@rr0/ufoathome 0.35.0 → 0.36.0

package/README.md

@@ -168,12 +168,17 @@ through the ordinary pinhole projection (`src/engine/instrument/`, `src/render3d
 A camera also has a FORMAT, which is part of what a photograph carries as evidence: an instrument
 states the image it exposes and the lens in front of it, in millimetres, so the ratio is the shape of
 the picture (a square 126 frame, a phone held upright) and `2·atan(h/2f)` is the field it takes in.
-And it has SETTINGS, each kept on the pose beside the heading and the pitch, each read-only where the
-device fixed it (an Instamatic's owner had one aperture, one shutter speed and one focal length):
+And it has SETTINGS, each read-only where the device fixed it (an Instamatic's owner had one aperture,
+one shutter speed and one focal length). The aperture, the focal length and the focus sit on the POSE
+beside the heading and the pitch — the same camera is set differently from one photograph to the next
+— while the shutter belongs to the RECORDING as a whole (`Sighting.exposureSeconds`): one observation
+was photographed one way, and a shutter speed that changed halfway through would be a second
+photograph rather than a moment of this one. The four:
 the focal length, which is the field written the way a photographer writes it; the aperture, which
 decides both the depth of field and whether the diaphragm's blades throw a star at all; the shutter,
-which accumulates a moving light into a STREAK and a blinking one into a dashed streak; and where the
-lens was focused. Two of them are drawn rather than merely stated — the blur (`DepthOfFieldPass.ts`,
+which accumulates a moving light into a STREAK and a blinking one into a dashed streak — and, once the pose is
+long enough, turns every star into the arc the turning Earth draws (see [A pose long enough draws the
+sky](#a-pose-long-enough-draws-the-sky)); and where the lens was focused. Two of them are drawn rather than merely stated — the blur (`DepthOfFieldPass.ts`,
 from the thin-lens geometry in `DepthOfField.ts`) and the streak — because that is what makes them
 evidence a reader can compare against a photograph: a sharp object bounds its own distance, and a
 blurred one in a sharp frame was close.
@@ -634,12 +639,24 @@ car headlights, car hazards, emergency beacons, streetlamp. A catalogue of speci
 never more code. Nothing about this is aircraft-specific: what dots a long exposure for an airliner's beacon dots
 it for a car's hazards too, at a different rate.
 
-That rate is the point. On a long exposure the spacing of the dots along a streak is the flash rate times the
-object's angular speed, which is exactly how a photograph of a passing airliner is told from a photograph of
-something that does not blink — and it is why the model exposes `lightOnFractionBetween` rather than only "is it
-on?". A wingtip strobe is lit for a hundredth of its cycle; sampled instant by instant it would be missed almost
-every time, and the dots that did appear would be an artefact of the sampling rate. Integrating the fraction of
-each interval is exact however coarsely it is sampled.
+That rate is the point, and it is now DRAWN. On a long exposure the spacing of the dots along a streak is the flash
+rate times the object's angular speed, which is exactly how a photograph of a passing airliner is told from a
+photograph of something that does not blink — and it is why the model exposes `lightOnFractionBetween` rather than
+only "is it on?". A wingtip strobe is lit for a hundredth of its cycle; sampled instant by instant it would be
+missed almost every time, and the dots that did appear would be an artefact of the sampling rate. Integrating the
+fraction of each interval is exact however coarsely it is sampled.
+
+The photograph this reproduces is Gennevilliers, 5 November 1990 (`/science/crypto/ufo/enquete/meprise/aeronef/
+avion/`): an airliner on a ten- or thirty-second pose, published as a monumental craft — its steady lamps drawing
+LINES and its flashing ones DOTS AT REGULAR INTERVALS. Reconstructed here at 6.7 km on a 20 s pose through a 50 mm
+lens, an aircraft leaves an 800 px streak carrying 21 dots a median 41 px apart, where 60 flashes a minute over
+20 s across 800 px predicts one every 40. `ExposureSampling` is what makes that possible: the sky drifts a single
+pixel in ten seconds and would ask for two instants, so the pose is sampled instead for what MOVES in the scene
+(an instant per two pixels of travel) and for what FLASHES in it (two instants per flash, which is what tells one
+dot from the next rather than dotting the line at the sampler's own rate). A lamp is also written in real units —
+`LAMP_RADIANCE`, about twenty times white for a position light, and `DecorLight.intensity` as a ratio of peak
+candela (a wingtip strobe really is some twenty times one) — because a pose spreads a lamp's light over hundreds
+of pixels, and at white a whole aircraft trail came out at a thousandth of white, i.e. invisible.
 
