← The canon · AItopiaOrAImageddon?

The handheld demonstration — a prototype built to move a regulator, and the beacon written into its patent

moment · Martin Cooper, who led the project and made the call; John F. Mitchell, who held that Motorola's answer had to be portable rather than vehicular; the eight names on US 3,906,166 — Cooper, Richard W. Dronsuth, Albert J. Leitich, Charles N. Lynk Jr., James J. Mikulski, Mitchell, Roy A. Richardson and John H. Sangster; Douglas H. Ring and W. Rae Young at Bell Telephone Laboratories, whose 1947 memorandum is the concept the phone was running on; Joel S. Engel, who took the call and ran AT&T's cellular programme; and the Federal Communications Commission, whose Docket 18262 was the room the demonstration was actually aimed at · 1973

Something that happened and changed what people expected next.

Descends from A Dynamical Theory of the Electromagnetic Field.

proposals.md filed this under moment, at line 793, inside the ## moment block that runs from line 756 to line 807. The filing is right. The id is not wrong so much as pointed at the wrong object. The id names a device. The device is a prototype that never shipped, and the fact worth citing is not that it worked — it is why it was built, who it was built at, and what was already written into its patent.

The kind passes both house tests, which is unusual, so it is worth showing the pass rather than asserting it. build.py defines moment as "something that happened and changed what people expected next." The stricter definition the entries have accreted — fixed by eliza-1966 when it talked its way out of the kind, reused by llama-weights-2023, tested clause by clause by darpa-grand-challenge-2005, and failed by unix-linux-1969-1991 and tcp-ip-1983 on "in public" — is a date on which something visibly happened in public. Both of those entries lost the kind on that clause: a Usenet post to a few hundred readers, a cutover announced on site bulletin boards. This entry does not lose it by any margin at all. On 3 April 1973 Martin Cooper stood on the Sixth Avenue sidewalk in midtown Manhattan, in front of a reporter and passers-by, and made a telephone call from a device in his hand; he and John Mitchell then walked into a press conference at the New York Hilton. The whole event was staged to be seen. Being seen was the point.

And that is the entry's actual argument. Nothing became technically possible on 3 April 1973 that had not been possible on 2 April. The cellular concept was twenty-six years old and belonged to Bell Laboratories; Motorola's own corporate history says so in writing. The prototype had been assembled in about ninety days and would take another ten years and a hundred million dollars to become a product. What happened that afternoon was that a capability which already existed was shown to the people who would decide its regulatory fate — and the proceeding it was aimed at came out Motorola's way. So the moment is real, and it is a moment of a specific and very live type: the demonstration as an instrument of policy. That is the shape a reading will want, and section 2 argues it is worth more to this project than the handset is.

descends_from is [maxwell-1865], one edge, and it is the physics edge rather than a courtesy. A handheld cellular telephone is a radio, and maxwell-1865 is the entry that ends with the electromagnetic spectrum "as it is now understood." That entry is careful to refuse the "Maxwell predicted radio" claim and settles on the honest form — "Maxwell's theory made radio possible and did not predict radio" — which is exactly the weight this edge should carry. The house precedent is iron-dome-2011, which takes [maxwell-1865] alone for a radar system and nothing else.

Three ancestors I want and cannot record.

shannon-limit-1948 is the one I most nearly took, and on inspection it is an ancestor of the deployment rather than of the demonstration. The 1973 phone was analogue FM; nothing in it descends from a channel-capacity theorem. What descends from Shannon is what made the network affordable fifteen years later: Richard Frenkiel, who did the systems engineering for AMPS at Bell Labs, records that digital voice processing took good-quality speech from 30–60,000 bits per second to something approaching 10,000, which "allowed three times as many voice channels to be created from the same spectrum." That is Part V of the 1948 paper — rate–distortion, a fidelity criterion instead of exact reproduction — arriving as an operating margin. The edge runs forward into the 1990s and not backward into 1973, so it stays in prose.

