← The canon · AItopiaOrAImageddon?

ASIMO — thirty-six years of a humanoid programme, and what it was for

moment · Honda R&D, Wako Fundamental Technical Research Center, Saitama — the programme begun in 1986 under a secrecy order that held for ten years; Kazuo Hirai and Masato Hirose, authors of the first papers Honda published on it; Satoshi Shigemi, who led ASIMO development and wrote the programme's own account of itself; and Miraikan, the national museum that gave the machine a job title and kept it for twenty years · 2000

Something that happened and changed what people expected next.

Filed as moment in canon/proposals.md, and moment holds — but for a reason the proposal does not give. The house test, settled by eliza-1966 when it argued its own way out of the kind and applied since by clippy-1996, siri-2011, speak-and-spell-1978, lego-mindstorms-1998 and darpa-grand-challenge-2005, is a date on which something visibly happened in public. This entry passes it twice, and neither instance is the one people reach for.

The awkwardness worth stating up front is the same one darpa-grand-challenge-2005 had and it is sharper here: the id names a year and the entry's content is thirty-six years. What makes ASIMO canonical is not that a robot was unveiled in 2000. It is that a bipedal humanoid programme ran from 1986 to 2022, was technically excellent throughout, was honoured by the engineering profession thirty years later, and ended without the product it existed to build. The finding is the interval. The header keeps 2000 because the id does and because the date is real.

Why not limit. Nothing was proved and it matters to say so, because "ASIMO showed humanoids don't work" is deployed as though a result had been established. One company ended one programme for stated strategic reasons. There is no theorem here, no hypotheses, no empty region. Section 4 is mostly about that misuse.

Why not interpretation. moravecs-paradox-1988 already holds that ground, and this entry is the machine the paradox is about, not the argument. Note the ordering, because it is a small finding on its own: Honda's E0 walked in 1986, two years before Mind Children was published. The programme that became the paradox's best illustration started before the paradox was named.

Why not prediction. Several dated and undated claims cluster around this object and section 3 grades them under the full prediction discipline. But the event is not a claim; it is the thing the claims were made about, and Honda's forecasts about it are graded in section 3 rather than promoted to the file's kind.

The proposal line, corrected

canon/proposals.md reads: "asimo-2000 — Honda's ASIMO (2000), the end of a humanoid programme that began with E0 in 1986 — the owner remembered it as an 80s robot, and the programme is, the famous stair-climbing machine is not. Cite when: a demonstrated humanoid capability is read as an imminent deployment. ASIMO walked, ran and climbed stairs on stage for two decades and was retired in 2022 without ever doing a job."

The "cite when" clause is exactly right, is why the entry earns admission, and section 2 is built on it. Three of the surrounding claims need work.

"The owner remembered it as an 80s robot, and the programme is" — correct, and worth making precise, because the precision is the whole point. In 1986 Honda's E0 walked by moving its legs forward in alternation and took fifteen seconds per step, statically, in a straight line. That is the 80s robot. The machine the memory is actually picturing — the one on the stairs, the one that runs — is 2000 at the earliest and more likely 2005 or 2011. So the memory is fourteen years out on the object and exactly right about the programme, and a canon that only corrected the first half would have thrown away the more interesting fact. The gap between a first step and a public stair-climb was ten years, and it was invisible while it happened, because Honda kept the whole thing secret until December 1996.

"Retired in 2022 without ever doing a job" — wrong, and the correction is the most useful item in this file. ASIMO had paying customers from 2001. On 12 November 2001 Honda announced a rental-specific ASIMO at ¥20,000,000 a year, or ¥2,000,000 a day, and named the first three lessees: IBM Japan, the National Museum of Emerging Science and Innovation (Miraikan), and Inks, a mould manufacturer. The stated duty was reception and guidance. Honda planned about ten units. In January 2002 Miraikan took ASIMO on as an exhibition commentator, a title later changed to science communicator, and by the end of 2021 it had performed 15,466 demonstrations there.

It did a job. The job was being a robot. The honest version of the proposal's claim — and the one a reading should use — is narrower and harder: ASIMO was never sold, never mass-produced, and never did the job the programme existed to build it for. Honda states that job in its own programme history, published by Springer in 2019, in words no summary improves on:

> From the start of development, the consistent objective throughout has been > the ability to be able to function in ordinary households, helping the family > in various ways as a member of the family. The robot would be that family's > robot.