 An aircraft in a scene is a **hypothesis**, not testimony — "here is what a flight at that altitude and heading
 would have looked like" — and belongs to the decor for that reason, next to the buildings and trees whose
@@ -674,6 +691,63 @@ One residual worth naming: an angular extent is stored as its *on-axis* value, a
 sits. For an object 9° off-axis subtending 9°, that is about 2.5% out. The overlay draws axis-aligned boxes and
 cannot express more; it is a fifth of the error it replaces, and it shrinks towards the centre of the frame.
 
+#### A pose long enough draws the sky
+
+The shutter accumulates the object (`UfoElement.exposureInstants`), and past a certain length it accumulates the
+**sky** too: the Earth turns under it at 15.041° an hour — a sidereal day, not a solar one — so every star is drawn
+out into the arc a tripod really records. A photograph of "lights that moved" is quite often exactly this, the
+lights having held perfectly still while the camera did not.
+
+`SkyDrift` (`src/engine/astronomy/SkyDrift.ts`) states what the SKY did, in pixels, and how many instants that
+takes to draw — `ExposureSampling` (`src/engine/model/ExposureSampling.ts`) answers the same question for what
+stands against it (see [Decor that moves](#decor-that-moves-and-lights-that-blink)), and the pose is drawn at
+whichever asks for more: one per pixel of the longest trail, so the arc lands on touching pixels instead of dashing, and
+**one** — no accumulation at all — whenever the sky moved less than a pixel, which is every ordinary frame (a
+snapshot renders in a tenth of a millisecond and never comes near this). Each instant is a whole sky RECOMPUTED at
+its own moment (`SceneElement.applySceneAt`, pushed through `SceneRenderer.setExposure`) rather than one sky nudged
+sideways, so the arcs curve towards the pole exactly as they should and the Moon and the planets travel their own
+way through the frame. `ExposureAccumulation` adds them on a half-float film in linear light and bends the sRGB
+curve only once, at the end — the same rule `colorSpace.ts` states, and the reason both fullscreen passes can now
+be told not to encode what they hand to it.
+
+The pose itself is one setting for the whole recording (`Sighting.exposureSeconds`, held to the device's own range),
+not a keyframed one: what the shutter did is how this observation was photographed, and a value that varied along
+the timeline would make the same recording two different photographs at two instants. A file written while it lived
+on each pose is read back through the first pose that stated one.
+
+A pose is drawn in two moves, because a sky costs about 8 ms to rebuild and a five-minute pose is 37 of them: the
+**viewfinder** immediately (one instant, a twentieth of a millisecond) and the **photograph** as the scene settles,
+a dozen milliseconds of instants per animation frame, shown as the film fills and gained up to stay properly
+exposed — so it goes from beady to smooth rather than from black to bright. Anything the witness does interrupts it
+and starts it again, which is why the editor stays answerable while a long pose is on. Doing it all inside one call
+is what made it crawl: the dozen setters a single tick touches would each rebuild the whole pose, and the per-frame
+animation loop (twinkle, rain, lightning — none of which survives a pose of minutes anyway) restarted it sixty
+times a second, so the picture never finished at all.
+
+Each instant carries its share of the light rather than its whole, so the picture stays exposed as it was taken and
+the movement shows as a trail. That has a consequence worth stating plainly, because it is the physics and not a
+shortcoming: a trail is FAINT, and the longer it is the fainter it gets — one star's light spread over more and
+more pixels. Measured against a night sky of 30.5: a first-magnitude star peaks at 181 as a point at 1/250 s, 52
+after ten minutes, 39 after thirty, 35 after an hour. This project draws what the pose collected, not what would
+read well.
+
+Only the 35 mm SLRs offer poses like that, and they offer them because they really had them: **B**, where the
+shutter stays open as long as it is held. An Instamatic had one shutter speed and a phone's night mode stops at ten
+seconds — neither can draw a trail, and neither is offered one: a pose typed past what the device could do is
+brought back inside its range rather than kept, since it would put a trail in the picture that the camera named on
+the same recording could never have drawn.
+
+Two consequences of drawing a pose rather than an instant, both of which cost an evening to find:
+
+- **What is on screen is not where the object is at the playhead**, so that is what a pointer has to be aimed at.
+  `UfoElement.shapeAt` hit-tests every instant of the pose, and without it an object with a ten-second pose became
+  impossible to select at all — the click fell straight through the streak onto the landscape behind. The selection
+  handles stay on the playhead's own instant: where they sit against the streak is the answer to "which moment am I
+  editing".
+- **How finely the pose is sampled is a distance, not a duration.** Counting time alone (a painting every fiftieth
+  of a second) leaves a fast object visibly BEADED — 48 paintings across three hundred pixels land six apart, and a
+  reader sees the paintings instead of the streak. The object's own travel decides it, one instant per two pixels.
+
 #### Where meters do come back
 
 The only real distance a testimony can support is an **inequality**, and only where the witness saw the object