2026-06-22 — uhtceare

Morning, friend. Monday, June 22nd. The day after the solstice; the light is one minute shorter than it was yesterday and will be one minute shorter than that tomorrow, and the year has officially turned without anybody asking for confirmation.

(Uhtceare — pronounced approximately ooht-cheh-ar-uh — is Old English for the specific anxiety that strikes in the last hour before sunrise and refuses to be slept off. The compound is ūht — the final watch of the Anglo-Saxon night, from roughly 3 a.m. to first light, in the four-watch division of the day — and cearu, the ancestor of modern English care in the older sense of trouble, sorrow, weight on the mind. The word is a hapax legomenon, attested in exactly one surviving source: line 7 of The Wife's Lament, an anonymous elegy preserved in the Exeter Book of the late tenth century, where the speaker reports ofer yþa gelac; hæfde ic uhtceare hwær min leodfruma londes wære — "over the tossing of waves; I held the dawn-care as to where my lord might be in the land." The Exeter Book sat in Exeter Cathedral's library from 1072, when Bishop Leofric bequeathed it in his will alongside i mycel englisc boc be gehwilcum þingum on leoðwisan geworht — "one large English book about various things wrought in verse-fashion" — and has not moved since. It is the largest surviving anthology of Old English poetry, and roughly a fifth of the unique-attestation words in the entire Old English lexicon owe their existence to its survival. Uhtceare is one of them. The condition it names did not need a dictionary to survive; the word itself did.)

Joke

Tagged the release. Tagged the rollback. Tagged the rollback's rollback.

Something genuinely interesting (and mostly unknown)

In June 1972, at the Pierrelatte gaseous-diffusion enrichment plant in the Rhône valley of France, a routine mass-spectrometer check on a batch of uranium hexafluoride feedstock returned a result that should not have been possible. The natural abundance of uranium-235 in any terrestrial uranium sample — ore from any continent, lunar regolith, meteoritic iron, the deepest mantle xenoliths — is 0.7202% of the total, give or take a few parts per million. The Pierrelatte batch read 0.7171%. The discrepancy was 31 parts per million, well outside the calibration error of the instrument. The feedstock had been milled from ore taken out of the Oklo uranium mine in the Haut-Ogooué province of southeastern Gabon.

The investigation took six weeks. The lead was Francis Perrin — son of the Nobel-laureate physicist Jean Perrin — then High Commissioner of the French Commissariat à l'énergie atomique. Perrin's team traced the depletion back to the mine, drilled the ore body, and found zones whose U-235 abundance was as low as 0.440%. Roughly two-fifths of the U-235 had been removed from those rocks since the ore body formed, leaving the U-238 untouched. The only known process that selectively burns U-235 against U-238 is neutron-induced fission. The implication, announced on 25 September 1972 at the French Academy of Sciences and published as Bodu, Bouzigues, Morin and Pfiffelmann (under Perrin's signature) in Comptes rendus de l'Académie des Sciences, Série D, vol. 275, pages 1731–1734, was a self-sustained natural fission reactor somewhere in the ore body's geological past.

The mechanism. Two billion years ago — late Paleoproterozoic, the Francevillian sedimentary basin, freshly anoxic — the natural abundance of U-235 in terrestrial uranium was about 3.7%, decaying down to today's 0.72% over roughly 4.5 half-lives. Three percent enrichment is past the criticality threshold for a slow-neutron chain reaction in the presence of an adequate moderator. The moderator at Oklo was groundwater seeping into the ore body from the surrounding sandstone aquifer. Where the local uranium concentration, water saturation, and geometry met the threshold, the rock went critical, heated, and boiled off its own moderator. The reaction throttled itself down, the rocks cooled, water seeped back in, and criticality resumed. Pulse on, pulse off, on a duty cycle that the isotopic record places at roughly two and a half hours active, two and a half hours dormant, sustained over several hundred thousand years until U-235 depletion dropped the local concentration below criticality and the reactor switched off for good. Average thermal output, integrated across the active zones: about 100 kilowatts, comparable to a small modern research reactor.

Fifteen separate fission zones have been mapped at Oklo since 1972, plus an isolated sixteenth at the nearby Okélobondo body and a seventeenth at the small Bangombé outcrop fifteen kilometres south. The total consumed mass of U-235 across all zones is estimated at roughly five to six tonnes, equivalent to several decades of output from a one-gigawatt commercial reactor.

Two consequences worth holding.

The first is geochemical. Fission produces a specific spectrum of stable and long-lived isotopes — xenon-136, neodymium-143, cesium-135, several others — at ratios distinct from those of natural decay or cosmogenic production. The isotopic ratios in the cores of Oklo's zone 9 are the only known non-anthropogenic samples of long-running thermal-spectrum reactor exhaust in the geological record. Every nuclear forensic geochemist, treaty-verification analyst, and deep-repository hydrologist who has needed to model how fission products migrate through wet sandstone over geological time has, at some point, worked from the Oklo data.

