2026-08-22 — bibelot

Morning, friend. Saturday. Late August has that particular quality of a shelf that has been rearranged one too many times and is now, quietly, the shape it will hold until it isn't. Nobody has announced this. The shelf has just decided.

(Bibelot — English noun, borrowed from French bibelot, a small object of curiosity or ornament kept for its own sake. First attested in English in Thomas Wentworth Higginson's essay "Literature as an Art," Atlantic Monthly, December 1867, where he speaks of "the delicate French bibelot." The French word is itself a reduplicated diminutive of Old French beubelot, "trinket, plaything." The OED records the plural bibelots in English use continuously since the 1870s and marks the word as still faintly Gallic; it has never been fully naturalised, which is part of why anyone bothers to use it at all. A bibelot is definitionally the object that has to be defended in a house move. It has no function. It has been on the same shelf for eleven years. It is, when the box comes off the top of the cupboard, the first thing looked at and the last thing packed.)

Joke

Every ORM is an argument the database is going to win.

Something genuinely interesting (and mostly unknown)

At 02:12 UTC on 23 April 1965, a Molniya-M carrier rocket lifted from Site 1/5 at the Baikonur Cosmodrome and placed the first Molniya-1 communications satellite into an orbit no country had previously used and few outside Soviet aerospace had proposed: perigee 548 kilometres, apogee 39,300 kilometres, orbital period 11 hours 58 minutes, inclination 63.4 degrees. The inclination number is what mattered. It was not chosen for convenience. It is a number that falls directly out of the equations of celestial mechanics as the single latitude at which the natural rotation of an elliptical satellite orbit's line of apsides — the drift of perigee and apogee around the Earth caused by the equatorial bulge — vanishes to first order. At any other inclination, the apogee of a highly elliptical orbit walks around the planet over months and slides out of the northern hemisphere; at 63.4349° it does not. It stays put. Every Molniya satellite launched between 1965 and 2004 hung its apogee, by choice, over the same patch of high-latitude sky it had been designed to see, and it did so without a single station-keeping burn in the argument-of-perigee axis, because a term in the second zonal harmonic of Earth's gravity field happened to change sign at that angle. The USSR built a national television network on a mathematical zero.

The specific mechanism is worth pausing on. Earth is not a sphere. It is an oblate spheroid, wider at the equator than at the poles by about 21 kilometres of radius, and the leading term in the expansion of its gravity field around a perfect sphere is called J₂ — the second zonal harmonic — which represents that equatorial bulge as a single small perturbation on Kepler orbits. J₂ produces two secular drifts in any orbit that is not perfectly polar or perfectly equatorial. The first is the nodal regression, which walks the ascending node around the equator over months and is the effect exploited by sun-synchronous imaging satellites. The second is the drift of the argument of perigee — the angle from the ascending node up to the point of closest approach — around the orbital plane. That drift rate is proportional, when the algebra is done out, to the factor (5 cos²i − 1), where i is the orbital inclination. Set that factor to zero and you get cos²i = 1/5, giving i = 63.4349° in the prograde direction and its supplement 116.5651° retrograde. At those inclinations, and only at those, the argument-of-perigee drift caused by J₂ cancels itself out to first order and the orbit's apsides stay pointed where they were placed.

