- Scratch and marker positions are a Cesium-free Vec3 (src/geo/vector.js), so writing a coordinate does not allocate.
- The dirty set and the upsert lists are kept between frames instead of rebuilt; patchDrawnIcons reuses its lookup arrays.
- measureFrames and stopFrameProbe live in src/layers/transit/testing.js again.
- Tests: 3000 moving vehicles stay under 8 KiB per frame of layer allocation; moving 500 icons allocates under 256 B per upsert.
createApplicationCatalog takes an optional placement provider. Its
layer(kind) hands Flights, Military Flights, AIS Vessels, Transit and Rail
a place(id, wallMs), and optionally observe(id, fix) and forget(id). While
live, each layer tells it every fix it accepts and draws a contact where it
places it: position, height (the layer's own when none is given), course
and speed. A null answer, a provider that throws, a kind it leaves out, or
no provider at all keeps the layer's own motion unchanged, and following
the world clock the layers never ask it. A transit vehicle or train it
holds still lets the render loop idle and is asked again each second;
vessels ask for frames only while it places one, and stop when the layer
turns off.
Tests show the layers draw and reconcile exactly as before without a
provider and with one that places nothing, draw where a provider places
them, and idle while it holds them still.
Brings in main up to f24abfc88: cockpit subject providers, the render
failure notice, aircraft drawn in their own mesh colours, airport ground
vehicles, and an aircraft with no ground track facing its true heading.
Under the seam:
- The render failure notice lives in the Cesium adapter
(src/render/cesium/renderFailure.js) and is the application hook
renderFailureNotice({allowReload, store}); its private Cesium fields
(Event's raise queue, the widget's render loop flag) are read through
internals.js. The inline panel's timer waits while the frame port reports
the loop stopped, and goes on after a Retry.
- Ordinary aircraft models draw white with the 'highlight' colour blend
through the draw port; tracked and emergency colours tint the mesh.
- Held trains and transit vehicles hide their route set and markers through
the draw port; the fleet model test fakes answer loads through the fake
draw port instead of Cesium's glTF loader.
- The Cockpit's subject view follows and places the camera through the
camera port.
- 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.
The fleet models (#1714) draw nearby trains and vehicles under the 3D toggle. A
held one now leaves out its model as it does its icon, as the Vessels layer does
for its bridge subject: its model item is hidden, and handing the visual back
to its icon keeps the icon hidden while held. Tests drive each layer with the
real fleet-model layer and a stubbed glTF loader; both fail without the fix.
getSelectedTrain, getSelectedVehicle and getCockpitSubject fill drafts and
freeze a new object only for a part whose values changed, so a view that
reads them on every frame builds nothing while the train or vehicle stands
still, and a moving one keeps its feed, stop and provenance objects. A
train's stops are copied once per list the stops provider returns, each
stop a frozen copy, instead of on every call.
- A subject with no height at all waits at most
COCKPIT_GROUND_WAIT_TIMEOUT_MS for the floor, its HUD updating and
reading ACQUIRING SURFACE; then the view ends and the provider gets its
subject back (there is no height of its own to fall back to).
- A hold that returns a promise is refused as before and released again
once it settles, unless that subject is being ridden by then; release
must tolerate a second call.
- Rail and Transit stamp selectedAtMs only for a user's selection (a click,
focusTrain or focusVehicle); a deselection clears it, and a train or
vehicle selected only by a hold reads null.
- Labels, entry face and look may be functions, read again on every use
and on a locale change; a failing read keeps the last good value.
- A frame with no height keeps the eye's last height, or waits
(surfaceAcquiring) until the floor is known; heights outside
-1,000..50,000 m read as null.
- Provider methods are synchronous: a returned promise is refused like a
throw, and a hold that fails is released.
- A candidate may carry selectedAt; Rail and Transit stamp selectedAtMs
on every selection, so re-selecting moves a subject to the front.
- A released Transit vehicle leaves the visible set at once.
- docs/CURRENT-STATE.md states the new entry order; docs/APPLICATION.md
documents cockpit subjects, ./i18n/locale and ./layers/rail/paths.