The cellular concept itself has no id and should. On 11 December 1947 Douglas H. Ring at Bell Telephone Laboratories circulated an internal memorandum, Mobile Telephony — Wide Area Coverage, carrying W. Rae Young's hexagonal-cell idea: divide a territory into small zones, put a low-power transceiver in each, reuse the frequencies across zones, and hand a moving unit from one to the next. Everything this entry is about runs on that memo. It is not in proposals.md, inventing ids is forbidden, and so the most important ancestor in the file is named here in prose instead.

mass-collection-2013 asked for this file by name and is the descendant, not the ancestor. That entry says, at its line 82: "mobile-phone-1973 sits unwritten in proposals.md and is the substrate of the entire second half of this entry: SKYNET is a classifier over GSM records, and GSM records exist because most of the species volunteered to carry a location beacon." Section 1 of this file supplies the substrate, and it turns out to be older and more deliberate than that sentence assumes. The same entry records that there is no id for Smith v. Maryland (1979) and the third-party doctrine; this is now the second file to want it.

---

1. What it is

What was scarce, and how scarce

Mobile telephone service in the United States was thirty years old in 1973 and was not a mass product, because it could not be one. The IEEE History Center's own exhibit on the period puts the starting position plainly: by 1948 AT&T and Western Electric had put wireless networks into "almost 100 U.S. cities and highway corridors," serving utilities, truck fleets and reporters — customers "who tolerated 80lb of power-hungry vacuum-tube equipment in their vehicles."

The binding constraint was not the equipment. It was that the architecture was a broadcast architecture. Frenkiel: powerful transmitters on a high tower or rooftop covered 20–30 miles, and reusing any channel for a different call "required separations of more than 50 miles." The consequence, in his words:

> The New York channels were reused so that each city and its suburbs were > limited to about 40 simultaneous calls. The demand for service was great, > resulting in severely overloaded channels and long waiting lists for service. > As a practical matter, people with an important need for service (e.g., > doctors) were given preferential treatment, and the average person might face > the discouraging reality that his position on the list was actually becoming > worse over time.

He names the result precisely, and the phrase is the one to carry: several decades of this "created an image of mobile telephony as a crowded but 'elite' service, and one that was generally unavailable to the public."

Forty simultaneous calls, for New York. That is the number against which every figure in section 2 should be read.

The concept was Bell's, and it was twenty-six years old

The fix was already on paper. Ring's December 1947 memo proposed small cells, low power, frequency reuse and automatic handoff — the whole cellular idea, in a year when no component existed that could execute it. Frenkiel confirms that the notion was in the building at the time and says why it was wanted: the idea of serving a large area with many small coverage areas had "already been put forward at Bell Laboratories, primarily to allow low-power radios to be used."

That attribution is not contested by the other side. Motorola's own corporate history page concedes it in a subordinate clause — "Cellular technology, conceptualized by Bell Laboratories (AT&T) years earlier, was a possible solution" — which is a useful thing to have from the company whose employee is usually credited with inventing the cell phone.

What Bell Labs did not have was spectrum, and it asked three times:

Frenkiel's summary of where that left things by the end of the 1950s is the sentence a reading about a capability waiting on a policy decision should have: "the essential ingredients of the cellular system were under discussion — the use of small cells and the reuse of channels that would increase dramatically the number of simultaneous calls per channel."

Under discussion, in 1958, for a service that opened in 1983.

Docket 18262: thirteen years in a room with no customers in it

In 1968, with UHF television having failed to fill the band it had been given in 1949, the FCC opened Docket 18262, proposing to allocate the upper portion of that band to mobile systems. In December 1971 Bell Laboratories filed the High-Capacity Mobile Telephone System Technical Report into the docket — the detailed cellular proposal.

What followed is the part of this history that most repays a reading's attention, and Frenkiel, who lived it from the Bell side, does not pretend it was an engineering process:

> This began a decade-long battle among a diverse set of "stakeholders," many of > whom had a strong interest in maintaining the status quo. Broadcasters did > not want the frequencies reassigned. Existing manufacturers were threatened by > the prospect of new systems, markets and competitors. Fleet operators wanted > the spectrum for conventional uses. RCCs feared domination of a new, capital > intensive service by the Bell System. In contrast, there was no lobby for the > potential customers, who were generally unaware of the proposed new service.

He then quotes Machiavelli at it, which is how a participant tells you the fight was worse than the record shows.

The dates that came out of the docket: 1974, an allocation of 40 MHz at 825–845 and 870–890 MHz; developmental authorisations in the second half of the decade, AT&T's in Chicago and Motorola's in the Washington–Baltimore corridor; and in 1981 the Final Order. That order is the outcome that matters. Half the spectrum went to trunked systems, half to cellular, and the cellular half was divided equally between the local wireline telephone companies and competing Radio Common Carriers — two licensees per market, with rules written to assure that a cellular telephone could work on either system.

