2026-07-21 — fossick

Morning, friend. Deep into July. The kind of morning where the air conditioner in the office was already on when you got there, quietly, since sometime yesterday.

(Fossick — verb, Australian and New Zealand English, first attested in the Melbourne Argus around 1852 to describe gold-rush prospectors picking over abandoned diggings for missed ore. The etymology is uncertain and probably from the English dialect fussack — "to bustle, to rummage" — carried out to Victoria by Cornish and West Country miners. The word has since generalised into the everyday sense of searching for something in a small, close, painstaking way, which is the exact mental motion of grepping a legacy repo for the specific commit where a comment used to say something helpful.)

Joke

The dashboard has been green since November. The query it runs targets a table that was dropped in October.

Something genuinely interesting (and mostly unknown)

On the afternoon of 17 May 1973, at a press conference held in a ballroom of the Steigenberger Hotel in Bonn, an East German-born historian named Arno Peters unveiled a rectangular equal-area world map of his own design. He described it as "die neue Weltkarte" — the new world map — and he framed the project, explicitly, as a political correction to the Mercator projection, which he argued flattered the temperate colonial powers by inflating the size of their landmasses. Peters's map preserved area exactly, at the cost of shape: it elongated the tropics into pale vertical ribbons and squashed the polar regions into wide flat sheets. Africa, on his map, appeared roughly the size of the continental United States plus China plus India combined — because that is, in real square kilometres, roughly what Africa is.

Coverage in Der Spiegel followed within a week. Within a year the Peters map had been picked up by the Lutheran World Federation (1974), Oxfam and Christian Aid (across 1975), the World Council of Churches (1978), and, eventually, UNICEF and several UN development agencies. Print runs by the mid-1980s were in the low millions. It was a genuine hit outside the discipline. It was the map that development NGOs put on the wall behind their fundraisers.

The cartographic profession's response, when it arrived, was incandescent.

Arthur H. Robinson, professor of geography at the University of Wisconsin–Madison and the leading North American academic in map-projection theory, published a rebuttal in The American Cartographer in 1985. He raised three separate objections. First, that Peters had wildly misdescribed the Mercator projection's actual use in modern cartography — Mercator had been marginal for anything except marine navigation since the mid-twentieth century, and no serious atlas of the postwar period had presented it as a general-purpose reference map. Second, that Peters's rendering of the continents was, aesthetically, an atrocity: continents that in Robinson's own phrasing (American Cartographer 12(2), 1985, p. 104) resembled "wet, ragged, long winter underwear hung out to dry on the Arctic Circle." Third and most damaging, that the projection was not new. The same equal-area cylindrical projection had been published a hundred and eighteen years earlier by the Reverend Dr. James Gall of Edinburgh, in a paper delivered to the British Association for the Advancement of Science at its Glasgow meeting of September 1855 and reprinted in the Scottish Geographical Magazine. Peters had, at best, independently rediscovered a projection whose trigonometric derivation was already in the record; at worst, he had rediscovered it while sitting in a library that had a copy of Gall on the shelf. The projection is referred to in every subsequent technical publication as Gall–Peters.

The dispute escalated. In 1989, seven major North American cartographic organisations — the American Cartographic Association, the National Council for Geographic Education, the Association of American Geographers, the National Geographic Society, the American Congress on Surveying and Mapping, the Canadian Cartographic Association, and the Special Libraries Association's Geography and Map Division — issued a joint resolution against all rectangular world maps. The resolution, printed in Cartographica 26(3–4), 1989, is one of the very few occasions on which the discipline's major professional bodies have collectively told the general public to stop using something they liked. It is written in the careful register of a group of academics who have been arguing with the same well-meaning laity for over a decade and are exhausted. The tell is a line about how "no rectangular map is a satisfactory representation of the entire earth" — the standard cartographer's response to the entire genre of the argument, delivered as though for the last time.

The dispute reached its widest audience through a 2001 episode of The West Wing — "Somebody's Going to Emergency, Somebody's Going to Jail," aired 1 February 2001 — in which the fictional Organization of Cartographers for Social Equality briefs the White House press secretary on the Gall–Peters projection and its political implications during a Big Block of Cheese Day. The scene, written by Aaron Sorkin, is largely faithful to Peters's original argument and entirely omits the cartographic profession's response. It is, to this day, the most widely-encountered exposure to map-projection theory in the English-speaking public, and every geography professor in North America has spent a class explaining to a student that the scene is missing the last twenty-five years of the story.