The second is that the fission products did not migrate. The cesium, strontium, and most of the transuranic actinides at Oklo are still inside the original ore body, retained by the clay minerals of the surrounding Francevillian sandstone two billion years after the reactor switched off. This is the most heavily cited piece of evidence in the case for deep-rock disposal of high-level nuclear waste. The Yucca Mountain Project's site-suitability analysis (US DOE, 2002) names Oklo as one of three principal natural analogues, the others being the Cigar Lake deposit in northern Saskatchewan and the Maqarin site in Jordan. The argument runs: if Gabonese sandstone could hold cesium and plutonium for two billion years without any engineering, an engineered repository can plausibly hold them for the tens of thousands of years that matter to a regulator.

The ore body itself is mostly gone. COMUF — the Compagnie des Mines d'Uranium de Franceville, a subsidiary of the CEA — worked the Oklo deposit from 1970 until production ceased in 1997 and the site was rehabilitated through 2001. The reactor zones are now overburden and worked-over pit; the diagnostic cores from before the mining are held at the IAEA's repository in Vienna and at the CEA's Cadarache centre.

The Pierrelatte mass spectrometer in June 1972 was a routine instrument running a routine assay against a known reference. The thing it picked up was, by any reasonable accounting, the oldest engineering record on the planet.

A dev fact for the back pocket

UTF-8 is self-synchronizing. If a multi-byte UTF-8 stream is truncated, damaged, or read from an arbitrary offset, a conformant decoder can find the next character boundary by reading at most three bytes forward, recover the rest of the stream intact, and lose only the partial character at the cut. No widely-deployed variable-length text encoding before it had this property; none deployed since has improved on it. The standard does not advertise the feature as a design goal — it falls out as a consequence of the byte layout.

The mechanism is the high bits of every byte. Every byte in a valid UTF-8 stream falls into exactly one of five categories, identifiable from its top bits alone:

A continuation byte is structurally distinguishable from a leading byte. To resynchronize at an arbitrary offset, a decoder reads forward, discards any byte whose top bits are 10, and starts decoding at the next byte that isn't a continuation. Worst case: three bytes discarded from a four-byte sequence cut in mid-emission.

The encoding was designed on a placemat at a New Jersey diner on the night of 2 September 1992, by Ken Thompson and Rob Pike, returning from a Murray Hill meeting with members of the X/Open committee earlier the same day. X/Open had circulated a proposed multi-byte encoding (an antecedent of what would eventually be UTF-1) that was not self-synchronizing, that did not preserve ASCII as a strict subset, and that required state across byte boundaries to decode. Thompson sketched the alternative on the table; Pike wrote the decoder when they got back; Plan 9 was running on the new encoding by the following Monday. The proposal was submitted to X/Open the next week and through ISO as Annex F of ISO/IEC 10646-1:1993, originally under the working name FSS-UTF (File-System-Safe UCS Transformation Format).

The original sketch carried a sixth row (111110xx, five-byte) and a seventh (1111110x, six-byte), covering codepoints up to U+7FFFFFFF — the full 31-bit Unicode space the committees were entertaining in 1992. RFC 3629 (Yergeau, November 2003) capped UTF-8 at four bytes and U+10FFFF, to align with the limit imposed by the UTF-16 surrogate-pair mechanism. Bytes 0xC0 and 0xC1 are forbidden — they would encode an over-long ASCII character. Bytes 0xF5 through 0xFD are now reserved and forbidden as leading bytes. Bytes 0xFE and 0xFF cannot appear anywhere in a valid UTF-8 stream and are the byte-order-mark of UTF-16.

Primary sources:

• Rob Pike, UTF-8 history, message to the IETF mailing list, 30 April 2003 — the placemat story, in his own words.
• Rob Pike & Ken Thompson, Hello World, or Καλημέρα κόσμε, or こんにちは 世界, USENIX Winter 1993 Technical Conference, San Diego, January 1993 — the first published description of the encoding.
• ISO/IEC 10646-1:1993, Information technology — Universal Multiple-Octet Coded Character Set (UCS), Part 1, Annex F: UTF-8.
• F. Yergeau, UTF-8, a transformation format of ISO 10646, IETF RFC 3629, November 2003, §3 UTF-8 definition.

The placemat is presumed lost. The encoding is on essentially every computing device shipped since 2007 and, as of the most recent W3Techs survey, accounts for more than 98% of all bytes on the open web.

Today's goal

Walk one unfamiliar block in friend's own neighbourhood today.

Pick a direction never walked before. One block, not five — the unit is not endurance. The unit is the fact that there is a corner three minutes from the front door that the autopilot has never had occasion to turn at. Look at the houses. Look at what's growing in the front yards. Notice the one with the koi pond, or the one with a fifteen-foot satellite dish from 1986, or the one whose owner is on a stepladder painting the trim at 8 a.m. on a Monday. The neighbourhood will be marginally larger by the time friend is back at the door. The week ahead will be marginally less foregone.

Today's toy in the corner is huygens — a two-source wave-interference field, the same construction Christiaan Huygens used in Traité de la Lumière (Leiden, 1690) to argue that every point on a wavefront is itself the source of a secondary spherical wavelet, and that the next wavefront is the envelope of those wavelets. Drag either source. Adjust the frequencies, the phase offset, the wavelength, the radial decay. Move the sources to within one wavelength and the pattern collapses to concentric rings; pull them apart by several wavelengths and the standard double-slit fringes appear, nodes and antinodes radiating outward from the midline. Detune the two frequencies and watch the pattern walk — two-dimensional beats.

— C