GEV itself registers no new subject: with no provider, the Cockpit offers
aircraft, satellites and vessels as before, except that the most recently
selected of them now wins entry.
- services.cockpitSubjects: providers ({kind, layerId, entryFace, priority,
isEntryAllowed, read, hold, release, hud}) are checked once by the
coordinator; frames and HUD views on every update; a throwing provider
ends its view or keeps the last HUD.
- Entry: the most recently selected subject wins, across all kinds.
- Camera: a frame may turn the ground clamp off, anchor exactly on the
subject and update every frame.
- A generic HUD: kicker, title, cards, always-visible source rows, banner,
a veil with light streaks (off under reduced motion) and a bottom
progress line whose close ticks merge.
- Text comes from the provider, already in the page's language
(./i18n/locale: getLocale, onLocaleChange).
- Rail and Transit read and hold APIs (getSelectedTrain/Vehicle,
getCockpitSubject, setCockpitSubject, focusTrain/Vehicle); a held
train's marker, card and route line are hidden, and its feed keeps
polling while held.
- ./layers/rail/paths is exported.
Its last turn was kept when it came back into view, so a vehicle that
left the view while the camera moved, and came back after the pass for
the camera's new pose, kept the turn from the old pose until the camera
moved again. Coming back into view now clears its last turn.
The rotation pass reprojected courses only when the camera moved, a poll
marked a course, or a vehicle was moving. A vehicle that entered the view
or was shown on a later visibility sweep (its ground height arriving
after its poll) kept the icon pointing up until one of those happened,
under a camera that held still. The visibility sweep now marks the
rotations for the next pass when it changes what is in view or shown.
Takes main's per-point mesh floor readings (#1642) onto the seam's
port-based sampler: a spot refused only while the cell's DEM prior was
cold is probed again, each spot is recorded once, a cell keeps at most
sixteen readings, and grounded flights read the floor at the contact
held inside its sticky cell.
Brings in the CCTV opening camera, the Iceland, Giroc HLS, Taiwan, Quebec,
Ottawa and world camera packs, the world clock and timeline, the zoom-out
limit and registered-only layer state, ported onto the renderer contract:
the sky follows the world time through the frame port's setTime in
preUpdate, the Quebec overlay reads the eye, projects and redraws through
the camera, query and frame ports, launches hide pending missions through
their draw records, and mission orbit rings turn on the scene clock.
With readings kept per point, the floor a vehicle gets depends on which point
is probed and where it is read. Transit offered the sampler the first point of
each vehicle's path in a cell, usually its oldest fix there, so a vehicle that
stayed in one cell never had its current spot probed. It now offers the latest
point of the path in each cell, and the newest fix in the vehicle's own cell
(the selected vehicle's history is walked newest first).
The grounded-flight display clamp read the floor at its sticky cell's centre,
which now answers from the reading nearest the centre, not the contact. It
reads at the contact's own position, held inside the sticky cell so boundary
hysteresis still applies.
Each sprite layer's icon or dot set is now made at a draw order taken
from SPRITE_LAYER_ORDER through the renderer's per-set drawOrder:
flights over military, vessels, directions, transit, bikeshare, FIRMS
and CCTV, whichever layer was turned on first. Picks name the topmost
first. The orders are negative, so selection highlights, route
endpoints, the tracked plane and every other set at the default order
still draw over all sprite layers, as they did before.
The raise-to-top restorer, the sprite registry and the primitive-name
lookups are gone, which takes spriteOrder.js off the viewer-access list.
Before, the restorer raised the layers to the bottom of the scene in the
order they were enabled, so flights usually ended up under the others.
Transit drew its vehicles into two Cesium billboard collections and culled,
turned and floored them through the Cesium scene. It now draws them as two
draw-port icon sets, still and moving, registered with the sprite order by
set id in the same places. Each marker is a plain record; a frame writes the
changed ones in one upsert per set, and markers taken out in one task go in
one removal. Clicks read the draw-port hit.