Peters died in 2002. Robinson died in 2004. Their respective projections — Peters's Gall–Peters and Robinson's own compromise projection of 1963 — remain in general use, in different contexts, and are quietly agreed by working cartographers to be both fine, provided the reader knows what a map is for and what it is doing. The current professional recommendation, when a general-audience presentation calls for a whole-earth map, is the Winkel Tripel projection published by Oswald Winkel in 1921 and adopted by the National Geographic Society as its standard reference in 1998. It is neither equal-area nor conformal and it distorts nothing spectacularly. It is the projection you have seen on the back of an atlas for the last twenty years, and it exists because everyone else's map had a fight.

Primary sources:

• Arno Peters, Die neue Kartographie / The New Cartography, Universum Verlag / Friendship Press, 1983 (bilingual edition), ISBN 0-377-00147-5. The book-length version of the 1973 press-conference argument. The English translation is by Ward Kaiser.
• Arthur H. Robinson, "Arno Peters and His New Cartography," The American Cartographer 12(2), pp. 103–111, October 1985. The definitive academic rebuttal; contains both the Gall priority documentation and the "long winter underwear" line.
• James Gall, "On Improved Monographic Projections of the World," Report of the British Association for the Advancement of Science, Glasgow meeting, pp. 148–149, September 1855; reprinted in the Scottish Geographical Magazine vol. 1, 1885, pp. 119–123. The original 1855 publication of the projection Peters would later be credited with. Two pages. Trigonometry all present and correct.
• American Cartographic Association et al., "Geographers and Cartographers Urge End to Popular Use of Rectangular Maps," Cartographica 26(3–4), pp. 79–80, Winter 1989. The joint resolution.
• Mark Monmonier, Rhumb Lines and Map Wars: A Social History of the Mercator Projection, University of Chicago Press, 2004, ISBN 0-226-53431-6. The book-length social history; chapters 8–10 cover the Peters affair with sourced interview material from participants who were still alive when Monmonier began the project in 2000.

The Monmonier book is the one to read if there is only time for one. The Robinson paper is the one to read if there is time for two.

A dev fact for the back pocket

Ariane 5 Flight 501 was destroyed thirty-seven seconds after launch by a floating-point-to-signed-16-bit-integer conversion inside a subsystem that was not needed for flight.

On 4 June 1996, the maiden flight of the European Space Agency's Ariane 5 launcher lifted from the ELA-3 pad at the Centre Spatial Guyanais in Kourou, French Guiana. Onboard were the four Cluster magnetospheric-research satellites, and behind them roughly a decade of ESA development at a programme cost estimated at the time as several billion dollars. At about T+36.7 seconds, the vehicle veered sharply off vertical, tore itself apart under aerodynamic load, and was destroyed by the flight-termination system at around T+39 seconds. The debris field, spread across mangrove immediately east of the pad, was recovered and reassembled in a hangar over the following days.

The failure was traced to the Inertial Reference System — the Système de Référence Inertielle, or SRI — the box that told the on-board computer which way was up. There were two SRIs on Ariane 5, running identical software, cross-strapped for redundancy.

The SRI software was directly inherited from Ariane 4, which had by June 1996 been flying operationally for eight years. The inquiry report describes the inheritance as a deliberate reuse decision: the module had run reliably across the Ariane 4 launch history and was believed to require no further qualification. The specific piece of inherited code was a background alignment routine which computed the vehicle's horizontal bias and continued to run for a fixed period after liftoff, so that a scrubbed launch could be recovered without a full SRI reset. On Ariane 4 this timing was harmless. On Ariane 5, whose trajectory was substantially steeper and whose horizontal velocity built up much faster in the first minute of flight, it was not.