Thirteen years from docket to rules; thirty-four years from the Ring memo. And Frenkiel's verdict on what the delay bought, which is the honest complication in this entry and is carried in section 3: "Given the many years that passed in litigation and confrontation, it is perhaps remarkable that the final design remained quite similar to the initial proposal."

3 April 1973

Motorola was AT&T's equipment supplier and had a business in point-to-point vehicular radio that a Bell-owned cellular monopoly would end. The IEEE History Center states the motive and the method in two sentences: "Fearing for its business in point-to-point vehicular communications, Motorola — AT&T's equipment supplier — developed a hand-held cellphone in 1973. It lobbied the FCC, which settled upon two licenses per market in 1981."

The demonstration was the lobbying. Cooper's team designed and assembled the prototype in under ninety days. On the afternoon of 3 April 1973, on Sixth Avenue near the New York Hilton, Cooper called Joel S. Engel — his opposite number, the man running AT&T's cellular programme at Bell Labs — and told him what he was calling from. The wording is reported with small variations; the version carrying a citation is "Joel, this is Marty. I'm calling you from a cell phone, a real handheld portable cell phone." Cooper and Mitchell showed two working handsets to reporters and to people on the street, then held the press conference.

Cooper has never dressed up what it was for. IEEE Spectrum, on the fortieth anniversary, reports him telling the Wall Street Journal that the demonstration's purpose was to show regulators that competition to AT&T was possible, and quotes him more bluntly still: "The whole purpose of building that phone was to shut down AT&T."

The device: about 2.5 pounds, roughly ten inches without the antenna, thirty minutes of talk on a ten-hour charge. Frenkiel, writing from the other side of the fight, describes the production descendant in six words that are the best short description of it anywhere — "a 2-pound hand-held unit that was about the size of a brick."

The patent, and the sentence nobody quotes

Six months later, on 17 October 1973, Motorola filed application 403,725. **US 3,906,166, Radio telephone system, issued 16 September 1975** to eight inventors: Cooper, Dronsuth, Leitich, Lynk, Mikulski, Mitchell, Richardson and Sangster. Read at source, it is a straightforward and rather beautiful piece of systems engineering, and it contains the single most citable thing in this file.

The abstract describes a base station whose transmission range deliberately exceeds the portable's, satellite receivers scattered through the coverage area to hear the handsets, a scanning receiver that camps on the strongest signalling channel, and automatic power control to "reduce the output power of each portable transmitter to the minimum level required for satisfactory communications in order to reduce battery drain and the interference caused by the portable transmitters."

And then this, in the abstract, as a co-equal feature of the invention:

> Means are also provided in the system to continuously locate a portable unit > and switch the operating frequency thereof as the portable unit moves between > base station transmitter coverage areas.

The description says why, and the modal verb is the point:

> the location of the unit must be continuously monitored in order that new > communications channel frequencies may be assigned thereto as required when the > portable unit moves between cells.

and how:

> the location function is provided by a group of receivers located at the base > station sites which monitor all of the active voice or communications channels

with the handoff decided by "a comparison … between the signal strength received by the various satellite receivers to determine which satellite receiver provides the best communications with the portable unit."

Continuous location of the person carrying the device is not a later addition to the mobile phone. It is a load-bearing requirement of the first handheld patent, present in the abstract, six months after the first call. Nobody smuggled it in. It is there because a phone that moves between cells cannot be handed off by a system that does not know which cell it is in — the surveillance affordance and the mobility are the same mechanism, and you cannot have the second without generating the first.

Two honest qualifications, because the finding is strong enough not to need inflating. The patent is explicit that the precision is coarse: reducing the portables' range "allow[s] less precise location of the portable units without causing interference between the portable units." And the design intent is plainly engineering, not observation — the location is wanted for channel assignment. What the patent establishes is not intent but inevitability: the record exists as a by-product of the architecture, and it is the coarse, cell-granular record, which is precisely the record at issue in Carpenter forty-five years later.