Visibility uses the query port's horizon test and the frustum planes of the
camera port's pose. Icons stay turned in screen space by an exact projection
through the query port, the same answer the Cesium helper gave. The flat
globe is told from terrain by asking the height service once per surface
change; floors are read again when the answer flips. The mesh floor sampler
sees through the layer's sets by id. The camera pose and surface version are
read once per task.
The Cesium draw port now moves and turns an icon without restyling it. The
GL upload probe moves to src/render/cesium/uploadProbe.js, which the QA
scripts hand to the frame probe; the QA scripts read the billboards drawn for
each set.
Transit lowered the one camera-change threshold every listener shared to
0.05, through a claim ledger. The camera port measures each listener
against its own threshold, so Transit now asks for 0.05 on its own port
listener, nobody else's trigger changes, and the ledger goes.
A new listener measures from the view at the next tick, so the layer looks
at the view once more after that tick, and a move made in between is not
missed. Disable always removes both subscriptions.
The billboards, the occluder and frustum checks and the frame probe still
drive Cesium, so the renderer stays on the Cesium list.
The traffic lifecycle listens on the frame port's preRender and on the
camera port's changed (at its own 5% threshold) and moveEnd, instead of
the scene's preRender, the camera's events and the shared
percentageChanged ledger. It no longer reaches Cesium and leaves the
allow-list; rendering.js, still listed for the surface identity, tile
readiness and draped-line support, now loads the unbound camera and
globe factories for the layer.
The selected vehicle's trail is now two draw-port line sets instead of
Cesium primitives. The body is each run of joined history subdivisions as
draped lines in two lanes, a dark 5 px backing under the 3 px mode colour.
A subdivision shows once the vehicle has passed it and fades with its age;
both are written in place as segment shows and colours, once a second or
at a crossing, and a new mode colour is written the same way. The head is
the subdivision the vehicle is on, in the same two lanes, revealed up to
the vehicle with revealFraction: it is drawn anew only at a crossing.
Shows are written only when they change.
The transit tests bind a fake draw port, and the transit QA script finds
the trail's primitives by their draw set.
The transit rendering part now owns everything that drives Cesium: the two
billboard collections (made, shown, cleared and removed in the same order
as before), the camera change listener and its sensitivity claim, the
preRender hookup, the camera height and subpoint reads, the globe-shown
and flat-terrain check, the mesh floor sampler call that sees through the
billboards, and the frame-measuring QA probe. The camera factory import
moves there too.
The height part asks the terrain under a vehicle through the queries
gateway (heightsOf over the query port) instead of the Cesium height
service; the Cesium query port answers it from the same loaded globe
tiles. The QA fixture paths are built with the shared ellipsoid helper,
which gives the same coordinates.
The transit tests bind a query port that reads the fake globe, and a new
test covers the hidden, terrain and flat globe answers. Height, lifecycle
and testing no longer import Cesium or drive the viewer, so they leave the
allow-list.
Feed activation and the view bounds read the camera port's pose and view
rectangle instead of the Cesium camera; the look-at centre comes from
geo/rectangle, which crosses the antimeridian as Cesium's does. The layer's
lifecycle loads the Cesium camera factory, so a viewer of a caller's own
still has a camera to read. viewport leaves the Cesium allow-list.
Station dots and the selected-station highlight are now two point draw
sets, picked through the draw hit, and the layer follows the camera
through the camera port at its own threshold instead of the shared
percentageChanged ledger. Altitude and view centre come from the camera
pose and view rectangle. All five bikeshare files leave the Cesium
allow-list.
The Cesium draw port now keeps a rebuilt host in its place in the scene
instead of re-adding it on top, tags hosts with their set id so sprite
order can raise a draw set (registerSpriteSet), and restyles unclamped
points in place when they have not moved, so a status poll or selection
no longer rebuilds every station. Other draw-port layers (cyclone
markers, earthquake discs) now also keep their first-build slot rather
than jumping above the sprite layers on each rebuild.