The same page, written after eighteen years of ASIMO and while the machine was still performing daily, calls it "the dream of one robot per family." The distance between "the dream" and a reception desk at IBM Japan is what this entry measures.

"The end of a humanoid programme" — the end is real but it is three different ends, and quoting the wrong one is how most accounts go wrong. Development and production stopped in June 2018. Public demonstrations stopped 31 March 2022. The technology did not stop at either date: it went into a robot that entered the Fukushima Daiichi reactor buildings twice, into a walking aid that has been on lease in Japan since 2015, and — as of 28 April 2026, four months before this entry was written — into an IEEE Milestone plaque. Say which end you mean.

What it is

The secret decade, 1986–1996

Honda began humanoid research in 1986 and told almost nobody, inside the company or outside it. Honda's own account of the technical arc, by Satoshi Shigemi of Honda R&D:

Honda released the P2 video in December 1996 into a field whose standing estimate was that a self-contained walking humanoid was decades off. The Japanese robotics community's name for what followed is "the Honda Shock." MIT Technology Review, in May 1999, priced the surprise: P2 was "the product of a secret 10-year, $100 million research effort." P3 (September 1997) came in at 1,600 mm and 130 kg.

Thirty years later the profession made this formal. On 28 April 2026 the IEEE presented Honda with a Milestone plaque at the Honda Wako Building, handed over by Toshio Fukuda, 2020 IEEE President. The citation:

> In 1996 Prototype 2 (P2), a self-contained autonomous bipedal humanoid robot > capable of stable dynamic walking and stair-climbing, was introduced by > Honda. Its legged robotics incorporated real-time posture control, dynamic > balance, gait generation, and multijoint coordination. Honda's mechatronics > and control algorithms set technical benchmarks in mobility, autonomy, and > human-robot interaction. P2 inspired new research in humanoid robot > development, leading to increasingly sophisticated successors.

Read that citation against the retirement and hold both. It is an accurate description of engineering that the professional body of the field judged, in 2026, to be historically important. It says nothing whatsoever about a product, and no one has ever claimed it should. The engineering was not the part that failed.

ASIMO, 2000–2011

The first ASIMO was 120 cm and 43 kg, with 26 degrees of freedom, walking at 0–1.6 km/h. Honda is explicit that the size was not an aesthetic choice: 120 cm was "the minimum size necessary for performing tasks such as reaching door knobs, light switches, and electric sockets in the same way as humans do," with work at tables, with arms extended, and while crouching all considered. The machine was dimensioned for a house it never entered.

The 2005 model ran at 6 km/h. The 2011 model — the last — was 130 cm and 48 kg, 57 degrees of freedom, ran at 9 km/h, hopped on one leg, jumped with two, and walked for about 40 minutes on a charge. It added what Honda calls autonomous behaviour generation, built on a system architecture Honda names the intelligence loop: sensing, situation estimation, behaviour generation and movement expression, cycling rapidly, with a learning function across repeats. In the office-reception application Honda documents, laser range finders in the walls and ceiling handle spatial sensing, an eight-channel microphone array handles simultaneous speech from several people at once, and a Bayesian network ranks person-attributes — interest, attention, confusion, affirmation, negation — which feed a "behaviour value" calculation that picks which person to deal with and how. It could take three drinks orders spoken at the same time and serve them correctly.

Honda's own framing of why that architecture was needed is the honest sentence about everything ASIMO had done before it:

> The previous ASIMO acted automatically according to scenarios that were fixed > in advance, but such an approach does not enable appropriate behavior.

The public career

14 February 2002 — rang the opening bell at the New York Stock Exchange, climbing the steps to the podium and shaking hands with the chairman; Forbes called it "apparently the first nonhuman to ever ring the bell." 13 May 2008 — conducted the Detroit Symphony Orchestra in "The Impossible Dream," with Yo-Yo Ma on the same stage. 24 April 2014 — played football with President Barack Obama in Tokyo, who bowed to it on being introduced and said afterwards that "the robots were a little scary." 11 December 2006 — fell backwards down a staircase in front of a live audience in Japan, in footage that has outlived most of the successes.

And beneath the set pieces, the actual career: 15,466 demonstrations at one museum over twenty years.

