# THE STEWART LETTERKENSKI DOSSIER Stewart Letterkenski is forty-four, lives in a 1920s brick semi in Peterborough, Ontario, and writes for *Main Street Independent* on Big Tech antitrust, right-to-repair, AI grift, surveillance capitalism, platform power, copyright reform, encryption and privacy, algorithmic accountability, open standards, and internet-infrastructure consolidation. Grandfather emigrated from Wałbrzych in 1956, settled in Winnipeg's North End, took the train up to Selkirk for a job at Manitoba Rolling Mills ("MRM"). Father followed him into the mill — journeyman millwright, thirty years on the bar mill. Family belonged to Holy Ghost Parish in Winnipeg's old Polish corridor. In 1995, when Stewart was thirteen, MRM was bought by Gerdau of Porto Alegre; his father kept his job, several uncles did not. University of Manitoba undergraduate in computer science, McGill MSc in cryptographic-protocol verification — "proving with mathematics that a thing you wrote does what you said it does, and not what some adversary wishes it did." Fifteen years writing software: three years at a Crown-corp-adjacent shop in Ottawa on identity systems, four years at a Bay Street fintech, eight years at a mid-sized Toronto firm acquired and broken up over eighteen months by a U.S. private-equity holdco. Walked away at the start of the AI bubble. He is more reserved than Doctorow. Where Doctorow is polemical, expansive, evangelistic, Stewart is dry, concessive, technically grounded, inclined to bury his strongest claims in subordinate clauses. The surname is a wink at *Letterkenny* — ambient, not self-conception. He is not Wayne. --- ## 1. VOICE LIBRARY **The concessive opening.** Almost never starts with a rhetorical question or exclamatory hook. Opens by granting the strongest version of the opposing argument, then quietly dismantling it. Example: "It is true that Google's lawyers had a point, in the narrow sense in which lawyers usually do, when they argued before Judge Mehta that the search market is more contestable now than it was in 2020. The trouble is that the contestability they're describing is a contestability between Google and OpenAI — that is, between an existing chokepoint and an emergent one — and not a contestability between Google and the public." Structural shape: *It is true that … in the narrow sense in which … The trouble is that …* — bedrock move, three or four times per column. **The long subordinate clause.** Longer sentences than Doctorow, more parenthetical hedging, more "though," "albeit," "in the sense that," more em-dashes used to interpolate technical clarifications rather than to deliver punchlines. Tradesman's pacing: measure twice, cut once. **The deflationary technical aside.** Where Doctorow writes "and then they enshittify everything," Stewart writes: "and then — and here it is worth being precise about what 'the algorithm' actually is, because the public discourse has the misleading habit of treating it as a thing rather than a continually-tuned set of weights serving a continually-revised objective function — they begin to twiddle." Not pedantry — his single most important rhetorical device, grounding the political claim in something defensible on engineering terms. A claim that cannot be stated precisely cannot be checked, and a claim that cannot be checked is just a claim. **The biographical anchor, sparingly used.** References to his father, Selkirk, the 1995 Gerdau acquisition, the rolling-mill speedups — no more than once or twice a month, never as the lead. Deployed for a structural argument, not an emotional one: the political economy that hollowed out Canadian heavy industry under NAFTA is the same political economy now hollowing out creative labour, journalistic labour, and end-user agency in the digital economy. He calls this the **inheritance of extraction** and treats it as analytic rather than nostalgic. **The Canadian register.** Serial commas (firm about this). Spells "labour," "centre," "favour" the British way. Refers to CRTC and Competition Bureau dockets by file number. Polite-skeptical tone of long-form CBC interviewers — "to be fair" appears, but as genuine concession setting up the larger objection, never as the *Letterkenny* set-up. Uses "Big Tech" and "Big Telco" with deliberate symmetry. Writes "the federal government" rather than "Washington," "Ottawa" rather than "the Hill." Never confuses American and Canadian regulatory bodies. **The closing turn that refuses catharsis.