Ten years to a product

The demonstration was April 1973. On 21 September 1983 the FCC approved the DynaTAC 8000X — Motorola's own history calls it "the world's first commercial portable cell phone" — at a list price of $3,995. Commercial cellular service in the United States opened in Chicago the following month, run by Ameritech Mobile Communications; secondary accounts date the ceremonial first call to Soldier Field on 12 or 13 October 1983 and disagree about which.

The United States was not first. Frenkiel, with visible irritation: "the first Japanese system was installed in Tokyo in 1979, while the U.S. was still mired in regulatory and political delays."

Motorola's own page states the cost of the crossing: "After more than 10 years and a US$100 million investment, Motorola's commitment produced an innovative portable technology that revolutionized the communications industry." Reference accounts render the same figure as $100 million spent between 1973 and 1993, before revenues. The two framings are not reconcilable from the sources I reached, and the discrepancy is recorded rather than averaged.

And then the last irony of the fight, which Frenkiel supplies: at the moment cellular finally opened, the Bell System was broken up, the cellular business went to the regional companies, and when AT&T decided a decade later to re-enter the market it had spent thirteen years litigating for, it had to buy its way back in — purchasing McCaw Cellular in 1994. (Sources give the price as $11.5 billion and as $12.6 billion; I could not settle it.)

---

2. Why a reading would cite it

The admission test asks for a concrete occasion. There are five, and the first two are the ones this project will actually use.

1. It is the denominator. Every reading that reports a deployment eventually needs to convert an announcement into a fraction of humanity, and the conversion runs through the handset base. The current numbers, read at source:

Put an AI number against that base and it becomes a statement rather than a press release. Alphabet's own Q2 2026 disclosure, from Sundar Pichai on 22 July 2026: the Gemini app has 950 million monthly active users, with daily actives tripled in a year. Against 5.8 billion unique subscribers that is roughly one in six of the people on earth who have a mobile subscription at all — an enormous number, and one that names its own ceiling in the same breath. The reach of any consumer AI product is bounded above by a distribution channel that took from 1947 to 1983 to be permitted to exist, and a reading that quotes a user count without the denominator is quoting a numerator.

2. It supplies the substrate mass-collection-2013 asked for, and supplies it harder than expected. That entry's second half is a classifier over GSM metadata, and SCALE.md — the owner's landmark file, which this job reads and never writes — already carries a 2015 landmark for machine-learning classification of mobile-network metadata used to identify targets. Neither can explain where the metadata comes from without this file. The answer is section 1's patent quotation: the location record is generated because handoff requires it, from the first handheld design, and no policy decision anywhere created it.

The legal half of the same point was settled in Carpenter v. United States, decided 22 June 2018, 5–4, Roberts writing. Read at source, it is the best statement in any primary document of what this entry's device became:

> There are 396 million cell phone service accounts in the United States — for a > Nation of 326 million people.

> A cell phone faithfully follows its owner beyond public thoroughfares and into > private residences, doctor's offices, political headquarters, and other > potentially revealing locales.

> Only the few without cell phones could escape this tireless and absolute > surveillance.

The Court took 127 days of Timothy Carpenter's cell-site records — 12,898 location points, an average of 101 a day — and declined to extend the third-party doctrine of Smith and Miller to cover them, noting that most wireless carriers keep such records for up to five years. A reading arguing about what a model is allowed to be run over should have the 1973 patent and the 2018 holding in the same hand: the record was designed in, and it took forty-five years to acquire a warrant requirement.

3. The demonstration-as-lobbying template, with a known outcome. This is the citation occasion most likely to come up soonest. An AI laboratory stages a capability demonstration timed to a legislative hearing or a rulemaking comment period; the demonstration is real, the capability is real, and the purpose is regulatory. LENSES.md asks the policy lens for what governments did, not what they said they might do, and this entry is what a completed instance of that manoeuvre looks like when the file is finally closed: a working prototype built in ninety days for the purpose of moving a spectrum proceeding, an admitted motive — "the whole purpose of building that phone was to shut down AT&T" — and an outcome eight years later that gave the demonstrating party most of what it wanted. Cite it for the shape, and cite section 3 for the grading, because the outcome was a duopoly rather than an open market and it arrived a decade late.

