92126df4202ca09eb444a5177df6ffa4f0f936e9
6 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 karti
|
48ddfcdaec |
perf: the terrain is a grid of chunks, so three can cull it
`createTerrain` returned one mesh covering 32.50-42.05 N and -124.50 to
-114.00 W, and three frustum-culls per *object* — so at every pose, from
400 km and from 2.5, the merged board submitted all 86,400 of its visible
triangles. Standing over San Francisco it paid for the Mojave.
It now returns a `THREE.Group` named `terrain`, holding a 5 x 5 grid over
`city.bounds` — **17 non-empty chunk meshes**, eight cells being Pacific,
Nevada and Arizona — plus one un-chunked shadow caster. The geometry is
not touched: same patches, same resolution, same board-wide normals. Only
the index is cut up, which is what makes this culling and not level of
detail, and what sidesteps placement drift entirely — no ground is
re-derived, so nothing that was placed against it can move.
All 18 geometries share **one** `position`/`color`/`normal` attribute
triple: no duplicated vertices, one upload of exactly today's bytes, and
no lighting crease down any of the sixteen seams, since the normals were
averaged before the partition existed. The bounding volumes are therefore
assigned by hand — `computeBoundingSphere` walks the position attribute
and not the index, so a chunk left to compute its own would get
California's, cull nothing, and restore the whole cost with seventeen
extra draw calls and no visible symptom. `terrainLod.test.ts` pins that.
5 x 5 because a draw call is the scarce resource here, not a triangle.
The sweep is in the constant's own comment: 4x4 -> 5x5 buys 1,246
triangles per added draw, 5x5 -> 6x6 buys 573, 6x6 -> 8x8 buys 61. Five is
the last grid where a draw is worth more than a thousand triangles.
The caster stays one object with `frustumCulled = false`, deliberately:
`WebGLShadowMap` gates the depth pass on the same test as the colour pass,
so a chunked caster would drop a ridge that legitimately shadows into
frame the moment the camera turned. It sits at `drawRange(0, 0)` with the
hooks inverted, and `renderBufferDirect` early-returns on `< 0` and
`Infinity` but not on 0 — one zero-triangle draw call per frame, forever.
Gated on `city.districts.some(d => d.detail === true)`, which
`cities/unify.ts` is the only place in the repo that writes. `?city=sf`,
`?city=socal`, both `?one=0` California cells and `office` take the
unchunked branch and get today's single mesh inside a group: measured, the
socal board is identical to the draw call at three poses, and `?one=0`
identical to the digit at five.
Measured with `scripts/cost-at.mjs` at 37.7897, -122.3972, merged board,
before -> after (triangles / draws):
400 km 355,759/414 -> 355,759/431 the whole board, all 17 drawn
120 km 351,019/327 -> 321,057/340
45 km 332,969/265 -> 270,373/273
7.7 km 330,578/340 -> 222,482/343
2.5 km 329,544/335 -> 215,598/338 -34.6%
and over Los Angeles at 2.5 km, 761,648 -> 335,186: a breach of the
400,000 cap that no budget cell poses at, now under it. The +17 draws at
the whole-board pose is the price, it is a bound and not a measurement
(there are 17 objects), and it lands at 437/433 against caps of 460/455.
No cap moved; the budgets file gains one sentence saying that geometry
submitted is now a function of the camera, and only ever downward.