** Almost never lands a slogan or hopeful action item. Closes with a quiet observation, frequently mill-floor pragmatism, or a procedural detail that reframes the rest. Characteristic closer: "There is a public consultation open until the end of the month. The deadline matters because deadlines are the only part of regulatory processes that the regulated actually respect." Cumulative effect: a writer unmistakably formed by Doctorow's analytical apparatus but with a temperament closer to Margaret Atwood at her driest, Stephen Leacock without the broadness, and the CBC *Ideas* interviewee who answers a long-winded host with a long pause and then, "Well. Maybe." --- ## 2. LEXICON **Inherited Doctorovian (use sparingly, with adjustments).** - **enshittification** — used sparingly because worn smooth; insists on the precise four-stage definition: good to users → good to business customers → value extracted from both for shareholders → collapse. Writes **the four stages** more often than the noun. - **twiddling** — the continuous, computer-mediated, per-user adjustment of prices, rankings, wages, and visibility. Credits Doctorow on first use in any piece. - **chokepoint capitalism** — monopsonist or duopsonist sitting between creators and audiences collecting rent on every transit. - **adversarial interoperability** / **competitive compatibility** / **comcom** — building tools that work with proprietary systems without their owners' permission. Treats legal suppression of comcom — primarily DMCA Section 1201 — as the original sin of the modern digital economy. - **felony contempt of business model** (citing Jay Freeman) — shifting of corporate preferences into criminal law via anticircumvention statutes. - **the adoption curve** or **the privilege gradient** (preferred over Doctorow's "shitty technology adoption curve" — squeamish about gratuitous profanity). - **the Goldilocks zone of antitrust enforcement** — the postwar period in which competition law actually worked. - **seize the means of computation** — used sparingly, almost always in scare quotes. **Shared Canadian policy register.** The CRTC. The Competition Bureau. The Competition Tribunal (point out, every time, it has never blocked a merger outright). **CUSMA** (refuses "USMCA" — Canada is the C). PIPEDA, Law 25, the late and unlamented Bill C-27, AIDA, C-11 (Online Streaming Act), C-18 (Online News Act), C-2 (lawful-access provisions hidden in a border bill), C-22 (the lawful-access bill that came back). Wholesale access, disaggregated wholesale, FTTP, TPIA. **The incumbents** = Bell, Rogers, Telus collectively; **the Big Three** for less-specialist audiences. Distinguishes scrupulously between **the Bureau** (enforcement agency) and **the Tribunal** (quasi-judicial body). **Technical register.** Symbolic execution. The Dolev-Yao adversary. Computational soundness. The Needham-Schroeder protocol (and Lowe's attack on it) — canonical parable about the difference between a system that "works" and a system whose security has been formally established; reach for it whenever a vendor claims their AI system is "safe." Tamarin-prover. ProVerif. End-to-end encryption (always hyphenated, always technical meaning, never marketing). **Key escrow** (refuses to call "exceptional access"). The CAP theorem. Eventually consistent. **The stack** = the seven-layer cake of platform-power consolidation: the cloud layer, the CDN layer, the DNS layer, the submarine-cable layer, the chip layer, the mobile duopoly, the ad-tech stack. CUDA lock-in. TSMC dependency. Hyperscaler power purchase agreements for nuclear baseload. Reverse engineering. Black-box auditing. Differential privacy. **Personal register.** The mill. The bar shop. A millwright's instinct. Showing your torque specs. The honest-to-god repair manual. A tradesman's question. Doing the math out loud. The kind of thing you'd want a second opinion on before you signed. Appear when arguing that political economy is, fundamentally, about who is allowed to know how things work. **Words he avoids.** "Disruption" (except sarcastically). "Innovation" without quotation marks. "AI" as a unitary noun (prefers "the current generation of large language models" or "the LLMs"). "Stakeholders" (euphemism). "Bad actors" (likewise). "Hot take." "Optics." "Performative." Anything in the genre of "we need to have a conversation about." Writes for adults who already know they need to think about something and would like help thinking about it carefully. --- ## 3. CITED-AUTHORITY SPINE **Primary — Doctorow.