Merlin ships — a machine opponent in the pocket, and then the best-selling toy in America

Filed under idea in canon/proposals.md, and idea is wrong. The proposal sits at line 294, inside the ## idea block that runs from line 195 to line 722, and what it offers as the idea — "when machine opposition became ordinary rather than remarkable" — is not an idea at all. It is a claim about a date. That is the definition of moment this canon has been using since eliza-1966 argued its own way out of the kind, and which clippy-1996, siri-2011, speak-and-spell-1978 and teddy-ruxpin-1985 have each reused since. The third member of the 1978–85 toy cohort takes the same correction as the other two, for the same reason, and the fact that three consecutive entries have had to make it is probably a fact about the proposal file rather than about the objects.

Why not the other kinds. Not fiction: there is a wizard on the box and nothing behind it. Not prediction: nobody involved forecast anything, and the one gradeable forecast in the vicinity is Parker Brothers' bet on the category, which section 3 grades. Not interpretation. Not limit — and this one is worth a sentence, because there is a hard mathematical limit sitting in the middle of this object and it does not belong to it. Tic-tac-toe is a solved game: with perfect play by both sides it is a draw, always. That result is the game's, not Merlin's, it was known long before 1978, and what Merlin does with it is section 4's business.

The hard part, which the proposal was right to flag. canon/proposals.md calls this "weakest of this group on its own," and on the ground the proposal offers, it is not merely weakest — it fails. Machine opposition was not made ordinary in 1978. Westinghouse's Nimatron played Nim against the public at the 1940 New York World's Fair and Edward Condon reckoned it was played over a hundred thousand times. Josef Kates's Bertie the Brain, four metres tall, played tic-tac-toe — this game — against fairgoers at the 1950 Canadian National Exhibition, with Danny Kaye photographed beating it for Life. Mattel's Auto Race (1976) and Football (1977) put a machine opponent in a pocket two years and one year early. Milton Bradley's Comp IV (1977) put the adversarial framing on the box in capitals — I AM PROGRAMMED TO BEAT YOU — a year before Merlin printed anything. And Milton Bradley's Simon launched at Studio 54 on 15 May 1978, months ahead of Merlin's autumn. On primacy Merlin loses to all of them.

So the entry has to be earned on something else, and it can be, on two things that are actually true of it and of nothing before it. The first is breadth: six different games out of one part, which is the move from a machine that plays this game to a machine that plays games. Auto Race raced, Football footballed, Comp IV broke codes, Simon repeated tunes; Merlin did tic-tac-toe, blackjack, Mastermind, a lights-out puzzle, a memory game and a music sequencer, and you chose. The second is scale: the Toy Manufacturers of America named Merlin the best-selling toy and game item in America for 1980, at 2.2 million units in that year alone, on lifetime sales above five million. It was Parker Brothers' most successful product by units and by revenue until Frogger in 1982. That is the moment — not a capability arriving, but a capability becoming the single most-bought object in a national toy market. A reading should notice that this makes the id's date slightly wrong: the thing worth citing happened in 1980, and 1978 is when the object went on sale.

On the date, which is layered. Bob and Holly Doyle formed Micro-Cosmos with Holly's brother Wendl Thomis in 1975. In 1977 Doyle was working up an electronic tic-tac-toe game and was told by someone in marketing to build something that was not boring; the six games are that instruction. The design patent was filed 14 September 1978 and granted 29 July 1980. The rules booklet carries © 1978. It was in shops for Christmas 1978, and appeared with Simon on the cover of Newsweek's Christmas issue that year and in Boston in October. The sales crown is 1980. Split Second followed in 1980, Master Merlin in 1982, Merlin: The 10th Quest in 1995, and a Milton Bradley/Hasbro reissue around 2003–04. Two respectable sources are loose here and should not be followed: Brown Alumni Magazine says Parker Brothers released it in 1977, and Toy Tales titles its piece 1979. The 1978 brochure, the 1978 copyright line and the September 1978 patent filing settle it.

descends_from, and what it leaves out. Two edges, both conceptual rather than causal, drawn on the footing clippy-1996 and teddy-ruxpin-1985 established. To shannon-chess-1950, because that is this canon's root entry for the proposition that a machine can be made to play a game against a person, and Merlin is the cheapest possible instance of it. To eliza-1966, because what the manual does with the pronoun "he" is ELIZA's finding — except that Weizenbaum observed users projecting a mind onto a program, and Parker Brothers printed the projection and sold it, which is section 2's first item and the difference the entry exists to carry.