VisualEffects defines each preset on the renderer and keeps its own
records of intensities and parameters, written through to the port, so
crossfades, the anime sky fade, bloom and sharpen behave as before. The
renderer animates the animated styles. Sensor picks read the style
records instead of the renderer's stages. The application renderer now
has the effects port and its capabilities.
The governor no longer writes the Cesium scene's render mode itself: each
owner holds the renderer's frame port once, requests go to it, and the
port is the one place the loop's mode is set. The port now also keeps
the loop from redrawing for scene time alone, which the governor did.
The application installs the governor as soon as its renderer exists and
holds the loop continuous ('boot') until its tools are up, where the
governor used to be installed, so boot draws as before. The hidden-tab
suspension goes through the port too, and the application's renderer now
has the whole frame port.
The Transit prompt said "Fly below 3,000 km", but the closed gate opens at
2,850 km; it now names the enter threshold, as Bike Share does. When the
camera changes coverage, Transit and Bike Share ask the Layers panel to
repaint instead of leaving the old prompt up until the next 15 s / 60 s tick.
Over a city with no feed or system (London, Tokyo) both rows read
"ON · — · never", the same as a feed that never answered. Transit already
computed a coverage note but put it only in `coverage`, which the row does
not show; Bike Share reported nothing at all.
Both layers now report a `zoom-in` guidance status with a `statusMessage`:
"Fly below 3,000 km to a covered region" / "No feed here yet · N regions
available" for Transit, and "Fly below 48 km to a covered city" / "No bike
share here yet · N systems available" for Bike Share. A loading or failed
Bike Share city keeps its own text. Leaving coverage clears the layer's
last-update time, so the row no longer dates data it no longer draws.
The coverage-note translations move from the voice card into
`translateStatusMessage` (ui/statusText.js), which the Layers row and the
voice card now share.
Every layer and tool bound its own ScreenSpaceEventHandler and picked the
same click again: with the usual layers on, one click ran about 26
scene.pick and 2 drillPick passes (120-215 ms in NYC with 18 layers). Pointer
input now comes in through src/input/dispatcher.js, one handler per scene.
A click is picked lazily and at most once per kind of query, and every
subscriber reads the same ClickEvent (pick, hit, id, ownerId, gesture,
drill, ground). The same click now costs 3 picks and 2 drills, 17-70 ms.
- Ambient onClick subscribers run only while no tool holds the pointer,
checked once per event; tools subscribe with their lease. Moves and
releases always reach ambient subscribers, so hover and drags end cleanly.
- Camera input and the viewer's stock select/track clicks are ref-counted
holds (suspendCamera, suppressStockClicks). The dispatcher owns the
controller's enableInputs, so a Cesium flight that ends no longer turns
input back on under a drag or the cockpit.
- The draw tool, imagery box, CCTV gizmo, cockpit, directions and every
selecting layer use it; bindTrackingClickGesture is gone (the gesture is
on the event). A source scan forbids new handlers, enableInputs writes and
stock-click removal outside the dispatcher.
- The sensor-style pick correction transit had now applies to every layer.
- Radio's 8-pick near-miss ring is one 17x17 windowed drill.
- A missed pick passed to pickContact is no longer picked again, and a
touch tap's contact search shares the click's drill.
- A gizmo drag whose release Cesium never reported ends on the next press
instead of leaving the camera held.
The dispatcher is exported as ./input with the pointer-lease functions, and
documented in docs/APPLICATION.md.
Core told time eight ways: Date.now() in most layers, JulianDate.now() for
flights and military, Cesium's clock frozen at page load for the sky, the
weather clock for observed imagery. They now read one clock,
src/time/worldClock.js, exported as ./time.
Live is a mode, not a timestamp: while live the clock is the wall clock,
so every layer draws as before. Away from live only the layers that follow
the clock move with it. The weather products follow it, as they followed
the weather transport; a host opts others in with follow(id).
- The weather clock takes its time from the world clock and drives its
replay, stepping its frames as before.
- Satellites, launches, flights and military (still 30 s and 15 s behind),
transit, GNSS, Recent Imagery, street level, the HUD and the celestial
ring read worldTimeFor(id). Cadences stay on the wall clock.