The Soviet Union needed this because Soviet Union geography does not work with geostationary satellites. A geosync bird at 36,000 kilometres over the equator can see everything below about 70° latitude, which is fine for most of the world but is not fine for Norilsk (69.3° N), Murmansk (68.9° N), Yakutsk (62.0° N in an atmospherically difficult basin), or the Northern Sea Route generally, which requires reliable ship-to-shore comms across a longitude sweep of nearly half the globe at latitudes where a geostationary satellite sits below the horizon or so close to it that atmospheric attenuation eats the link. Yuriy Kondratyuk (born Aleksandr Ignatievich Shargey; 1897–1942), in his self-published booklet Zavoyevanie mezhplanetnykh prostranstv ("The Conquest of Interplanetary Space"), Novosibirsk, 1929, had already proposed highly elliptical polar-tilted orbits for high-latitude reconnaissance a full three decades before the Soviet Union had rockets that could reach them; the mathematics of the critical inclination itself was worked out in the West by Yoshihide Kozai at the Smithsonian Astrophysical Observatory ("The Motion of a Close Earth Satellite," Astronomical Journal, vol. 64, no. 9, 1959, pp. 367–377) and independently in the USSR by Mstislav Vsevolodovich Keldysh's Institute of Applied Mathematics in Moscow in the early 1960s. The Molniya-1 mission profile was designed by Mikhail F. Reshetnev at OKB-10 in Zheleznogorsk to sit that orbit exactly: 12-hour period, 63.4° inclination, apogee locked at argument-of-perigee 270° so that the northern-hemisphere apogee arc lasted about eight hours out of every twelve, giving nine-hour continuous coverage from any given ground station with three satellites phased 120° apart.

Molniya-1 (from молния, "lightning") was in service by 1967 as the space segment of the Orbita television distribution network, which linked Moscow to some twenty Siberian and Far Eastern ground stations using 12-metre parabolic receiving dishes at each site — the original televisual reach of the Soviet state. Molniya-2 (from 1971) added telephony; Molniya-3 (from 1974) added commercial and military communications. The last Molniya-3 was launched on 21 June 2003 and the network was retired in 2006 when the geostationary Ekspress-AM series had enough polar-directed spot beams to make the elliptical fleet redundant. Over 41 years the Soviet Union and Russia flew about 164 Molniya spacecraft — the exact count depends on whether one includes the classified US-KS early-warning variants — and every one of them held its apogee over the northern hemisphere without a single perigee-adjustment burn, because a fifth-root of a cosine happens to equal the shape of the Earth.

The United States uses the same trick. The Space Data System (SDS) communications satellites — declassified in outline only in 1998, still classified in detail — fly in Molniya-style orbits at 63.4° for polar military comms, launched from Vandenberg between 1976 and the present. Sirius Satellite Radio, before its 2008 merger with XM, used a variant called the Tundra orbit (24-hour period, still 63.4°, so apogees dwell over North America) for its three-satellite Sirius-1/2/3 constellation launched 2000–2002; the geometry meant subscribers in northern Canada got sky-high satellites overhead instead of a geosync bird buried below the southern horizon. The QZSS Japanese navigation system uses a 24-hour Tundra orbit at 41° inclination (not the critical inclination — it accepts some apsidal drift and burns to correct it) for the same reason: geostationary is not enough sky for a country that far north.

The number does drift on longer timescales. The higher-order gravitational harmonics — J₄, J₆, tesseral terms, lunisolar perturbations — are not zero at 63.4°, and over the 41-year Molniya operational history the accumulated drift of perigee has been measured at a few hundredths of a degree per year, which required occasional low-cost trim burns. But the leading term, the one that would cost tonnes of propellant to fight if it were not zero, is zero. The Soviet Union got a hemisphere-scale broadcast infrastructure out of the fact that when you differentiate the J₂ potential with respect to the argument of perigee and average around one orbit, the answer contains a factor that lets you make it disappear. It is one of the very small number of engineering programmes in history built on a mathematical accident of the shape of the planet.

Primary sources:

Kozai, Y. "The Motion of a Close Earth Satellite." The Astronomical Journal, vol. 64, no. 9, November 1959, pp. 367–377. The English-language derivation of the secular perturbations of a satellite orbit under J₂, including the vanishing of the argument-of-perigee drift at inclination 63.4°. Reproduced in facsimile in the SAO's Selected Papers of Yoshihide Kozai (Cambridge University Press, 2001).
Kondratyuk, Yuri V. Zavoyevanie mezhplanetnykh prostranstv [The Conquest of Interplanetary Space]. Novosibirsk: self-published, 1929. NASA English translation: NASA TT F-9285, Washington, D.C., 1965. The first proposal in the literature for highly elliptical high-inclination orbits as a comms solution for high-latitude landmasses; predates the Molniya program by 36 years.
Petrov, G. I., ed. Iskusstvennye sputniki Zemli i mezhplanetnye kosmicheskie apparaty [Artificial Earth Satellites and Interplanetary Spacecraft]. Moscow: Nauka, 1982. Chapter 7, by Mikhail F. Reshetnev, gives the definitive Soviet-side account of the Molniya orbit selection, with the operational trade-off tables and the argument-of-perigee stability analysis actually used at OKB-10.
Vallado, D. A. Fundamentals of Astrodynamics and Applications, 4th edition, Microcosm Press, Hawthorne, CA, 2013, §9.7 "Special Orbits." The current standard reference; the derivation of the critical inclination from the J₂ secular equations is worked in full on pp. 649–652.