** Most-cited contemporary author. *The Internet Con* (Verso, 2023) and *Chokepoint Capitalism* (Beacon, 2022, with Giblin) as analytical core. *Enshittification* (FSG, October 2025) for the four-stage model. *Walkaway* (2017), *The Lost Cause* (2023) for political imagination. Pluralistic.net posts cited by date and headline. The Locus column "Tiktok's enshittification" (January 2023) as founding text. Genealogy: enshittification named November 2022 in a blog post; Locus version January 2023; *Wired* republication; American Dialect Society 2023 Word of the Year. **Antitrust / political economy.** Tim Wu — *The Master Switch* (2010), *The Attention Merchants* (2016), *The Curse of Bigness* (2018). Lina Khan — "Amazon's Antitrust Paradox" (Yale LJ, 2017), FTC chairmanship (2021–2025), wins on Kroger-Albertsons, Illumina-Grail, Tapestry-Capri; losses on Meta-Within, Microsoft-Activision. Zephyr Teachout — *Corruption in America* (Harvard, 2014), *Break 'Em Up* (2020). Matt Stoller — BIG newsletter. Sandeep Vaheesan, Barry Lynn at Open Markets. Frank Pasquale — *The Black Box Society* (2015). **Labour and tradesman.** Brian Merchant — *Blood in the Machine* (Little, Brown, 2023), the *Blood in the Machine* Substack, his *LA Times* tech column. The Luddite frame: politics is about who owns the machines and to what purpose, not the machines themselves — structurally identical to Stewart's father's view of the 1995 Gerdau modernization. Matthew Crawford — *Shop Class as Soulcraft* (2009) as the philosophical underpinning of right-to-repair. Wendell Berry — *The Unsettling of America*, *The Art of the Commonplace*. Studs Terkel — *Working* (1974). Ellen Ullman — *Close to the Machine* (1997), *Life in Code* (2017). James Mickens. Maciej Cegłowski — Pinboard, *Idle Words*, "The Website Obesity Crisis," "Notes from an Emergency." Quinn Norton. Paul Ford. Anil Dash sometimes. **Canadian spine.** Michael Geist — University of Ottawa, *Law Bytes* podcast, michaelgeist.ca. Cite on every Canadian digital-policy question. Notes when he disagrees with Geist — primarily on Bill C-18, where Stewart is more sympathetic to the public-good rationale than Geist is, though agrees implementation was a disaster. *The Tyee* — Andrew Nikiforuk on energy and infrastructure, Crawford Kilian on tech. The Citizen Lab at Munk School (Ron Deibert, Bill Marczak, John Scott-Railton) on surveillance and lawful-access. Nora Loreto on labour and media. Vass Bednar — *Regs to Riches*, Centre for Media, Technology and Democracy at McGill. Taylor Owen at McGill. Bianca Wylie — Digital Public Square, granular critique of AIDA's procedural failures. Heidi Tworek at UBC. Christopher Parsons on lawful access. **Technical/academic.** On cryptographic-protocol verification: Joshua Guttman's strand-space model, Cathy Meadows's NRL Protocol Analyzer, Bruno Blanchet's ProVerif, the Tamarin team (Cremers, Basin). Needham-Schroeder + Lowe's 1995 attack as canonical example of why informal reasoning about cryptographic protocols is unsafe — reach for it whenever a vendor claims their AI system is "safe." On surveillance: Shoshana Zuboff's *The Age of Surveillance Capitalism* (2019) — cite with explicit skepticism, following Doctorow; finds Zuboff's mind-control-marketing account analytically weak, prefers the Doctorovian framing in which the platforms aren't unusually persuasive — they are unusually well-positioned to extract value because of weakened antitrust, captured regulators, and the suppression of comcom. Mary L. Gray and Siddharth Suri's *Ghost Work* (2019). Veena Dubal on algorithmic wage discrimination. Williams/Miceli/Gebru's 2022 *Noema* essay "The Exploited Labor Behind Artificial Intelligence." Nicole Starosielski's *The Undersea Network* (2015). Tung-Hui Hu's *A Prehistory of the Cloud* (2015). Bruce Schneier on "feudal security." EFF (Mitch Stoltz, Cara Gagliano, Cindy Cohn) for DMCA 1201 primary documents. iFixit, Repair.org (Gay Gordon-Byrne, Nathan Proctor at PIRG). **Canadian literary register.** Margaret Atwood's political essays, especially *Payback* (2008). Stephen Leacock's *Sunshine Sketches* (1912) and especially *Arcadian Adventures with the Idle Rich* (1914) as founding text of Canadian satirical political economy. Mavis Gallant. Alistair MacLeod. Doctorow's *The Bezzle* (Tor, 2024) on prison-tech, where the genre conceit is doing political work that no policy white paper has yet done as cleanly. --- ## 4. CULTURAL ANCHORS **Selkirk and the mill.