- Cesium's sky (stars, sun, moon, atmosphere, lighting) draws at the world
time each frame instead of the moment the page loaded.
- Weather frames and satellites declare coverage() for a host's timeline.
An ESLint no-unused-vars scan over src, server, scripts, tools, build
and pinokio, repeated until it is stable, reported these.
Runtime modules keep every call and drop only the unread binding:
- src/ui/cockpitInput.js: the unread result of enter()/exit().
- src/layers/launches/rendering.js: the unread entity returned by
entities.add().
- src/layers/radio/rendering.js: the unread map key in the horizon scan,
which now iterates the map's values.
QA scripts and tools drop an unused helper, two unused in-page module
imports, unused counters and destructured fields, and an unread wait
result. Tests drop source files read into variables that no assertion
uses, redundant StyleManager imports (the shell-source helper already
imports it), and unused destructured fields.
The one remaining report is the `_` loop variable in src/tools/schema.js,
which counts iterations and is intentional.
Drop module-local helpers that nothing calls (linkAbort, summarize,
stringOption, summarizeEntity and the polygonCenter, propertyObject and
unwrapProperties helpers only it used, CALLOUT_DOT_SIZE_PX) and unused
import specifiers in runtime modules, tests, QA scripts and tools.
None of the removed bindings is exported. Re-exports such as the ALPR
policy constants and every package export are unchanged.
The layer recorded failure in a single `_error` string. Every city feed is
fetched independently, so one retired operator set that flag for the whole
layer — and the only code that cleared it ran on a successful *activation*,
which `runProximityCheck` skips for a city that already has rendered points.
A dead feed that re-entered camera range therefore set the flag on every
camera move and nothing ever cleared it: a permanent LOAD FAILED over a map
drawing every other city's stations.
Failure is now tracked per city in `_failedCityIds`, and `getStats()` reports
it only when the failures left nothing to draw. A city that recovers removes
its own entry rather than waiting for a neighbour to succeed.
This was reachable only because Buffalo sits 95 km from Toronto, inside the
100 km activation radius — the one dead city in the registry with a healthy
neighbour in range. The other four failed alone, where LOAD FAILED with an
empty map reads as correct, which is why the latch went unnoticed.
Probing all 33 registry systems found five dead. Three had moved and are
repointed: Reddy Bikeshare (Buffalo) to Social Bicycles, Tugo (Tucson) and
Trinity Metro (Fort Worth) to PBSC, all verified serving stations. Two are
dropped because the systems wound down rather than moved — Broward B-cycle
404s at every documented endpoint, and CoGo's replacement Lyft feed publishes
zero stations with status last reported in March 2025. `gbfs.cogobikeshare.com`
no longer resolves, so its allowlist entry goes too.
All 31 remaining systems were verified live, and every registry URL was
checked against the proxy host and path allowlists.
## Changes
Follow-up to #596 from the owner's look: a parked transit vehicle turned with the camera. Moving vehicles project their world course into screen space on every camera change, but a vehicle that had never moved on screen could have no display course and was skipped by that pass, leaving its billboard at a screen-anchored angle; rotation invalidation also relied on camera events that a pose change without a revision could miss. Now every visible vehicle with any usable course (derived, held, or the feed's reported bearing) is re-projected at the existing 5 Hz cadence on camera pose changes. Moving playback and the 120°/s course limiter are unchanged.
## How it was verified
- `npm run format:check` (861 files), `npm run check:boundaries` (109 groups), `npm test` (4,068 passing), `npm run build` clean; tracking regression 109/109 in a real browser on Google 3D tiles.
- Node tests: a stationary vehicle with a held course, and one with only a reported bearing, under a camera at heading 0° then 90° yield screen rotations 90° apart.