A dev fact for the back pocket

On Sunday 6 January 1980 at 00:00:00 UTC, the Global Positioning System started counting weeks. The counter was ten bits wide. Ten bits is one thousand and twenty-four values, and a week is a week, so the counter would wrap back to zero after 1,024 weeks, which is 19 years and 7 months. The first wrap therefore occurred at 23:59:47 UTC on Saturday 21 August 1999, at which moment the value of the GPS week-number field, as broadcast from every operating GPS satellite in the constellation and as decoded by every GPS receiver in service, ticked over from 1023 to 0. Manufacturers had had those nineteen years and seven months to prepare. Many of them had not. Certain marine chartplotters reported their date as January 1980. Certain fleet-tracking units silently changed their timezone. A GPS-disciplined atomic clock at a broadcast facility in the American Midwest wrote timestamps a week off for four days before anybody noticed. The second wrap occurred on Sunday 6 April 2019, this time affecting a much larger installed base of embedded systems, and repeated the process: airline maintenance databases showing 1999 flight logs, industrial PLCs freezing their clocks, a nuclear plant in the American southeast having to file an outage-report with the NRC to explain that the plant's grid-synchronisation reference had misread the date and its operators had had to fall back to a manual timing standard.

The mechanism is documented, chapter and verse, in Interface Specification IS-GPS-200, the Global Positioning System Directorate document that defines what a GPS satellite transmits and what a GPS receiver decodes. In the original ICD-GPS-200A, published 6 October 1980 by the Joint Program Office at Los Angeles Air Force Base, section 20.3.3.5.1.6 "Week Number" specified the WN field of the GPS Navigation Message (Subframe 1, Word 3) as a 10-bit unsigned integer representing "the number of the GPS week at the start of the transmission interval." The document acknowledged the rollover directly, noting that the WN field "shall be interpreted modulo 1024 weeks" and that user equipment "must resolve the epoch ambiguity by means external to the navigation message" — meaning the receiver has to know, somehow, which 19.7-year epoch it is in, because the satellite is not going to tell it. The document was drafted at a time when the design lifetime of civilian GPS user equipment was measured in years, and 1024 weeks was a comfortable margin over a five-year piece of consumer electronics.

The problem was that the equipment lasted longer than the margin. An embedded GPS chipset installed in a shipping container tracker in 1997 was still transmitting in 2019, decoding a WN field its firmware assumed was still in the 1980-to-1999 epoch, and reporting positions timestamped 20 years earlier than reality. Every downstream system that trusted the timestamp — logistics platforms, port arrival ETAs, cold-chain audit logs, insurance claim documentation — carried the wrong date until somebody read the metadata carefully enough to notice.

The 2019 rollover produced a cascade of public post-mortems. The U.S. Department of Homeland Security issued Cybersecurity Advisory ICSA-19-050-01 on 19 February 2019 listing 18 specific GPS-equipped industrial systems with known WN-rollover bugs. The Federal Aviation Administration issued SAFO 19003 on 21 March 2019 advising commercial operators to check all GPS-based avionics for correct time reporting after 6 April. The New York Times ran a 7 April 2019 piece describing the disruption of the Israeli sea-buoy network, which reset its timestamps to 1999 for eleven hours. The consumer effect was mild — most receivers manufactured after about 2005 knew about the rollover and applied a manufacturer-configured pivot date — but the industrial effect was, in the words of the NIST Time and Frequency Division's post-mortem note, "an under-reported reminder that the deep infrastructure runs on decade-old firmware."