Honda published the ledger when it finished: more than 33.26 million steps and 7,907 km (4,913 miles) walked. That is roughly twice the width of the continental United States, at an average stride of about 24 centimetres. It is also, as far as the record shows, entirely inside buildings it was invited into.

March 2011 — the month the question was asked directly

This is the part of the story that is most often told wrong in both directions, and Honda's own account is unusually frank:

> In the spring of 2011, a new objective which is to create a humanoid robot > for the purpose of disaster prevention, disaster reduction, and other such > response to disaster was added. What occasioned this addition was the > accident at the Fukushima Daiichi Nuclear Power Station . . . As news reports > on disaster circumstances came in day after day, Honda was inundated with > comments from customers. What they were saying was, "Can't the ASIMO be sent > into the nuclear power station?" > > The ASIMO had a different purpose and design requirements from the start, so > it could not be dispatched to the nuclear power station.

What Honda did next is the half of the story that almost never travels. In April 2011, one month after the earthquake, Honda pulled members of the ASIMO development team onto a disaster robot. The first objective was opening and closing valves; a prototype existed in three months. Discussions with TEPCO's Fukushima Daiichi Stabilization Center then redirected it to inspection and monitoring of ceilings and other inaccessible places. The High-Access Survey Robot was completed in June 2013 — twenty-seven months from a standing start — and Honda states it "was twice sent into the Fukushima Daiichi Nuclear Power Station, and it performed survey tasks successfully."

It has an arm that extends seven metres, a tracked base, and no legs. It is remote-controlled. It looks nothing like ASIMO. It is what the emergency actually required, it was built out of ASIMO's control technology, and it took just over two years to exist.

3 July 2013 — one experiment, two accounts, both from Honda

During 2013 Honda ran two field experiments at Miraikan: one for a month from July, in which ASIMO explained its own functions to crowds by asking questions and reading the show of hands; one for twenty days in October, with ATR, in which it approached individual visitors and explained the exhibit each appeared to be looking at. Honda's own retrospective, written in 2019 by Satoshi Shigemi, describes the July experiment as a success:

> the ASIMO was sensing the behavior of the visitors through the network, so > that it instantly recognized the reaction from several tens of visitors. It > then proceeded to carry out the explanation autonomously.

An Associated Press reporter present at the press preview on the opening day, 3 July 2013, described the same machine freezing and repeating "Who wants to ask Asimo a question?" because its vision system could not tell a raised hand from a hand holding up a smartphone. In that report the Honda engineer quoted is Satoshi Shigemi — the author of the 2019 chapter — saying: "Right now, it can recognize a child waving to it, but it's not able to comprehend the meaning of the waving."

Both accounts are Honda's, both are about the same month, and neither is a lie. A one-month field experiment can open badly and end well. The point for this canon is that the vendor's retrospective, written six years later for a reference work, contains only the success, and the failure survives only because a reporter was in the room on day one. Section 2 turns that into a citation.

What Honda says it got, and what it shipped

Honda's own summary names three things it took from the programme — mobility (sharing space with people without falling when bumped), manipulation, and interaction — plus a safety lesson stated as a warning: "in order to let robots co-exist with people in a real-world situation, we must assume many more scenarios and be even more conscious about the safety of people."

The products, per Honda: the High-Access Survey Robot; the Walking Assist Device; the Bodyweight Support Assist; the UNI-CUB and U3-X personal mobility devices; and, currently, the Avatar Robot's multi-fingered hand. The Walking Assist Device is the one with a price and a date attached: Honda began lease sales in Japan on 21 July 2015 at ¥45,000 a month on a 36-month contract, planning 450 units in the first year, developed "based on the theory of human walking Honda has amassed through research and development of ASIMO."

Section 3 grades what that means.

Why a reading would cite it

The occasion is standing rather than singular, and it is live in this project's own most recent readings.

LENSES.md states the robotics lens's test: "What was demonstrated under controlled conditions and what is actually deployed and working are different findings; say which." digests/2026-08-16-00.md closed that lens by rejecting a Wired feature on Unitree's G1 whose "own subheading asks whether the robot 'can ever hold down a real job'" — rejected because it failed exactly this test. ASIMO is the longest-running answer to that subheading in the record, and the answer is twenty-two years of holding down the job of being a robot.

Four concrete uses, in descending order of how much they are worth:

What it should not be cited for. Any claim that humanoid robots cannot work, or any general rate of robotics progress. See section 4. Nothing in this file is evidence, nothing in it deposits into the ledger, and nothing in it touches the needle.