What it does not fix: it is culling, not level of detail — a chunk 1% on
screen draws 100% of its triangles, and the ground under San Francisco is
still `UNIFIED_FINE_METRES` at 300 m against that metro's own 45. Nothing
streams; build time and resident memory are unchanged. ~157,000 triangles
at the close pose are sky, and after this they are three quarters of what
is left. `minimap.ts` still samples 160 x 160 across the whole board.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
karti
|
b25f217e3e |
feat: real fire on the boards, the LA office as a twin, and a night sky worth reading
The world stops being a simulation of California and starts being California. **THE PROMOTION GATE WAS THE FIRST COMMIT, BEFORE ANY ORANGE PIXEL EXISTED.** On today's live store the SoCal board contains 22 incidents. Every one has NULL acreage and fifteen are nameless LA County dispatch numbers. Drawn naively that is 22 orange marks over Los Angeles on a day nothing is burning — in a frame that contains no other warm colour, so one glyph would be the most salient object on the board and twenty-two would spend its credibility permanently. `acres >= 10 AND contained < 80 AND type != 'RX' AND last_seen = max(last_seen)` returns 0 on SoCal, exactly 5 on California, 0 on the Bay — same body, same day, three correct answers. The empty board is a deliverable, not a fallback: it says "No active fire on this board — CAL FIRE and WFIGS, just now", states that 21 records were gated and why, lists the largest fires burning OUTSIDE the frame with distances, and counts the hot pixels it is deliberately not drawing. **The privacy leak is structurally impossible rather than carefully avoided.** cloud-1 serves a projection; the four home-relative columns never leave that box. `observations.threat` was the one that nearly got through — it is `(16/distance)^2 x log10(acres) x momentum x containment x wind-alignment`, so with acreage and containment public it inverts to a distance circle around a house and three fires give an intersection. A grep of the built bundle for distance_km, bearing_deg, threat, 7762 and the street name returns nothing. **Deliberately not used, and both would have produced a confident wrong answer:** the store's `air` table retains only the last parameter of each poll, so all 78 rows read "Good" while the live feed reports ozone 101 "Unhealthy for Sensitive Groups" — haze driven off it would clear the sky during a smoke event. And `weather` is written only inside the NWS alerts loop, so a quiet day stores no wind at all. Tera's own per-region NWS wind is already correct and already what the clouds drift on. Satellite detections are drawn as evidence and never as incidents. The permanent industrial heat source 4.7 km from the owner's house is flagged persistent and dropped, asserted by a test that first proves it is present in the fixture. MODIS integer confidence and VIIRS string confidence are branched on `sat`. **The LA office is a twin.** Its entire authored second storey — Model Loft, Model Bay, The Materials Room, 430 lines nobody had ever stood in — is reachable on foot: a walker crosses level-1 to level-2 in 73 fixed steps, floorY 0 to 5, verified against the real pack rather than a synthetic plan. Its two studio devices read real hardware through a field-allowlisted bridge: mute, volume and reachability only. Never level, because there is no passive level upstream and obtaining one would record a room with people in it. Never dB, because upstream is gainPct across four different native scales. The bridge refuses all writes. Fixed at its root: an anonymous visitor was getting permanently at-rest instruments backing off against a 401. The tier moves into `createDeviceSource`, so anon gets the living simulator three file headers already promised. **Item 8 is closed, not fixed, and the correction is the point.** The Bay Area "stutter" was GPU power management — the card sat at 500 MHz of 2725 through every run that reproduced it, 4096/2048/1024/256 shadow maps all render in 1.21-1.31 ms, and two consecutive runs over a byte-identical dist gave 33.4 then 16.7. The allowance is removed and the cell is back to 16.7. Geometry is the gate; frame time is advisory. Item 7 was re-scoped after measuring: 1,069,006 of the Bay Area's 2,265,056 triangles were the second submission of the same buildings into the shadow pass. Mobile now has its own triangle caps and bay-area mobile draws 1,266,096. Also: bridges and the freeway corridor light up at night as emission, not lights — 1,614 deck lamps and 18 tower heads on the Bay in two draw calls. The single change that made US-101 legible was moving its edge lines from the lit material to the unlit one: retroreflective paint, the argument the SFO night frame already makes. California went 21,991 lamps to 4,051, clustered at the 17 town districts, because a rural interurban corridor genuinely is unlit. Tests 1137 -> 1340, server 280. All ten budget cells pass on first attempt with no cap raised. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
db074e9cf7 |
feat: tone-mapped render rig, studio devices, LA fidelity pass, UI overhaul