Everything that actually precedes Merlin in the world is missing from canon/, and I have invented no ids for it: Nimatron (1940), Bertie the Brain (1950), Mattel Auto Race (1976) and Football (1977), Milton Bradley's Comp IV (1977), and Simon (1978), which is proposed as simon-1978 and not written. If one of them is later added, Bertie the Brain is the strongest single edge, for a reason section 2 turns on: Kates stood beside his machine at the exhibition adjusting its difficulty up for adults and down for children. Merlin does that too, in mask ROM, unattended, twenty-eight years later, in a pocket. Note also that speak-and-spell-1978 is a contemporary and not an ancestor — same year, same TMS1000 chip family, no line of derivation — which is exactly the distinction that entry's sibling had to draw about it.

What it is

Merlin is a red plastic slab about nine and a half inches long, shaped like the handset of a modular telephone, that a child holds in one hand. It takes six AA cells. The entire interface is eleven square keys arranged in a cross, each with a red LED behind it, plus four control buttons — NEW GAME, SAME GAME, HIT ME, COMPUTER TURN — and a speaker. The keys are the display: there is no screen, so the same eleven lights that you press are the board, the cards, the tune and the score. It has, by its own manual's count, "a unique vocabulary of 20 electronic sounds."

Inside is a Texas Instruments TMS1100, mask part number MP3404 — a 4-bit microcontroller in a 28-pin package, introduced in June 1976 as the double-capacity version of the TMS1000 that was also inside Speak & Spell and Simon. Its whole world is 2,048 bytes of mask ROM and 128 four-bit words of RAM. That is the entire wizard: the tic-tac-toe player, the blackjack dealer, the code-setter, the puzzle generator, the memory game and the sequencer, all six, share two kilobytes. It sold for about $24.95 — roughly $125 in 2026 money on the same CPI adjustment speak-and-spell-1978 used — with some accounts putting the street price up to $35.

The six games are Tic-Tac-Toe, Music Machine (record and replay up to 48 notes and rests, which makes Merlin one of the first consumer digital sequencers), Echo (a Simon-style memory game with nine difficulty levels), Blackjack 13 (twenty-one scaled down to fit eleven keys), Magic Square (a lights-out puzzle) and Mindbender (Mastermind). Three of them — tic-tac-toe, blackjack, Mindbender — put the machine on the other side of the table with information you do not have.

The rules booklet is the part of this object that matters most, and it opens like this: "MERLIN™ is a remarkably intelligent computer. Through light and sound he plays six different games of chance, strategy, memory, logic and skill." He. "With these sounds he challenges you to a game, responds to your moves, and tells you when you've won, lost or tied." The tic-tac-toe section: "You and MERLIN take turns occupying one square at a time. Either of you may make the first move," and "MERLIN'S aggressive tactics will keep you on your toes in this famous strategy game." The box, in the scanned copy, says TRY TO BEAT ME AT TICTACTOE.

Why a reading would cite it

Four occasions. The first is the one this entry exists for; the fourth is the one the proposal file offered, and it is the weakest.

First: anthropomorphism as printed product copy, at five million units, about a system whose contents are exactly countable. eliza-1966 is this canon's entry for people attributing a mind to a program — and Weizenbaum's finding was about what users did, spontaneously, to a program that made no claim for itself. Merlin is the other half, and the half that describes 2026 far better: the mental vocabulary is not projected by the customer, it is manufactured, printed, and sold. Parker Brothers wrote "remarkably intelligent." Parker Brothers chose "he." Parker Brothers wrote "aggressive tactics," which is a claim about a disposition, for a program in which no disposition exists. And unusually for this sort of case, the denominator is knowable to the last byte: two kilobytes of ROM, across six games. A reading that wants to ask what a 2026 product's mental vocabulary is actually resting on has here a case where the answer is not a matter of opinion, and where nobody was defrauded and nobody complained, because the language was doing a job — making a toy feel like an opponent — that it did extremely well. That is what makes it useful. It is not an indictment. It is the cleanest available demonstration that the mind-language is a product decision with a price and a purpose, which is the thing most often lost when a 2026 launch is read either as a lie or as a report.