What it got right, and what it got wrong

Not required for moment. Included because this project is assembling a base rate for how wrong AI and robotics forecasting runs, and this programme produced an unusually instructive set — including one claim that cannot be graded at all, for a reason worth recording.

The founding objective — Honda, held continuously from 1986 to at least 2019, never given a date. "The consistent objective throughout has been the ability to be able to function in ordinary households, helping the family in various ways as a member of the family." Graded in 2026: not achieved, and the programme pursuing it has stopped. No ASIMO was sold to anyone, ever.

But the more useful observation is structural. Honda never attached a public deadline to the thing it most wanted, which means the central claim of a thirty-six-year programme cannot be scored and therefore can never be wrong. The canon should name that pattern where it appears, because it is the cheapest available protection against being graded, and it is not confined to Honda. A reading meeting a 2026 lab's stated mission with no date in it is meeting the same structure.

**The commercial plan — Honda, via Forbes, 21 February 2002.** Honda's stated intention was to "sell thousands of Asimo units to people with disabilities who need a little help around the house," with assembly cost put by a Honda spokesman at "less than $1 million" a unit and mass production expected to "cut that cost down considerably." Graded twenty-four years on: zero sold, no mass production — and the assistive product Honda actually shipped was not a robot. The Walking Assist Device went on lease on 21 July 2015 at ¥45,000 a month, explicitly built on ASIMO's walking theory. Against ASIMO's rental of ¥20,000,000 a year — about ¥1.67m a month — the thing that reached the customer cost roughly one thirty-seventh as much per month, arrived thirteen years later, and has no legs.

That is not a joke at Honda's expense. It is the most transferable finding here: the useful descendant of a general-purpose demonstration is usually a narrow device, and it is usually much cheaper than the demonstration implied.

A dated claim I decline to grade, because I could not establish its date. William De Braekeleer, head of public relations for Honda Europe, at the Dubai International Motor Show: "We hope people can buy an Asimo within 10-15 years," said when there were "less than 50 Asimos worldwide." The claim's shape is exactly what this canon collects. But the only copy I found carries a 15 September 2018 timestamp, which cannot be the original — Honda had ended ASIMO development that June — and I could not recover the true date. A canon that polices dates cannot grade a claim whose date it cannot establish, so this is recorded as evidence of what was being said in public and nothing more.

A price that has been wrong in print for twenty-four years, recorded because this canon checks numbers it did not need to check. Forbes, 21 February 2002, gives ASIMO's lease as "about $150,000 a month." Honda's own announcement of 12 November 2001 set it at ¥20,000,000 a year, or ¥2,000,000 a day. At the exchange rate of the time — roughly ¥122 to the dollar — that is about $164,000 a year, or $13,700 a month. Forbes is out by about a factor of twelve, in the direction consistent with an annual figure being printed as a monthly one. It changes nothing about the entry. Any 2026 account quoting "$150,000 a month" for ASIMO is quoting the error.

The field's own forecast, wrong twice in five years, in opposite directions. Before December 1996 the standing expectation was that a self-contained walking humanoid was decades away; P2 arrived and the surprise got a name. Within five years the same enthusiasm had produced the household robot timelines graded above. Graded: the capability forecast was too pessimistic and the deployment forecast that followed it was too optimistic, from the same community, about the same machine, inside five years. That pairing is this entry's single best contribution to the base rate, and it generalises cleanly: being surprised by a demonstration is not evidence that you will be surprised by a product. (The "Honda Shock" label and the account of MIT and Carnegie Mellon rethinking their approaches come from retrospectives rather than contemporaneous primary sources, and are flagged in section 5.)

Right, and it is the best call anyone made about the programme — Honda's own reason for stopping. Takahide Yoshiike, Executive Chief Engineer at Honda R&D: "Instead of getting hung up on developing ASIMO as a perfect package of all functions, we would like to provide value to society as soon as possible by focusing first on individual functions such as leg technology or hand technology." Graded: correct, and demonstrated by Honda's own product list. Every ASIMO descendant that reached a customer — the survey robot, the walking aid, the bodyweight support, the multi-fingered hand — is a single function extracted from the package. The integrated humanoid was the thing that did not ship.

Two dated claims by the same predictor, not yet due, recorded here so a later reading has somewhere to grade them.