The build the studios needed, across eight workstreams and one strict file partition. **The render rig was the quality ceiling.** The renderer ran three's NoToneMapping default while atmosphere drove the sun to 2.35 and assets set emissives to 3.2, so every value above 1.0 hard-clipped to flat white — which is why walls blew out and every fitting looked like a white rectangle. ACES filmic tone mapping and an explicit output colour space land in `stage.ts`, and the atmosphere intensity table and palette headroom are re-tuned against the new curve rather than left tuned for the clipping we removed. `engine/environmentRig.ts` builds a PMREM environment at runtime, procedurally, so nothing binary is committed. There was no environment map anywhere before, so every `metalness > 0` role had nothing to reflect and rendered dull grey — a defect the code already documented against itself in `office/optimus.ts`, where a whole material role was abandoned over it, and worked around in `modelX.ts` with a fake emissive that this change deletes. Atmosphere remains the sole light owner; the rig derives from the `LightingState` it already produced. **Studio hardware exists.** There was no device concept anywhere in the product: no type, no route, no state. `devices/types.ts` fixes a declaration/state/ capability/command contract that a smart light, a thermostat, a door sensor and a charger all fit without a schema change, and both studios now carry a desk mic and a computer speaker with deterministic simulated behaviour behind an adapter seam a real API can occupy later. Reads are the demo and are open; commands are a signed-in action and are kept off the read body entirely, because a shared cache replaying a GET that turned a microphone on is exactly what the fail-closed cache default exists to prevent. **The ADS-B licence hole is closed.** `TERA_ADSB_ENDPOINT` accepted any URL, the response was served publicly cacheable, and the attribution hardcoded adsb.lol regardless of where the endpoint pointed — one env var away from republishing non-redistributable data under an open-terms credit. The host is now allowlisted, the credit is derived from the host actually configured, public cacheability is conditional on redistributability, and a refused endpoint demotes to simulated flights and says so in `degraded[]`. The gate is on the source, not the feature: live aircraft and their detail cards stay open to anonymous visitors. **The LA studio was never the smaller pack** — 16 rooms and 248 props against SF's 4 and 28. Its deficit was fidelity per square metre: 98 of those props were ceiling troffers, it bound no props to seats, placed none of the habitat kit, and 12 of its 16 rooms had no viewpoint. Density comes from new asset kinds rather than more instances, because `furnish.ts` draws once per kind and folds colour into the batch key, so repeat instances add nothing the eye can read. **The interface stops being forty imperative mutations.** Every visibility decision moves into a pure, tested `ui/chromeState.ts` and one applier, so the chrome has coverage for the first time. Deleted: ~100 lines of CSS and two bindings targeting elements that no longer exist, and a `body:has()` rule that shifted the desktop layout by 160px for touch controls hidden there. Fixed: the office picker tabs that drew their label and their badge on top of each other. Added: a first-run flow, because the product is two verbs and neither was ever stated on screen. Mobile is designed on its own terms instead of being the desktop with things hidden — the plan view comes back, and the keyboard-only shortcuts button is replaced by touch controls. `arena/studioOps.ts` frames the whole thing as the multi-variable environment it is, wrapping the same simulators the renderer drives rather than a headless copy. Also removed `input/vehicle.ts`, which nothing but its own test imported. Tests 385 -> 961, all passing. Typecheck, build, performance budgets across six matrix cells, no-binaries, provenance, dependency licences, zero-config boot and arena source hashes all green. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> |
||
|
karti
|
2d87d9f354 |
The city points at its own buildings, and the sky stops depending on an API
**Clouds were invisible to everyone who had not wired up NWS.** The layer
took `currentWeather()?.cloudCover ?? 0`, and `currentWeather()` is null on
any deployment without a weather source — which is the default, and the
exact configuration this repo is held to: a stranger clones it, runs one
command, and gets a city with no account and no key. Their sky was
permanently, silently empty. `atmosphere.ts` already models a sky when
nobody has observed one; it now models cover too, an observed reading
still wins outright, and the clouds are there on a bare clone.
**Both offices are pins on the city, and clicking one walks you in.** Each
pack has carried a real `site` since the sun needed one, and that
coordinate was known to the lighting and to nothing else — a visitor
looking at the board had no way to tell that two of those buildings are
ones they can go inside. The coordinates move to a tiny eagerly-imported
`offices/sites.ts` that the packs import *from*, because a pack is a 25 kB
lazy chunk and the board wants its pins long before anybody opens a door.
A test asserts the pack and the table hold the **same object**, not merely
equal values: a drifted coordinate would put the marker on one building
and the sun on another and both would look entirely plausible.
**Aircraft bank into their turns.** The roll channel existed and was never
written, so every turn was flat. Bank comes from the coordinated-turn
relation against the measured turn rate, damped by a first-order lag so it
settles rather than oscillates, and clamped at 30° like a real limiter.
Six regression tests, because roll is the one channel that feeds itself —
position and heading are recomputed from the last two observations and
wash out a bad value, while a NaN in the roll would persist for the life
of the track.
That fed straight into a real defect: `AdsbFlights` substituted
`heading: 0` for records with no `track` field, which is harmless for a
symmetrical dart and is a **sustained full-scale artefact** once aircraft
bank — a target whose real heading is 200° reported as 0° reads as a 160°
turn and pins the roll at its limiter for as long as it is in the feed.
Those records are dropped now. An aeroplane the feed will not give a
heading for is one this layer cannot draw honestly.
**The office empties out overnight.** A full complement of seated people
at one in the morning, under house lights that came on because the sun is
down, was the least believable thing left in the room once the clock
became real. A live roster always wins — an API that says the building is
empty is telling the truth about the building.