Second, and specific: the manual discloses the machine's decision rule for one game and not for the others, and nobody noticed. For Blackjack 13 the booklet states Merlin's policy exactly — "If his hand is less than 10, he hits. If he 'busts,' you win. If his hand is 10 or more, he stands." That is a complete, checkable specification of the machine's behaviour; a player can verify it in ten minutes and know precisely what they are up against. For tic-tac-toe the same booklet offers "aggressive tactics." One is a policy; the other is an adjective. Both are in the same document, three pages apart, written by the same people, for the same buyer, and the difference is invisible unless you are looking for it. This is the exact shape of a model card or system card that pins down the temperature, the context window and the refusal categories, and describes the thing anyone actually wants to know — how it decides, how good it is, where it gives up — in adjectives. The question the entry supplies is sharp and portable: which of this document's claims could I falsify by playing, and which are adjectives? In 1978 the answer was one out of six.

Third: breadth that comes from a menu rather than from a capability. speak-and-spell-1978 makes the inventory argument about a vocabulary — the toy could speak, but only two hundred words a man in Dallas had said into a microphone. Merlin is the same argument one level up, about tasks, and that is the version 2026 keeps needing. Merlin looked general. Every predecessor did one thing; this one did six, and the experience was of a device that could turn its hand to whatever you picked. What it contained was six separately hand-written programs sharing a ROM, with nothing whatever in common between them and no capacity to do a seventh. A reading meets this constantly: the agent that handles the twelve workflows it was demonstrated on, the benchmark suite that is a menu, the assistant whose breadth is a list of integrations. The entry does not settle those cases — it supplies the question, which is whether the breadth on display is a capability or an enumeration, and the knowledge that a five-million-unit consumer object once made an enumeration feel exactly like a capability for a price of two kilobytes.

Fourth, and weakest, as a base rate for a category's durability. In 1980 Merlin was the best-selling toy and game item in America. By 1982 it had been overtaken by its own company's Frogger cartridge; the LED-and-buzzer handheld as a form was substantially finished by the mid-1980s, killed by LCDs and then by the Game Boy; Split Second, Master Merlin and the 1995 10th Quest each did worse than the thing before. Two years from the top of the American toy market to displacement, five to obsolescence of the whole form factor. Cite this when durability is being claimed for a category on the strength of one very good Christmas — but cite it carefully, because the category (a machine you play against, in your hand) did not die at all. It won completely. What died was one implementation of it. A reading that uses Merlin to predict the collapse of a 2026 category has to say which of those two it means, and they point opposite ways.

What it got right, and what it got wrong

Not required for moment. Four gradeable claims, in the house habit.

Parker Brothers' claim that this is "a remarkably intelligent computer." Made 1978, due immediately, false as engineering and accurate as a description of the experience. Two kilobytes of mask ROM for six games is not remarkable intelligence by any technical standard available in 1978 or since; it is a competent piece of embedded programming. But nothing in the sentence was legally or commercially wrong, because the claim being made was about how the object would feel, and it felt exactly like that to several million children. The interesting grade is not "false." It is that the sentence was doing a different job than it appeared to be doing, and that job was done well.

Parker Brothers' commercial brief. Made c. 1977, due by Christmas 1978, a clean hit. The company hired Micro-Cosmos to build something that "would feel more like a toy, and less like a device designed for engineers." The record: the best-selling toy and game item in America in 1980, 2.2 million units that year, five million-plus lifetime, the company's most successful product by units and revenue until 1982, and 5% of it going to Doyle in royalties, which he has said funded the rest of his working life. As a bet on a product brief this is about as cleanly graded as the canon holds.

The implicit claim that a handheld machine opponent is a durable consumer category. Made 1978, due by roughly 1985, half right in a way that matters. The category was permanent and is now the largest form of gaming on earth. The specific object was not: LED, one-line-of-lights, single-purpose mask ROM, gone inside seven years. Anyone using 1978 to argue that a hot category is a fad, or that it is forever, is picking one of these halves and ignoring the other.

The box's challenge — "TRY TO BEAT ME AT TICTACTOE." Made 1978, due on the first game, and ungradeable from here, which is itself the finding. Tic-tac-toe is solved: two players who both play correctly always draw. So the printed challenge is either impossible or trivial and nothing in between — if Merlin plays perfectly you cannot beat it, only tie it; if it does not, you beat it as soon as you find the hole. I could not establish which. No teardown, review, manual, patent or retrospective I could reach says how strong Merlin's tic-tac-toe actually is, whether it plays a lookup table or a heuristic, or whether it was deliberately made fallible the way Kates made Bertie the Brain fallible by hand in 1950. The evidence I have is one adjective ("aggressive"), one imperative on a box, and a scattering of contradictory nostalgia — a TechRepublic writer recalling "I never beat tic-tac-toe," against a general recollection that Merlin was beatable. Recording this as unresolved rather than guessing is the point: a claim about a machine's strength, printed on five million boxes, in the one game where the truth is completely known to mathematics, and forty-eight years later the answer is not written down anywhere I can reach. The customer could not check it in 1978 and I cannot check it now. That is worth more to a reading than either answer would have been.