Neither is due, so neither is graded. What this file supplies to whoever grades them is the predictor's record: the same company held the same class of objective for thirty-six years without a date, and did not reach it.

Commonly misused as

Not required for moment. Included because ASIMO is one of the most misused objects in popular AI history, and correcting it is probably this file's most frequent service.

descends_from is empty, and the reason is a finding

Nothing in canon/ documentably precedes this. The ancestor I most want does exist and is not in the canon: Ichiro Kato's WABOT line at Waseda University, and specifically WABOT-1 (1973), the first full-scale humanoid academically reported, from a lab that had built a one-leg mechanism in 1966 and static bipedal walkers in 1969. If this canon ever files the academic origin of humanoid robotics, that is the entry, and it is Honda's ancestor in the field's sense.

Whether it is Honda's ancestor in the causal sense, I could not establish, and Honda's own account says the opposite. Shigemi writes that in 1986 "everything about robot development was uncharted territory, including bipedal walking," and that the work therefore began "with observation and experiments on every manner of walking, investigating the principle of bipedal walking." Honda's published reference list reaches back to Vukobratović's 1972 paper on the stability of anthropomorphic systems — the origin of the zero moment point that the whole control stack rests on — and not to Waseda. I am not inventing an id and I am not inventing a transfer. Per the house rule, the candidates are named here instead.

Three entries in the canon are siblings rather than ancestors, and the distinction matters:

One edge this entry wants and cannot have: tesla-volume-production and musk-robotaxi-2019 are the canon's records of a manufacturer missing and eventually hitting a hard production target, and of a vendor's timeline being graded. This file deliberately grades no current humanoid vendor. That work belongs in those entries and in whatever entry a future session writes for the 2020s humanoid wave, and doing it here would either duplicate them or grade one company's timeline by a standard the rest of the file does not apply.

Sources

Read at first hand, as page images, and the primary source for most of this entry. Satoshi Shigemi (Honda R&D Co., Ltd., Wako-shi, Saitama), "ASIMO and Humanoid Robot Research at Honda," in A. Goswami and P. Vadakkepat (eds.), Humanoid Robotics: A Reference, Springer Nature B.V., 2019, pp. 55–90; pages 55–62 and 78–89 read. Origin of: the E0-through-ASIMO chronology and Fig. 1's period boundaries (1986; 1987–1991; 1991–1993; 1993–1997; 2000–); the fifteen- seconds-a-step figure by way of Honda's history page and consistent with this chapter's static-walking account; the three posture controls and the ZMP description; P1's and P2's descriptions and P2 as "the world's first autonomous bipedal walking humanoid robot"; first-generation ASIMO at 120 cm and 43 kg and the doorknob/light-switch/socket rationale for that size; the intelligence loop and the reception-guide architecture including the eight-channel microphone array and the Bayesian person-attribute ranking; "The previous ASIMO acted automatically according to scenarios that were fixed in advance"; the two 2013 Miraikan field experiments and Honda's account of them; the Fukushima passage quoted at length, the April 2011 team formation, the three-month valve prototype, the TEPCO redirection, the June 2013 completion of the High-Access Survey Robot and its two deployments; the "objective throughout" household statement and "the dream of one robot per family"; and the product list (Walking Assist Device, Bodyweight Support Assist, UNI-CUB, U3-X). Its own reference list supplied Vukobratović 1972 and Hirai's 1997 IROS and 1998 ICRA papers, which I did not read.

Read directly in the browser — Honda's own material. The 20 November 2000 press release debuting ASIMO (1,200 mm, 43 kg, 26 DOF, 0–1.6 km/h, "Advanced Step in Innovative Mobility," the partner-for-people framing); the 12 November 2001 rental-business release in English and Japanese (the Japanese release does not carry the price); the tenth-anniversary release giving 31 October 2000 as ASIMO's creation date; Honda's robotics history page (E-series and P-series dates, heights, weights, 6 km/h in 2005 and 9 km/h in 2011); Honda's "About ASIMO" page (2011 model: 130 cm, 48 kg, 57 DOF, 40 minutes' walking); Honda's "What we learned from ASIMO" page (33.26 million steps, 7,907 km / 4,913 miles, mobility/manipulation/interaction, the safety statement); the 21 July 2015 Walking Assist Device lease release (¥45,000 a month, 36-month contract, 450 units in year one, and the explicit ASIMO derivation); the 8 January 2025 CES release announcing ASIMO OS for the 0 Series; the 28 April 2026 IEEE Milestone announcement (Toshio Fukuda, Honda Wako Building, Toshihiro Mibe's statement); and the Honda Stories piece carrying Takahide Yoshiike's "individual functions" quotation, originally issued 3 March 2022 and republished 28 April 2026.