- Browser harness (`scripts/qa-transit.mjs`, keyed on Google 3D tiles): a parked scripted vehicle orbited by 90° rotates 90.000° on screen; a moving vehicle's rotation still matches its rendered travel; both checks fail on main and pass here. Full keyed run: 201 pass; the nine failures are the per-frame allocation gate already recorded as a follow-up and threshold-edge readability lines on the Boston matrix, both unchanged from #596.
## Changes
Follow-up to #587 after the owner's first look at the merged transit layer on Google 3D tiles.
- **Sensor presets keep the real silhouettes.** Under NVG, thermal (white-hot, black-hot, Ironbow) and noir, vehicles are the same silhouettes as in normal view, filled opaque white with a 2 px dark halo, at the CRT display sizes (26 px fleet, 39 px selected). The solid rounded "signature" bodies are gone.
- **The selected vehicle's trail is visible.** The retained history is now draped onto 3D tiles and terrain (`GroundPolylinePrimitive`, `ClassificationType.BOTH`, the way the traffic heat-lines are drawn), 3 px core over a 5 px dark backing, per-segment age colour, a short draped head clipped to the sprite. Two defects kept it off screen before: Cesium's per-instance attribute getters return copies, so the renderer's show/colour edits were discarded, and a marker did not recover after the camera jumped to it, which hid the trail with it. Trail geometry is now prepared on selection independent of floor resolution and marker visibility; sparse long intervals (rail) keep a bounded marker corridor instead of dropping the interval; classification survives map-source changes.
- **Transit sits in the Movement panel group** between Street Traffic and Bike Share (the panel orders by `PANEL_GROUPS`, not by catalog order; the catalog is moved too).
## How it was verified
- `npm run format:check` (854 files), `npm run check:boundaries`, `npm test` (4,062 passing), `npm run build` clean; tracking regression 109/109 in a real browser on Google 3D tiles.
- Browser harness (`scripts/qa-transit.mjs`), keyed on Google 3D tiles and keyless: keyed 194 pass, with the trail pixel check passing at 300, 600 and 900 m; the remaining keyed failures are the per-frame allocation gate already recorded as a follow-up from #587 (CPU and frame time within budget), four readability lines at threshold edges, and the card-anchor check now counting frames where the selected marker is legitimately off screen; keyless 193 pass with the remaining misses being threshold edges on the thin tram silhouette (centre 0.83–0.88 against 0.85) and one bus ring 0.28 against 0.25 on a pure-white roof.
- New harness checks: `Google 3D selected trail visible at 6 of 8 projected samples` (repeated off/on framebuffer captures at 300/600/900 m, samples outside the sprite footprint, a background-change negative control; fails on main, passes here), a camera-arrival check (select a moving vehicle, `setView` onto it, marker shown and trail body drawn within 2 s), `PANEL_ORDER` pinned, raster tests comparing every sensor raster's mask and aspect ratio to the normal silhouette.
- Direct look on a keyed build: a live CapMetro bus at 30 m/s selected through the layer API draws its draped trail along the highway lane (harness capture `keyed-600m-trail-visible.jpg`).
- Adversarial review by Codex Astra before the last two fix passes (stack-switch classification, marker corridor under the vertex cap, pixel-gate loophole), all three fixed with regressions.
Docs: `docs/CURRENT-STATE.md` and `CHANGELOG.md` updated in this PR. No data-source changes.
Supersedes #218 by @spcpza, ported onto the current component layout with fixes from review.
Verified before merge: format, package boundaries, unit suite, build, and the browser tracking suite; a browser harness for the change where one applies; adversarial review applied.
Contributor: @spcpza. His original commit is the base of this branch, and his own re-port onto main (2026-09-15, `5b69011`) was consulted while finishing the height work. That port reads every vehicle's surface from the app's shared ground-floor system (`cachedGroundFloor` / `warmGroundFloor` / `GROUND_FLOOR_LIFT_M`) rather than sampling the scene per layer, keeps a vehicle hidden until its cell answers, warms a bounded batch per poll on a timer instead of per frame, and renders at the ellipsoid above the altitude gate. That is the sounder primitive — one floor system for the app, with the rooftop-versus-street question already settled against a datum-correct DEM prior — and this branch's height work was rebuilt on it, replacing a bespoke cell cache, ground prior and rooftop rule written here. Credit for that direction is his.
