a server message carries
its status as sourceParams, so it is worded again with the number. The docs
say sourceDetail is the message as given, catalog text when it is the app's
own.
The Gemini session translated each status and error message once, when it
happened, so after a language change the voice control kept showing it in
the old language. The state event now also carries the English catalog text
as sourceDetail, and the control words it when shown. The local server's
known token errors stay catalog text until then instead of being translated
on arrival.
The Gemini token route now answers a missing key with code API_KEY_MISSING,
the session carries it to the error state as missingKey, and the voice
control hides "Check microphone permission" for that error, as it already
does for OpenAI. The OpenAI controller sends the same flag, so the shared
control and the controller agree whichever updates last.
The voice error was matched against the translated key message to decide
whether to show the microphone hint, so switching language brought the hint
back and left the message in the old language. The error now carries a
missing-key flag into the status, which words the message at render time.
A controller test covers the language switch, and CURRENT-STATE.md documents
the keyless voice response.
The decision no longer reports not-entity-context or channel-closed, which
its only caller rules out first, nor pointer-scale-not-useful, which the
crop path already checks and logs. A closed channel returns before any
record, as on main. The crop decision is pointer-crop-eligible, since the
crop checks after it decide whether the pointer is on ground, and a sender
called on a closed channel logs channel-lost.
In a dense downtown view from about 400 m up, the occlusion margin of
max(3 m, 2% of the range) came to 14 to 20 m, so a wall, awning, tree or
lower facade standing 10 to 20 m in front of a street was read as the
road itself: the vehicle's dot was hidden by the scene, but its bracket and
callout were still drawn. Checked against scene.pickPosition in one such
view, 18 of 470 vehicles in frame fell in that band.
The margin now depends on how steeply the camera looks down on the
contact. A point on the line of sight s metres in front of a vehicle
stands s * sin(elevation) above it, so a surface there hides the vehicle
once that rise passes 2 m. The margin stays at least 3 m and at most 2%
of the range, so grazing and street-level views are unchanged. In the
same views no vehicle with the drawn surface within 3 m in front of it
(its own road) is newly hidden.
From the review of the first change:
- A contact back in view after longer than a retest interval is held back
and retested like a new one, instead of being drawn from an answer
given before it left the view. It gets a fresh record, so a test still
in flight from before cannot end the hold-back either.
- The hold-back for a contact's first answer stretches with the retest
pace when a crowded view slows it.
- A batch whose control point failed no longer counts as a first answer.
- An unknown log-depth flag reads as "no depth" rather than as standard
depth, which would have put every surface in front of every contact.
- A shader that fails to build leaves no shader behind and is not built
again on the same context.
The detection overlay only hid a contact behind the Earth's curve, so a
street vehicle behind or under a building, a 3D tile or a hill still had
its bracket and "VEH-nnnn" callout painted over that geometry, while its
own dot was depth-tested and correctly hidden.
A contact a layer marks `depthTested: true` (only Street Traffic does) is
now tested against the depth the scene drew and left out of the overlay
while a drawn surface stands in front of it by more than max(3 m, 2% of
its range). Other layers are untouched: their markers are drawn over the
scene on purpose.
The test runs on the GPU and never waits for it. Each tested contact is
one point drawn inside an occlusion query; the vertex shader looks up the
frame's packed pick depth at the contact's pixel, decodes it and drops the
point when the surface is nearer than the contact's threshold. Query
results are collected on later frames. Reading depth back was measured
and rejected: `scene.pickPosition` per contact costs 0.2 to 1.4 ms each,
and Chromium answers `getBufferSubData` with a round trip to its GPU
process even behind a signalled sync object (1.5 ms on a quiet machine,
15 to 25 ms on a busy one).
- Contacts are retested every 100 to 200 ms, spread across frames, at
most 256 a frame; with more in view the pace stretches so none starves.
- Hiding is immediate. A hidden contact returns 600 ms after its last
"hidden" answer, so one on a silhouette edge does not blink, and a
contact new to the view is held back until its first answer (400 ms at
most), so orbiting does not flash brackets behind buildings.
- A callout whose contact goes behind something is dropped at once, not
carried to the next solve and not faded out on the facade.
- These times run on painted frames, so a hidden tab does not expire them.
- A control point in every batch must draw; if it does not, the batch is
ignored and after three the probe stands down for ten seconds. With no
pick depth (2D, Columbus view, depth picking off) nothing is hidden.
A WebGL 1 context reads depth back synchronously about 8 times a second.
The renderer internals this depends on are read in one function,
`readSceneDepthSource`, with the engine version they were checked against.
- layerStateTokenLedger.test: the retired-token reuse case now offers the
retired token (02) to the newcomer, not News's 00.
- he.json: the catalog's singular imperative, {count} first in the plural,
Wikidata read as one source, and smoother legend and info text.
- layerManifest: News after Shipwrecks, so the prompt-budget comment stays
next to the entries it explains.
- CURRENT-STATE: Prettier-formatted (table column, blank line before a heading).
Main added the News layer at the same places: both layers are kept in the
layer names, panel order (News stays last of Events), standalone sources,
voice manifest, credits, DATA_SOURCES and the CURRENT-STATE table. News took
share-link token 00 first, so Shipwrecks moves to 01 (ledger, registry,
CURRENT-STATE and the token tests). The catalog now has 40 layers and the
share-link registry 38. Shipwreck strings are added to the new Hebrew
catalog; the other catalogs keep both sides' strings.
Removing one source's wrecks on the page split merged records: a wreck both
sources list is one Wikidata record carrying EMODnet facts, so expiring
Wikidata dropped wrecks EMODnet still had, and expiring EMODnet kept its
facts. The page now keeps every shown wreck, marked stale, until the newest
source passes the limit, then clears them all. Adds a source test for the
reported ages and limit.
The page compared the proxy's fetchedAt with its own clock, so a skewed
browser clock cleared or kept wrecks wrongly; a snapshot without a timestamp
never expired; and one old source cleared the other's fresh wrecks. The proxy
now reports each source's age on its own clock and its seven-day limit; the
page turns them into per-source deadlines when the snapshot arrives (seven
days from arrival when no age is given) and, on a failed refresh, removes only
the sources past theirs. A refused short refresh is logged.
CURRENT-STATE said each source is asked at most once a day; a failed source
is retried after 15 minutes. It now also covers refused short refreshes and
the page clearing expired wrecks, and its two shipwreck paragraphs are
rewrapped. The card's comment no longer says the records carry no names.