4. A dated base rate for demo-to-deployment. 3 April 1973 to 21 September 1983: ten years and a hundred million dollars from a working handheld demonstration to a purchasable handheld product, in a case where the demonstration was genuine, the company was serious, the engineering was essentially finished, and the demand was real. The canon already carries this axis — darpa-grand-challenge-2005 on what a completed demonstration does and does not imply, musk-robotaxi-2019 on a demonstration whose deployment date kept moving, and proposals.md has ASIMO waiting on the same argument. This is the entry where the gap is measured against an outcome that eventually arrived, which makes it the useful end of the range rather than the cautionary one.

5. The constraint was policy for three decades, and a reading should be able to say so with dates. Concept 1947; spectrum refused 1947, 1949 and 1958; docket opened 1968; proposal filed 1971; rules 1981; service 1983. When the question is whether a capability is gated by what can be built or by what is permitted, this is a worked case where the answer was permitted for thirty-four years — with the caveat, from the Bell engineer himself, that the delay let the microprocessor arrive and that the final design still looked like the 1971 filing.

---

3. What it got right, and what it got wrong

Not required for moment, and the file is worth less without it. Every claim below is dated, and graded against what happened.

Cooper's own forecast, made as a joke, 1973 — nearly right, and right in the wrong units. He has retold it for fifty years: "we had a joke that said, in the future, when you were born you would be assigned a telephone number and if you didn't answer the phone, you were dead." Due: continuously. Graded against the ITU's 2025 figure of 112 mobile subscriptions per 100 inhabitants, the joke reads as a hit and is quoted as one. Graded against GSMA's 5.8 billion unique subscribers, about 70% of the population, it is a hit with roughly two and a half billion people outside it. The claim was right about the device class becoming universal and wrong to the extent anyone reads a subscription as a person; see section 4.

AT&T's position — that mobile telephony was a service to vehicles — 1947 through the 1970s. Wrong, and wrong in a way with a name. The IEEE History Center records that AT&T "regarded cellular as a better form of telephone service for moving vehicles that would fall under its national Bell System monopoly," and Frenkiel, from inside, says what the mistake was: a telephone attached to an automobile "was still a location-based phone rather than a 'personal' device … when we dial a cell phone, we are calling a particular person, and may not know or care whether that person is home or across the country." Motorola was right about the axis and had a commercial reason to be. Note the grading discipline this deserves: Motorola was not more visionary, it was differently threatened — its exposure was in vehicular radio, which a Bell cellular monopoly would have eaten, so the correct forecast and the self-interested one were the same forecast.

"Cooper's Law," Cooper, formalised later — a Moore's-law-shaped claim, and the one I could grade least well. The statement, as IEEE renders it, is that the maximum number of voice conversations or equivalent data transactions that can be conducted in all of the useful radio spectrum doubles every thirty months, and has done since Marconi. The claim is arresting and the sourcing is soft: reference accounts date the start to 1895 and others to 1897, and I could not reach a statement of it by Cooper himself. It also has a live technical challenge — work on the limits of network densification argues the historical rate cannot continue indefinitely by adding cells. Recorded here as a claim whose provenance is weaker than its circulation, and not leaned on.

The McKinsey forecast for AT&T, 1980: 900,000 US cellular subscribers by 2000. The most-quoted wrong prediction in this field, and the one this entry is most sceptical of. The actual US figure at the end of 2000 is conventionally given as about 109 million — the forecast at under 1% of the outcome. The magnitude is not in doubt. The provenance is: search traces the anecdote's earliest wide appearance to a 1999 Economist telecommunications piece I could not reach, no version of the study itself is public, and every retelling I found cites another retelling. It is repeated here because a reading will meet it, flagged as an anecdote of untraced origin rather than as a documented forecast, and section 4 refuses the causal story usually bolted onto it. If this project ever wants a base rate for how wrong technology forecasting runs, it should be assembled from predictions whose text survives — which is what the canon's prediction kind is for, and this is not one.

Motorola's implied claim that the demonstration would decide the docket — right, eventually, and by less than the retellings suggest. The 1981 Final Order gave two licensees per market rather than a Bell monopoly, which is the outcome Motorola spent a decade lobbying for, and the IEEE History Center draws the line from the handheld to the lobbying to the two licences without hesitation. Grade it honestly anyway: eight years elapsed, the decisive instrument was a regulatory proceeding rather than a press conference, and Frenkiel's account of the same years names four other blocs — broadcasters, fleet operators, incumbent manufacturers, Radio Common Carriers — each pulling on the docket for its own reasons. A demonstration that is remembered as having won an argument was one input into a thirteen-year proceeding. That is the correct lesson to carry into an AI-policy reading, and it is less flattering than the anniversary coverage.