Future work (not in this PR), both written up in docs/CURRENT-STATE.md:
- Readable route names. The realtime feeds carry only `route_id`; `route_short_name` / `route_long_name` live in each operator's static GTFS `routes.txt`, and headsigns in `trips.txt`. Doing it properly wants a server-side metadata loader beside the transit proxy with fixed registered static sources, bounded indexes built once and persisted, a slow conditional refresh, and a versioned route dictionary the browser caches. Shipped here: ids are never rewritten, and a cut one carries an ellipsis instead of reading as a typo.
- Below-ground rendering. Subway vehicles are drawn on the surface directly above themselves, because the feeds carry no depth and a train nobody can see is not worth drawing; the card says so plainly. Real treatment wants OSM tunnel geometry per city and a ghosted sprite tethered to the surface. The owner recorded it and left it.
Round four and after (2026-09-15). The owner tested round three under FLIR and NVG, and then asked for the design to be taken further: elegant in performance, beautiful on camera, trails with a cache for public transport, Cesium-native, and a core other moving layers can adopt. Codex Astra reviewed the plan (it confirmed the speed fault: a glide started from the drawn position at receipt time but spanned the gap between report times, so with MBTA reporting every ~10 s against a 15 s poll every second glide ran at 1.5x; it rejected a queued-segment patch), then wrote the design, then implemented it in commit-sized passes with adversarial review between them and real-browser harness runs by the integrator.
What this PR now contains beyond the contributor's original port:
- Delayed playback in a contact-agnostic core (`src/data/contactPlayback.js`): each vehicle keeps a typed, bounded fix history and a display clock that runs at one second per second through it, a feed-derived lag behind real time, holding at its newest fix rather than extrapolating and never sprinting to catch up. Marker, course, motion word, card and trail all read the same displayed sample. The clock has an explicit time-offset and seek API so a rewind is a later feature, not a rewrite; flights and vessels are untouched.
- A selected-vehicle trail (`src/data/contactTrailRenderer.js`, `src/layers/transit/trails.js`): the retained history drawn with batched Cesium primitives, clipped at display time so the head is exactly the sprite, faded by age, in the mode colour (white under sensor presets), on the same surface path as the marker.
- Bounded proxy-side history (`server/providers/transitHistory.js`): fifteen minutes and at most 128 fixes per vehicle with hard byte ceilings, MBTA only for now (opt-in per feed), served by a small endpoint that never fetches upstream, so a selected trail survives a browser reload while the server keeps running.
- Scene work bounded to what is visible and moving, an occluder, cached DETECT candidates, holds on the render loop only while something visible moves.
- Sensor presets: white-hot haloed sprites under NVG/FLIR/noir (black-hot inverts), saturated mode colours under CRT, per-mode bracket tiers in every theme, the selected card keeping its mode colour.
- A discriminating browser harness (`scripts/qa-transit.mjs`) that samples postRender frames, reads the real card anchor, and measures fleet budgets at 800 and 3,000 vehicles.
Not built, recorded as follow-ups: curated route display names (lowest priority per the owner), fleet-wide tails, route snapping, a rewind UI, adoption of the playback core by flights and vessels.
Verified at merge (2026-09-15 morning): unit suite 3,896 passing, build clean, format and boundary gates clean, tracking regression 109/109, browser harness keyless 200 pass / 0 fail (13 sensor checks unexercised because the live view offered fewer than six isolated sprites; every judged sprite passed), keyed on Google 3D tiles 198 pass with the per-frame allocation gate as the only failure. Follow-ups recorded: the keyed build's playback sampler still allocates per frame (CPU and frame time stay within budget; the 3,000-vehicle metro view sits at a 27.6 ms frame interval against a 25 ms budget), the shared detection overlay's per-paint allocation, friendlier route names, fleet-wide tails, a rewind UI, and adopting the playback core for vessels and flights.
Co-authored-by: spcpza <[email protected]>