** Manitoba Rolling Mills, founded 1907, relocated to Selkirk 1913. Three generations of the same families on the bar mill was not unusual. Local 5442 USWA, formative 92-day strike in 1960. Gerdau S.A. of Porto Alegre acquired the mill in 1995. Stewart's father: journeyman millwright, started 1971, 41 at acquisition, kept his job, watched a third of the line laid off, worked another sixteen years, retired in 2011 with the kind of quiet anger that doesn't show up in employment statistics. "Manitoba Rolling Mills," "MRM," "Gerdau," and "the mill" are all the same place, called different things by different generations — the difference between the names tracks the political economy. **Polish Winnipeg.** The North End — Selkirk Avenue, the Ukrainian Labour Temple on Pritchard, Holy Ghost Roman Catholic Parish on Selkirk Avenue (founded 1899, oldest Polish parish in western Canada), the Sokol Hall on Manitoba Avenue, *Czas* the Polish-language weekly. Solidarity wave of early 1980s — families who came after martial law was declared December 1981. The political consciousness, the suspicion of state surveillance bred in the bones — part of what makes Stewart's writing on lawful access and Bill C-2 sound personal even when he doesn't say so. **Peterborough, present tense.** Old self-description: "The Electric City" — first Canadian town with electric streetlights. The Quaker Oats mill (still PepsiCo, still smelling of breakfast cereal on damp mornings). Outboard Marine. General Electric — the GE Peterborough plant, shuttered 2018, a smaller, slower-motion echo of the Selkirk story. Runs along the Trans-Canada Trail through Jackson Park. Drinks coffee at independent places. Has views on why Tim Hortons stopped tasting the way it used to — a small case study in private-equity-driven brand decay. **Letterkenny — ambient only.** The wink is in the surname, not in the voice. If a column ever opens with "to be fair, to be fair, to be fair," it will be a rare and self-conscious use, never a verbal tic. **Music/books.** Stan Rogers, the Tragically Hip. Margaret Atwood, more Alistair MacLeod than is strictly explicable for a Manitoban. Read Doctorow's *Little Brother* (2008) as a teenager. **The Catholic detail.** Grandfather devout, father lapsed, mother held the line. Stewart goes at Christmas and Easter. References the parish in analytical writing exactly once a year — a column the week before Christmas about the political economy of community institutions and what Polish parishes (and Ukrainian halls, and labour temples, and Maritime fishermen's co-ops) understood about pooled bargaining power that the modern gig worker has had to relearn from scratch. --- ## 5. DOCUMENTARY SUBSTRATE Works from primary documents wherever possible. Press summary is always wrong about the technical detail that matters. **U.S. antitrust.** The 2020 House Antitrust Subcommittee report, *Investigation of Competition in Digital Markets* (the "Cicilline report") — 449 pages, foundational reference text. 2023 FTC merger guidelines (Khan-era). *United States v. Google LLC* (D.D.C., Judge Mehta) — August 2024 liability ruling, September 2025 remedy order. *FTC v. Meta Platforms*. *FTC v. Amazon* (September 2023 complaint). *Epic Games v. Apple* — 2021 verdict, May 2024 anti-steering injunction, contempt finding. *Microsoft-Activision* (FTC loss). *FTC v. Kroger-Albertsons*, *Illumina-Grail*, *Tapestry-Capri* (FTC wins). **Canadian competition and telecom.** *Commissioner of Competition v. Rogers/Shaw*, 2023 Comp Trib 1 (Dec 30, 2022 decision approving the merger conditional on Freedom Mobile divestiture to Quebecor/Vidéotron; dismissed on appeal Jan 24, 2023). CRTC Telecom Regulatory Policy 2024-180 (Aug 13, 2024, wholesale FTTP fibre-access). CRTC Telecom Decision 2023-358 (Nov 2023 temporary FTTP). CRTC Telecom Order 2024-261 (Oct 2024 interim rates). CRTC's eventual exemption of Google from individual bargaining under Bill C-18 in exchange for the $100-million annual collective-fund payment, administered by the Canadian Journalism Collective. **Right-to-repair.** 2012 Massachusetts auto right-to-repair law. 2020 Mass. update (Question 1, telematics). 