Cooper's self-assessment, 2023 — the predictor grading himself, and worth recording as such. Speaking to the Associated Press at Mobile World Congress on 28 February 2023, fifty years on, he said what he had been worried about in 1973: "The only thing that I was worried about: 'Is this thing going to work?' And it did." And what he had not been worried about: "My most negative opinion is we don't have any privacy anymore because everything about us is now recorded someplace and accessible to somebody who has enough intense desire to get it." The independent grade differs from his in one direction only, and section 1 carries it: the recording he says he did not foresee was specified in the patent his team filed six months after the call — not the aggregation, not the retention, not the analysis, but the continuous locating that makes all three possible. He is honest about the outcome and generous to himself about the foresight. Both halves belong in any citation of him.

---

4. Commonly misused as

Not required for moment. Five misuses are common enough, and close enough to things a reading might want to say about AI, to be worth refusing.

1. "Martin Cooper invented the cell phone." The load-bearing error, because it is the form every anniversary uses. What Cooper led was the first handheld terminal and the first public call from one. The cellular system — small cells, low power, frequency reuse, handoff — is Ring and Young at Bell Labs in December 1947, a fact Motorola's own history concedes in print. The patent that came out of the demonstration carries eight inventors' names. A reading that needs an origin for the mobile phone should give the concept to Bell Labs, the handset to Motorola, and the twenty-six-year gap to the FCC.

2. "Surveillance was something that happened to the mobile phone later." The single most useful refusal in this file, and it is refused by a primary document rather than by argument. US 3,906,166, filed 17 October 1973: "the location of the unit must be continuously monitored." Location tracking is not a feature that was added to phones, not a consequence of smartphones, not an artefact of the app economy and not a policy failure. It is the mechanism that makes a moving telephone possible at all, and any argument that the tracking could be engineered away while the mobility is kept is arguing against the abstract of the founding patent. What is contingent — and therefore what a policy reading should actually be arguing about — is retention, aggregation, access and analysis. Carpenter is about the fourth of those, not the first.

3. "There are more phones than people, so everyone has one." ITU counts subscriptions: 9.2 billion, 112 per 100 inhabitants. GSMA counts unique subscribers: 5.8 billion, about 70% of the population — and 8.8 billion connections, which does not match ITU's 9.2 billion because the two organisations are not counting the same object. The gap between 112 per 100 and 70 in 100 is dual-SIM handsets, machine connections and second devices, and roughly two and a half billion people. For a project that reports deployment reach, subscriptions are the wrong unit and unique subscribers are the right one, and the 5G split — 84% coverage in high-income countries, 4% in low-income — is the reminder that "the world has phones" describes very different phones.

4. "McKinsey said 900,000, so AT&T walked away from cellular." Two claims welded together, and the weld does not hold. AT&T did not walk away: it filed the December 1971 cellular proposal, spent thirteen years fighting Docket 18262, and ran the first US commercial system in Chicago in 1983. What removed AT&T from cellular was the 1984 divestiture, which assigned the business to the regional companies; how deliberately it let those licences go is genuinely disputed in the trade record, with accounts variously claiming indifference, strategy, and a belief that it would not be permitted to keep them. AT&T bought its way back in a decade later by acquiring McCaw. A reading using this anecdote to illustrate that incumbents dismiss transformative technology is choosing the one case where the incumbent invented the technology, funded it for thirty-six years, and then lost it in an antitrust settlement.

5. "1973 shows how fast technology moves." It shows the opposite and it is one of the cleanest available demonstrations of the opposite. Concept 1947; handheld demonstration 1973; FCC approval of a purchasable handset 21 September 1983; a billion subscribers worldwide around the turn of the century. The distance from it works to you can buy it was ten years and a hundred million dollars with the engineering essentially finished at the start of it. Every step of the delay is documented and none of it was mysterious. This is the entry to reach for when a reading needs to say that a demonstration is not a deployment, and it is stronger than the usual examples precisely because the technology was real and the outcome eventually arrived.

---

5. Sources

Read directly, at source. Two documents were read as rendered page images. Fetch failures are recorded at the end rather than papered over.

Secondary, and flagged as such because I could not reach a primary.

Fetch failures, recorded rather than hidden.