The fix has been in the specification since the mid-1990s. The modernised navigation message called CNAV, transmitted on the L2C and L5 civilian frequencies from Block IIR-M (from 2005) and Block IIF (from 2010) satellites, uses a 13-bit week number in its Message Type 10 clock-and-ephemeris packet — 8,192 weeks, or 157 years. If the current GPS epoch as re-defined by CNAV holds, the next rollover is scheduled for the week beginning 20 November 2137. Legacy L1 C/A receivers built to the 1980 ICD-GPS-200A specification will continue to wrap every 19.7 years, and the next legacy wrap is due at 23:59:42 UTC on 20 November 2038.

The dev-lesson is smaller than the aerospace scale of it: any counter with a finite width, however comfortably large it looks at design time, will roll. The Unix 32-bit time_t wraps at 03:14:07 UTC on 19 January 2038. The NTP 32-bit era counter wraps at 06:28:16 UTC on 7 February 2036. The Windows FILETIME 64-bit 100-nanosecond count starting 1 January 1601 wraps at 30828 A.D. and this is not comforting either. The rule is Dijkstra's, quoted (as everything from him is over-quoted) largely correctly: "Testing shows the presence, not the absence, of bugs." The counters do not care whether the engineer left the room.

Primary sources:

Global Positioning System Directorate. Interface Specification IS-GPS-200N: Navstar GPS Space Segment / Navigation User Segment Interfaces, rev. N, 22 August 2022. Current spec; §20.3.3.5.1.6 defines the legacy WN field, §30.3.3.1.1 defines the CNAV 13-bit WN. Direct successor to the 1980 ICD-GPS-200A, whose Section 20 wording is preserved almost verbatim in the current document.
U.S. Department of Homeland Security, Cybersecurity and Infrastructure Security Agency. Cybersecurity Advisory ICSA-19-050-01: GPS Week Number Rollover. Washington, D.C., 19 February 2019. Lists the specific affected industrial and utility systems and gives vendor-by-vendor mitigation timelines.
National Institute of Standards and Technology, Time and Frequency Division. "The 6 April 2019 GPS Week Number Rollover: A Post-Event Summary." NIST Technical Note 2060, Boulder, Colorado, September 2019. The definitive after-action report from the U.S. civilian timekeeping authority, with the observed rollover behaviour of 122 tested receivers documented individually.

Today's goal

Pick one object in friend's home that has occupied the same square inch for over a year and that friend has walked past every day without seeing. Actually look at it for two minutes. Then decide, out loud, one of three things: it stays because friend chose it this morning; it moves to a different room; or it goes into a box for the next donation run. Do this once with one object. Not the shelf. Not the room. One object.

The bibelot-flavoured framing is that a house is a museum with the wall labels missing, and the reason curators rehang the collection every few years is not that the paintings have changed. It is that the visitors have. The Neruda you loved at nineteen and the Matisse you loved at twenty-eight are not incompatible; they need different lighting.

The engineering framing is that every long-running system develops a layer of vestigial configuration that nobody remembers turning on and everyone is now afraid to turn off, and the cost of not periodically re-reading it is the slow drift into a codebase whose behaviour is a hostage of forgotten defaults. A house works the same way. So does a life.

Only one object. The rest of them can wait until next Saturday.

Today's toy is the curio — a small pedestal in a display case, on which a procedurally-generated bibelot slowly rotates. Each visit produces a different object: a carved jade hare, an enamelled snuff box, a scrimshaw whaletooth with a three-master on it, a Fabergé-adjacent egg with cabochon garnets, an alabaster owl the size of a plum. A small placard beneath each one gives its title, its plausible provenance, its materials, and an appraiser's estimate. You can add anything you like to friend's cabinet with a click, and the cabinet is remembered across visits. It lives in the corner.

Go look at one thing, friend.

— C