At T+36.7 seconds, the alignment routine attempted to convert a 64-bit IEEE-754 floating-point value representing the horizontal bias — a value that had by then exceeded the maximum representable positive 16-bit signed integer (32,767) — into a 16-bit signed integer variable, in Ada, without a protective range check. The runtime raised an Operand_Error exception. The exception was uncaught. The active SRI halted and put a diagnostic bit pattern onto the databus in place of flight data. The hot-standby SRI, running the identical software with the identical inputs, had raised the identical exception a fraction of a second earlier and had already halted. The on-board main computer received the diagnostic bit pattern, interpreted it as legitimate attitude data, and commanded full deflection of the boosters' hydraulic nozzles to counter what it believed was a violent tumble. The commanded deflection then produced the actual tumble. Aerodynamic breakup followed within about a second of the first nozzle command.

The Board of Inquiry was chaired by Professor Jacques-Louis Lions of the Collège de France, one of the founders of the modern French school of numerical analysis and at the time president of the French Academy of Sciences. The board's report, delivered on 19 July 1996 under the title Ariane 5 Flight 501 Failure — Report by the Inquiry Board, reads at twenty pages as one of the calmest engineering documents in the modern software-safety literature. Its findings, in summary: the reused code was correct for Ariane 4, and had been correct for Ariane 4 across the whole of that vehicle's operational history. It was not requalified against the Ariane 5 flight envelope. Nobody had asked whether the horizontal-bias value could exceed 16 bits under the new trajectory, and the specification of the SRI did not require the question to be asked, and so the question was not asked.

The report is worth reading directly. Its central conclusion is that the failure was caused by a specification error rather than a coding error — the Ada conversion did exactly what it was written to do; the software's authors had never been told about the trajectory in which their code would be asked to do it. The report also notes, in a passage that has been quoted in every subsequent introductory course on software safety, that the module that failed was, at the moment it failed, computing a value that was not needed for flight and was not being read by anything on the vehicle. Its only consumer was a ground-side alignment console that had been disconnected before liftoff. The module was still running because nobody had gone into the SRI software to remove it, and nobody had gone in to remove it because it was believed to be inert.

Subsequent revisions of Ariane 5, from Flight 502 onward, cut the post-liftoff continuation of the alignment routine and added a full audit of the Ada pragma Suppress directives across the flight software, with a formal proof obligation for every remaining suppression. The vehicle went on to a launch history that ran through July 2023, with a reliability figure in the mid-nineties percentile — one of the highest ever fielded by any orbital launcher. The reused module was corrected; the reuse practice was corrected; the specification-review process was corrected; and every graduate software-engineering course in Europe from 1997 forward has used the Lions report as its case study for the sentence "software inherited without full requalification against the new operational envelope is not tested software."

Primary sources:

• Jacques-Louis Lions et al., "Ariane 5 Flight 501 Failure — Report by the Inquiry Board," ESA/CNES, Paris, 19 July 1996. Twenty pages. Available in facsimile from the ESA public archive and mirrored at sunnyday.mit.edu/nancy-lynch/ariane5rep.html. The single most-cited engineering post-mortem of the last thirty years.
• Jean-Marc Jézéquel and Bertrand Meyer, "Design by Contract: The Lessons of Ariane," IEEE Computer 30(1), pp. 129–130, January 1997. The condensed English-language software-engineering interpretation, published within seven months of the report.
• Nancy Leveson, Engineering a Safer World: Systems Thinking Applied to Safety, MIT Press, 2011, ISBN 978-0-262-01662-9. Chapter 2 §2.4 contains the definitive academic re-analysis, situating the Lions report within the broader field of system-safety engineering.

If friend is inheriting a subsystem into a new vehicle — a service, a codebase, a hardware platform, a call site — the Lions report is the twenty pages to read before the merge.

Today's goal

Send one text message to someone you haven't heard from in three months or more.

No favour to ask. No news to share. No "sorry it's been a while." Six words if you want six: thinking of you, how are you. The friction is entirely in the opening — once the message is out, the reply-thread does its own work, and you will find on the far side of it that the other person was pleased to hear from you and has been meaning to text you and had also been finding the opening awkward.

Nobody who receives an unsolicited thinking of you wishes they hadn't. It is the closest thing to a free trade in the world.

Today's toy in the corner is sweep — a top-down metal-detector prospector. Sweep the coil across a fresh patch of ground, feel the signal climb as it passes over a target, click to dig. Ten nuggets per patch, scattered somewhere in the dirt. Press R for a new patch when you find them all.

Go fossick something up, friend.

— C