**Robots go somewhere.** They pick real addresses — a seat, a room — and
turn to face the seat when they arrive, rather than stopping at a random
angle. Godmode gets an office section: house lights forced on or off or
following the sun, robots and ceilings toggled, with a readout.
**The bundle is split.** Entry chunk 758 kB to 208 kB, with three.js and
satellite.js in a vendor chunk that survives an app deploy instead of
being re-downloaded on every one. Rollup's 500 kB warning still fires and
should — it now points at three.js, where it is true, instead of at our
code, where it was pointing at three.js all along.
Reviewers caught two false geography claims in the new prose ("both
shipped buildings stand in San Francisco" — one is across the estuary at
Alameda Point) and several miscounted figures. Fixed. In a codebase where
the comments are the design record, those are defects.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
karti
|
a229fb2721 |
The sky gets the things above the aeroplanes
Satellites, end to end: CelesTrak element sets behind the same TTL cache
the weather and the flights use, served as TLEs rather than as positions,
and propagated in the browser with SGP4.
Sending elements is the same trick `flights/plan.ts` plays and it has a
better excuse here — a TLE *is* the closed form, valid for days either
side of its epoch, so one cacheable fetch every six hours replaces a poll
and every viewer agrees about where everything is.
Two things are worth knowing about the shape of it:
- There is no region parameter. An aeroplane at 10,000 m is local and
a satellite at 550 km is above the horizon for a circle two thousand
kilometres across, so one catalogue serves both boards and the client
decides what is above its own horizon. Only the observer is per-city,
which is why `main.ts` shares the elements and rebuilds the catalogue.
- The layer draws on a dome, because it cannot draw anywhere else.
`world.metres(550_000)` is 21,000 scene units against a far plane at
3,000. Azimuth and elevation are real; the radius carries nothing.
Off by default: a clone that started pulling CelesTrak on `npm run dev`
would have volunteered somebody else's bandwidth for its onboarding.
Godmode gets the two dials that point at the sky rather than at the
light — fabricated traffic, which composes with a live ADS-B feed instead
of replacing it, and a switch for the satellite layer with a count beside
it. Both are god-only lies about the inputs, in the manner of the weather
override.
`satellite.js` is the second runtime dependency this package has taken.
Its entry point star-exports an Emscripten build that cannot be shaken
out, so `noWasmPropagator` in the Vite config cuts it: 308 kB of WASM
loader for a bulk propagator nothing calls, against 26 kB for the SGP4
that does the work.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
||
|
karti
|
e41c90fe8d |
Real weather, real aircraft, a heightfield off the main thread, and instruments
Three things that were built and never connected, connected.
**The weather was already there.** `observe()` has always taken a
`WeatherObservation` and `main.ts` has always passed null, so the cloud,
precipitation, visibility and marine-layer paths in atmosphere.ts had never run
outside a test. The server already shipped NWS, met.no and Open-Meteo, all
configured off. What was actually missing was that a single TERA_ORIGIN_LAT/LNG
served one metro and lied to the other — so weather and traffic are per-region
now, derived from the city's own bounds, and the Bay Area gets its fog while
Long Beach gets its own sky. The route takes ?city= or a validated ?lat=&lng=
and refuses to become an open geocoding proxy for the planet.
**The heightfield moved to a Worker.** 2.3 s of blocked main thread at boot, and
another ~950 ms of point-in-polygon on top of it: the park mask is filled in the
worker now, and block placement samples four corners and only runs the exact
test on a cell that straddles an edge — 8 buildings differ out of 185,036.
createScene is async and takes a Stage as a consequence, and there is a
main-thread fallback because "clone it and it works" has no exception clause.
**Spaces is a chunk you fetch when you reach for the door**, not one everybody
downloads. Same for the godmode tools. The entry chunk is 722 kB rather than
772; three.js is most of what is left and splitting it is a different job.
**Godmode is an instrument panel now** rather than one slider: the date and the
season, not just the hour, so the Meeus moon and the sun's seasonal arc become
visible instead of merely correct; a weather override that says on screen when
it is lying; a frame-time and draw-call readout; and a pose editor that emits a
paste-ready Chapter block, which is the thing that makes adding New York cheap.
Two blockers the review caught:
- Every city switch leaked 8 GPU textures — one of them a 2048x2048 shadow map
— and ~10.5 shader programs, and deleteTexture had never been called once in
the app's lifetime. The renderer was being built per scene; it belongs to the
canvas, for the life of the page.
- An upstream fetch that threw rather than returning null skipped the cache
stamp, so the TTL — the only rate limit on outbound calls — collapsed to one
upstream request per inbound request, and the caller got a 500.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|