validation.md

Validation

What is tested, against what, and how closely. All of these run with npm test; the numbers below are the assertions in the test files, not estimates.

The question these tests answer is "does OrbitView add error on top of SGP4?". They don't measure SGP4 or TLE accuracy itself — for that, see Vallado et al. 2006 and the TLE age warning in the app.

Frames and time — src/lib/__tests__/science-regressions.test.ts

What	Reference	Assertion
GMST every 30 min for a day	satellite.gstime	equal to 1e-12 rad
Geodetic round trip at 89.9999°, 90°, −90°	geodeticToEcf	height to 1 mm, latitude to 1e-4°
Mean elements of the ISS	its TLE	period, inclination, epoch match

Doppler

What	Reference	Assertion
Range rate at 20 times over 2 h 20 min	central difference of range, ±0.5 s	within 1 m/s
Geostationary satellite seen from Istanbul	≈ 0 by definition	below 20 m/s
Elevation and horizon flag	—	consistent

The second case is the one that caught the v3 bug: without the ω × r term a GEO satellite showed about 3 km/s of range rate, roughly 100 kHz at 12 GHz.

Passes

What	Reference	Assertion
AOS and LOS of every full pass in 24 h	1-second brute-force scan	within 2 s
Maximum elevation	5-second samples across the pass	none higher than reported + 0.01°
Pass already in progress at window start	—	flagged, starts at the window
Pass still in progress at window end	the same pass in a wider window	same LOS to 2 s
"Sunlit" flag over three days	5-second scan of sunlight and sky darkness	any sunlit stretch over 30 s is flagged; nothing in shadow is
Short passes between scan samples (ISS, 20° mask, 2 days, 4 scan phases)	2-second brute-force scan	every pass found, rise within 3 s

Earth's shadow — science-regressions.test.ts

What	Assertion
Day side, on the axis behind the Earth, edges at GEO distance	sunlit / umbra / penumbra as the cone geometry says
ISS over one orbit	~35 min of umbra, 3–20 s shorter than the old cylinder
Geostationary satellite at the March equinox	60–75 min of umbra

Lifetime

What	Assertion
B* parsing, including negative and malformed fields	exact
ISS-like orbit	between 2 months and 5 years, range brackets the estimate
Lower perigee	shorter lifetime
B* = 0	"drag not measurable"
1200 km perigee	beyond the 25-year horizon
Badge	weeks / months / years / long-lived from the lifetime, not the perigee

Close approaches

What	Assertion
ISS vs. GEO	rejected by the apogee/perigee filter
ISS vs. the same orbit 0.01° behind	found, miss distance < 5 km, GEO not reported

Rendering and new features — v4-features.test.ts

What	Assertion
Worker's inline TEME→ECF rotation vs. stateAt	under 1 m
Worker only touches the visible indices; failed propagation	untouched / NaN
Frame-rate rule	0 when paused, 500 ms zoomed out at ×1, 66 ms at high speed, ≥ 200 ms on battery saver
Device tier	static hints and measured medians classify as expected
Calendar export	UTC times, escaping, 75-character folding, reminder
SATCAT parsing	decayed objects dropped, quoted names, radar size
Night window (Istanbul)	Sun below −6° just inside both edges, above it just before dusk
Shared pass cache	one calculation per key; previous result shown while a new window loads

Catalog labels — catalog.test.ts

What	Assertion
Space stations	exactly the ISS and Tiangong, not modules, vehicles, cubesats or "ISS DEB"
Regime	geostationary needs i < 5°; BeiDou IGSO and QZSS are MEO; Molniya HEO; broken lines don't crash
Debris	"OBJECT A" is not debris; SATCAT's object type wins over the name
Chip counts	LEO + MEO + GEO + HEO + debris = everything, with and without SATCAT
GNSS	real catalog names of GPS, GLONASS (incl. K), Galileo, BeiDou, QZSS, NavIC

Data layer — tle.test.ts, db.test.ts, SpaceWeather.test.ts, route.test.ts

• IndexedDB read failure falls back to the network; write failure still returns the downloaded data.
• A partial or stale refresh never overwrites the full cached catalog.
• Duplicate catalog numbers get stable suffixes.
• The API returns 400 for unknown groups, 502/503 (not someone else's data) when the upstream fails, flags stale responses, and times out.
• Kp parsing handles both NOAA JSON layouts; failed fetches are flagged as fallback values rather than shown as live.