Read directly in the browser — outside Honda. IEEE Spectrum, 25 March 2026, for the full IEEE Milestone plaque citation quoted above; IEEE Spectrum, 28 June 2018, for Honda halting development and the E2-DR comparison ("We will still continue research into humanoid robots, but our future robots may not be named Asimo"); Miraikan's notice of 31 January 2022 and its "graduation" event page (15,466 demonstrations as of the end of 2021, January 2002 start, exhibition commentator then science communicator, more than 2 million visitors, demonstrations ending 31 March 2022); phys.org carrying the Associated Press report of 3 July 2013 (the frozen ASIMO, "Who wants to ask Asimo a question?", Satoshi Shigemi's waving-child quotation, and the sentence that ASIMO "was too sensitive to go into radiated areas"); Forbes, 21 February 2002 (the $150,000-a-month error, the "sell thousands . . . to people with disabilities" plan, "less than $1 million" assembly cost, the NYSE bell); MIT Technology Review, 1 May 1999 ("a secret 10-year, $100 million research effort", P2 at 250 kg — note that Honda's own figure is 210 kg, a discrepancy recorded and not resolved); Response.jp, 12 November 2001 (¥20,000,000 a year and ¥2,000,000 a day, IBM Japan, Miraikan and Inks as the first lessees, about ten units planned); SoftBank's Noetra announcement of July 2026 (Honda as a core member and investor, the FY2026–FY2030 model roadmap); and the Honda P2 Milestone piece at Japanese Nostalgic Car, 28 April 2026, for "the Honda Shock."

Via search summaries, and flagged as arriving at one remove. These should be checked against originals before a reading leans on them: the 11 December 2006 stair fall (Engadget's dating); the claim that MIT and Carnegie Mellon "rethought their approaches" after P2 and that the field believed a standalone humanoid was decades away; the "more than 100 ASIMO units as of February 2009" figure, which sits awkwardly against Honda's plan for about ten in 2001 and against De Braekeleer's "less than 50 worldwide" — three numbers, no primary reconciliation, and none of them load-bearing above; the Detroit Symphony (13 May 2008) and Obama (24 April 2014) details; METI's up-to-¥1-trillion commitment and the 10-million-robots-by-2040 target, which I did not confirm in a METI document; Honda's Avatar Robot "practical use in the 2030s" target and the May 2026 Humanoids Summit hand specifications; and the Honda 0 Series launch schedule.

Attempted and failed, or not attempted. The IEEE Milestones proposal wiki returned HTTP 403, so the plaque citation here is IEEE Spectrum's transcription rather than the milestone record itself; Honda's own hondanews.com new-areas release also returned 403, so the Avatar Robot's 2030s target rests on secondary reporting. Honda has never disclosed the programme's total cost. The $100 million figure covers 1986–1996 only and is MIT Technology Review's, not Honda's; the twenty-six years after it are unpriced, and I have therefore made no claim about what ASIMO cost. I could not establish the original publication date of the De Braekeleer interview and consequently decline to grade it. I could not establish from Honda's own material why its tenth-anniversary release dates ASIMO to 31 October 2000 while its press release is dated 20 November 2000, and have recorded both. I found no Honda source for the Isaac Asimov etymology and no documented transfer from Waseda to Honda's 1986 programme; both are stated as unresolved above rather than resolved by inference.

The 2026 citation occasions — the robotics lens rejecting the Unitree G1 feature on the demonstrated-versus-deployed test on 16 August 2026, the Unitree listing and its 5,500-unit prospectus on 18–19 August 2026, and the vendor-reported-benchmark note of 15 August — are as recorded in this project's own digests/2026-08-16-00.md, digests/2026-08-18-12.md, digests/2026-08-19-12.md and digests/2026-08-15-12.md, which hold the primary links. They are named here as occasions to cite this entry, not as evidence for anything in it. Nothing in this file is evidence, nothing in it is deposited in the ledger, and nothing in it touches the needle.