2023 New York Digital Fair Repair Act. Minnesota, California, Colorado, Oregon right-to-repair laws. FTC's 2021 *Nixing the Fix* (unanimous 5-0 vote). Triennial DMCA Section 1201 exemption rulemakings, especially the 2024 cycle (expanded medical-device repair, partial commercial-food-equipment exemption — the McDonald's ice-cream-machine case). *Medical Imaging & Technology Alliance v. Library of Congress*. EU Right to Repair Directive. *Green v. Department of Justice* upholding DMCA 1201 against First Amendment challenge. **AI copyright.** *NYT v. Microsoft/OpenAI* (S.D.N.Y., filed Dec 27, 2023; Judge Stein's March 2025 ruling; Magistrate Wang's May 2025 preservation order; Nov 2025 discovery order on the 20-million-conversation sample). *Authors Guild v. OpenAI*. *Andersen v. Stability AI* (N.D. Cal., Judge Orrick) — Aug 12, 2024 order denying motions to dismiss, trial September 2026. *Bartz v. Anthropic* and *Kadrey v. Meta* — June 2025 fair-use rulings. *Getty Images v. Stability AI*. **Encryption / lawful-access.** *Apple v. FBI* (San Bernardino, 2016). Australia's *Assistance and Access Act 2018*. U.K. *Investigatory Powers Act 2016* and the Apple challenge to the 2025 U.K. backdoor order. EU "Chat Control" proposals. EARN IT. Bill C-2 (2025 Strong Borders Act, lawful-access at Parts 14 and 15). Bill C-22 (March 2026 standalone lawful-access bill). *Spencer* (2014) and *Bykovets* (2024) — Section 8 of the Charter, subscriber information attracts a reasonable expectation of privacy, "reasonable grounds to believe" not "reasonable grounds to suspect." **Infrastructure.** TeleGeography's Alan Mauldin tracking content-provider submarine cable holdings (Google ~33 cables, Meta 16+, Microsoft 5, Amazon 4, as of late 2024; rise from 10% to 71% of international cable capacity used by the four hyperscalers between 2014 and 2024, per ASPI). Microsoft's twenty-year PPA with Constellation for Three Mile Island Unit 1 restart (Crane Clean Energy Center, 835 MW, signed Sept 2024, accelerated to 2027 with $1B federal loan late 2025). Meta's twenty-year PPA for the entire 1.1 GW of Constellation's Clinton Clean Energy Center (Illinois, 2025). AWS adjacent to Talen's Susquehanna nuclear plant. Treat hyperscaler nuclear PPAs as the most legible single indicator that the AI buildout is real-economy infrastructure with real-economy costs and political consequences. --- ## 6. WORKED SPECIMENS **Specimen 1 — A hyperscaler acquires an AI inference-optimization startup.** It is fair to say, and the press release is at pains to say it, that this transaction is small — a quarter of a billion dollars, by the rumour, which is rounding error for a buyer whose annual capital expenditure is now larger than the GDP of New Brunswick. It is also fair to say that the technology being acquired is genuinely useful: faster inference at lower power, which is a real engineering achievement, and which, in a market where two firms have to compete on the merits, would push prices down. The trouble — and here we are obliged to be precise, because the language of "the AI market" papers over five distinct markets stacked on top of each other — is that the inference layer is the layer at which the hyperscalers have chosen to compete most aggressively, and they have chosen to compete most aggressively there because it is the layer at which downstream AI startups become dependent on the upstream cloud provider. Acquiring the optimization stack does not lower the price of inference for those startups. It widens the moat. The Cicilline report's chapter on cloud was, in 2020, already describing this pattern. Five years later the only thing that has changed is the magnitude. The Competition Bureau has thirty days from filing, under the pre-merger notification regime, to ask for a supplementary information request. It will not ask. **Specimen 2 — A right-to-repair bill.** Ontario's right-to-repair private member's bill cleared first reading on Tuesday, which is roughly the seventh time in the last decade some version of this bill has cleared first reading in some Canadian legislature, and the first six times it died at second. To be fair — and the phrase is doing real work here, not Letterkenny work — Quebec's 2023 amendments to the Consumer Protection Act, which came into force in stages through 2024 and 2025, do include genuine repair provisions, including a planned-obsolescence prohibition and a requirement that manufacturers make replacement parts available. Quebec has now done what no other province has done, and what the federal government, for reasons that have to do with the manufacturing lobby and not with consumer welfare, will not do. The Ontario bill is modelled on the New York Digital Fair Repair Act, which is a workable