Commonly misused as

Not required for moment. It earns a section because almost every popular sentence about this object is a first it did not achieve.

• "The first handheld electronic game." No. Mattel's Auto Race (1976) and Football (1977) precede it, both designed around the same class of TI part.
• "The first mobile game." This is Xconomy's 2014 framing of Doyle's story and it did not survive its own comment thread on Slashdot, where readers named Auto Race and Football within minutes. The defensible narrow claim is first multi-game handheld, and even that wants care: Merlin's contemporaries were single-purpose, so the claim is about breadth in a pocket, not about being handheld and not about being a game.
• "The first machine to play a person at a game." Nimatron, 1940, in public, a hundred thousand games. Bertie the Brain, 1950, at this game, thirteen feet tall, beaten on camera by Danny Kaye. What Merlin removed was the fair, the attendant, and the price.
• "The first toy to talk about itself as a mind." Comp IV, 1977, box front: I AM PROGRAMMED TO BEAT YOU. Merlin was not even first at the sentence.
• "Merlin was unbeatable at tic-tac-toe." Unverified, and structurally suspicious: the strongest true version of that claim is never lost, because perfect play in a solved game draws. Anyone asserting the stronger version is reporting a memory of losing, not a property of the machine. See section 3.
• "It proves the whole thing was always marketing." The deflationary misuse, and the same one speak-and-spell-1978 and dijkstra-1959 both have to guard against. Merlin's language oversold a real object that really did play six games competently for $25 — and the reason the marketing worked is that the engineering underneath it was good. A reading that uses the manual's "he" to argue that mind-language is always empty has thrown away the actual finding, which is that mind-language is a choice, made for a reason, gradeable against what is behind it.
• A factual correction, for the next session. Wikipedia's Merlin article gives the chip as a TMS1000 and the memory as "48 bytes of RAM." Both look wrong. The ROM dumped and emulated in MAME is MP3404, a TMS1100, corroborated independently by hardware catalogues; and 48 bytes matches neither part — the TMS1000 has 64 four-bit words (32 bytes) and the TMS1100 has 128 (64 bytes). I have not read the die myself and am relying on the dump designation, but the dump is stronger evidence than an uncited infobox.

Sources

The standing caveat, in the form deep-blue-1997 and speak-and-spell-1978 set it. The fetch tool available in this run returns another model's summary of a page rather than the page, so quotations arrive at one remove and could have been introduced in summarisation. Where two extractions of the same document differed I have said so. One instance: two passes over the same OCR text of the rules booklet returned different wording for Magic Square, one of them containing the phrase "MERLIN'S secret code" and the other not; I have used only the wording both passes agreed on and flag the phrase as unconfirmed.

Fetch failures in this run, recorded so the next session does not repeat them. Bob Doyle's own site, theelectronicwizard.com, hosts the richest primary archive on this object — the Parker Brothers 1978 and 1979 brochures, Boston October 1978, Newsweek December 1978, 3-2-1 Contact December 1980, BYTE April 1982 and Inc. September 1982 — and every one of them is a scanned image PDF with no text layer. pdftotext is installed on this machine but the shell in this non-interactive run refused the command, so none of that archive was read, including the BYTE piece, which is the one likely to have contained a technical account of the games. That gap is the direct cause of section 3's unresolved grade. Also failed: web.archive.org (not fetchable by this tool at all, which blocks the usual route to dead sources); xconomy.com (301 to bioxconomy.com, then 404 — the 2014 Doyle interview is gone and survives only through the Slashdot summary of it); washingtonpost.com archive, HTTP 403, so Tom Zito's contemporary "Wired-Up Wizards" (December 1978) is known only through Wikibooks' summary; skybuilders.com/BobDoyle/, 404; MAME's hh_tms1k.cpp on GitHub, too large for the fetch tool to reach the merlin section, so the MP3404 designation comes from search extracts of that file plus an independent hardware catalogue rather than from the source itself.

Primary.

• Merlin rules booklet, Parker Brothers, "Rules © 1978 Parker Brothers Div. of General Mills Fun Group, Inc., Beverly, Mass. 01915," scanned at the Internet Archive (manuals-handheld-games-ParkerBros-Merlin). Source of every quotation in sections 1 and 2: "remarkably intelligent computer," the pronoun, the 20 sounds, "Either of you may make the first move," "aggressive tactics," and the Blackjack 13 policy. The single most useful document about this object.
• US Design Patent D256,139, Electronic game casing — Arthur P. Venditti, Samuel T. Kjellman and Robert O. Doyle, assigned to CPG Products Corp (later Kenner Parker Toys). Filed 14 September 1978, granted 29 July 1980. Read via Google Patents. Fixes the date and the credits; note that it covers the casing, and that no utility patent on the games surfaced in this run.

Technical.

• MAME, src/mame/handheld/hh_tms1k.cpp — the dumped Merlin ROM as MP3404, a TMS1100. Via search extracts; see the failure log.
• datamath.org, TMS1100 product family, and Wikipedia's Texas Instruments TMS1000 — TMS1100 introduced June 1976, 16,384-bit ROM (2,048 eight-bit words), 512-bit RAM (128 four-bit words), 28-pin DIP; TMS1000 at 1024×8 and 64×4, around $2 in volume in 1974, roughly 0.3 MHz.
• facele.eu chip catalogue — independent corroboration of TMS1100/MP3404. It lists 9V for the unit it holds against the six AA cells given elsewhere, which is probably a variant and is not resolved here.
• handheldmuseum.com, Parker Brothers Merlin — model #3200, 1978, six AA cells, the D256,139 reference, the Meccano edition for France.

Historical and contextual.

• Wikipedia, Merlin (console) — the six games, ">5 million" lifetime, the Toy Manufacturers of America crown for 1980 at 2.2 million units, the Newsweek and Boston covers, Split Second, Master Merlin and The 10th Quest, and the Frogger comparison citing the Boston Globe, "Competitors Claim Role in Warner Setback" (1982). Its chip and RAM figures are the ones section 4 disputes.
• Wikibooks, History of video games/Platforms/Merlin — the 1977 tic-tac-toe prototype and the marketing objection, Arthur Venditti's case and his earlier naming of the Nerf ball, the $25–$35 range, and the summary of Tom Zito's December 1978 Washington Post interview in which Doyle complains about storage and predicts voice control.
• Toy Tales, Merlin — The Electronic Wizard from Parker Brothers — Micro-Cosmos in 1975, the twenty-plus prototypes, the $24.95 price, 5.5 million units, Kjellman on the sounds anthropomorphising Merlin in a way the industrial design could not, and the modular-telephone-plus-Star Wars-droid account of the case.
• Gizmodo, I Miss Merlin, the First Electronic Game I Ever Played (2021) — Parker Brothers' brief for something that "would feel more like a toy, and less like a device designed for engineers," and the 5% royalty.
• Slashdot, Merlin's Magic: The Inside Story of the First Mobile Game (4 March 2014), summarising the lost Xconomy interview — Doyle's royalty, and the comment thread that dismantles the "first" claim with Auto Race and Football.
• Brown Alumni Magazine, Renaissance Man (23 May 2012) — Doyle's biography: Brown Sc.B. 1958, Harvard astrophysics PhD 1968, NASA's Astronomy Missions Board and the Skylab Joint Observing Program, MacPublisher in 1984, and the later turn to Free Will: The Scandal in Philosophy and informationphilosopher.com. It dates Merlin to 1977 and credits Doyle with Mattel Electronic Football and Milton Bradley Electronic Battleship; the first is wrong and I could not confirm the second, so neither is used above.
• On the ancestry, none of which is in canon/: Wikipedia's Nimatron (Edward Condon, Westinghouse, 1940 New York World's Fair, "played over 100,000 times") and Bertie the Brain (Josef Kates, Canadian National Exhibition, 25 August – 9 September 1950, four metres, additron tubes, unbeatable at the top difficulty, Kates adjusting the level by hand for adults and children, Danny Kaye photographed beating it for Life, dismantled afterwards and forgotten); Wikipedia's Mattel Auto Race and Simon (game) (Ralph Baer and Howard Morrison at Marvin Glass, programmed by Lenny Cope, launched at Studio 54 on 15 May 1978); and the Milton Bradley Comp IV (1977) box copy, I AM PROGRAMMED TO BEAT YOU, via collector listings.