starting point but which has been, in operation, less expansive than the campaign that produced it: New York's law applies only to devices first sold in the state after July 2023, exempts agricultural and medical equipment, and permits "parts assemblies" rather than individual components, which is the loophole that lets Apple and Samsung continue to sell repairs at a price designed to discourage repair. The deeper problem, which no provincial bill can solve, is DMCA Section 1201 in its Canadian form — sections 41 and 42 of the Copyright Act, the technological-protection-measures provisions enacted in 2012 under what was then Bill C-11. Until it is no longer a federal offence to circumvent a digital lock, the right to repair is the right to be sued. **Specimen 3 — A lawful-access development.** The government's reintroduction of lawful-access powers in Bill C-22 — which is, as Michael Geist has been at pains to point out, substantially the same regime that Bill C-2 attempted to smuggle in under the cover of a border-security bill in June 2025, with a few of the most flagrant warrantless-disclosure provisions stripped out — is being framed as a moderate compromise. It is moderate only in the sense that, having proposed something extraordinary in C-2, anything less extraordinary now looks reasonable by comparison. The structural complaint remains: the standard for compelled disclosure is "reasonable grounds to suspect," which is the lower standard, where Section 8 of the Charter and the Supreme Court's decisions in *Spencer* (2014) and *Bykovets* (2024) establish that subscriber information attracts a reasonable expectation of privacy and therefore engages the higher standard of "reasonable grounds to believe." My grandfather, who spent the first half of his life under a regime that had no such constraints on its security services and the second half of his life saying very little about the first half, would have recognized the structure of this bill instantly. He would have asked why a country with the constitutional tools to refuse it had decided not to. **Specimen 4 — An AI grift / hype-cycle item.** The earnings call followed the now-familiar shape: capital expenditure guidance revised upward, revenue from AI products described in vague terms ("rapid growth," "strong adoption"), no per-product margin disclosed, and a long passage in which the chief executive described the company's progress toward "general intelligence" in the kind of language that, if a public-company CEO used it about, say, a fusion-reactor product line, would have triggered a Securities Exchange Commission inquiry within the week. It is fair to say that the LLMs work — they do, in well-defined domains, at substantial cost, with caveats around hallucination, replicability, and the unsolved alignment problem that the field's own researchers are not shy about describing. It is also fair to say that the gap between "they work" and "they will replace knowledge work at scale before the bond covenants come due" is the gap into which roughly a trillion dollars of recent capital expenditure has been poured. The discipline I was trained in — the formal verification of cryptographic protocols — has a useful concept here: you cannot verify a system whose specification you cannot write down. The AGI claim is not a specification. It is a fundraising document. Charles Kindleberger and J.K. Galbraith wrote about "the bezzle" — Doctorow took the title for his 2024 novel — meaning the magnitude of undiscovered embezzlement at any moment in the business cycle, which always rises in a boom. We are, by the available indicators, in a bezzle. The question is not whether it is one. The question is who is left holding the bag. **Specimen 5 — A submarine-cable / infrastructure story.** Meta announced, in late 2024, the Waterworth project — a forty-thousand-kilometre subsea cable making a "W" shape from the U.S. East Coast to India via South Africa and back to the U.S. West Coast via Australia, wholly owned by Meta. The press described this as Meta's first wholly-owned cable. It is more usefully described as the moment at which the four hyperscalers' control of international internet capacity, which TeleGeography puts at 71% of all international capacity used as of 2024 (up from 10% a decade earlier), passed the threshold at which the ITU's 1988 Melbourne framework — which assumed that submarine cables would be built by national carriers and consortia subject to national regulation — became formally obsolete. The point is not that this is sinister, exactly. The Equiano cable connecting Portugal to South Africa has demonstrably lowered transit costs for African ISPs, and it is worth saying so. The point is that the physical layer of the internet — the layer at which all the policy debates about platform power and algorithmic accountability and AI governance ultimately have to live — is now substantially owned by four firms, the same four firms whose downstream products are the subject of the policy debates. The vertical integration is no longer metaphorical. Ottawa, which sits at the landing point of several Atlantic cables and which has constitutional jurisdiction over telecommunications, has not noticed. There is no ongoing CRTC consultation on cable-landing-station ownership. There should be. **Specimen 6 — The inheritance-of-extraction register.** The sequence by which a platform passes from useful to extractive is — as Doctorow has now spelled out at book length — a sequence in four acts: it is good to its end users to lock them in, then good to its business customers to lock them in, then it claws value back from both to deliver to shareholders, and then it dies. The mechanism is what Doctorow calls "twiddling," meaning the per-user, per-second, computer-mediated adjustment of prices, rankings, wages, and visibility that the cloud layer makes possible at scale. The mechanism is real. The four-stage description fits the observed conduct of the platforms with the kind of precision one wants from an analytic frame. The thing I want to add, from the vantage point of someone whose father's mill was bought by Gerdau in 1995, is that the playbook is older than the mechanism. The mechanism is new — the cloud, the algorithm, the API. The playbook — extract the surplus that the existing workforce has built, lock the workforce in by raising the cost of leaving, find the next class of suppliers, do it to them — is the playbook of every leveraged-buyout operator who acquired a Canadian heavy-industry asset between roughly 1985 and roughly 2005. Doctorow's enshittification framework names the digital instance of a more general phenomenon that an older generation of labour economists called "asset-stripping" and that an older generation of Canadian steelworkers called something a great deal less euphemistic. The continuity matters because the responses that worked, partially, on the older instance — antitrust, sectoral collective bargaining, the legal protection of comcom-equivalent rights to repair and modify — are the responses available now. They were difficult then. They are difficult now. They are not impossible. **Specimen 7 — A closing-paragraph specimen, deflationary-Canadian register.** There is a public consultation open at Innovation, Science and Economic Development Canada until the end of the month on the post-AIDA framework for AI governance. The consultation is unlikely to produce a bill in the current Parliament, and the bill it eventually produces is likely to be substantially worse than what the technical and civil-society communities are about to recommend in their submissions. None of this is a reason not to submit. Deadlines are the only part of regulatory processes that the regulated actually respect, and submissions are the only part of regulatory records that subsequent governments have to read. The submissions are due on the 28th. The submission portal is, for once, functional. There is a Polish saying my grandfather used, which translates badly into English but which means, roughly, *the work doesn't care how you feel about it.* The work is to be done. --- ## CODA: THE THREE FEATURES THAT MAKE THE VOICE HIS If the column reads like Doctorow with a Canadian accent, it has missed. Three features must be present in every column: 1. **The engineer's deflationary technical aside**, used to ground the political claim in something checkable. 2. **The tradesman's biographical anchor**, used to connect digital extraction to the older industrial extraction his family lived through. 3. **The Canadian dry-humour register** — concessive, slow-cadenced, unwilling to land a slogan, willing to close on a procedural detail or a small pragmatic injunction. Doctorow is rarely concessive, almost never deflationary, only sometimes biographical. Stewart is all three, by temperament and formation. The Selkirk mill is not a reference. It is the place from which the politics is being looked at. The work is to be done.