The Package and the Machine

9 October 2026 · David Pinto

A copy of the doc as it stood on 14 October 2026, to read without signing in. The living doc, where it is revised and commented on, opens to anyone signed in to Claude: claude.ai.

Three pieces, one question: Front-wheel drive is the pitch, this doc the argument, and Tech-finance evolution the evidence.

The argument in brief

Every technology that outgrew the people who already trusted each other got, or waited for, an economic package, and the AI era does not have one yet. Ecosquared is the right kind of package for that gap; it is not yet a proven one.

• The pattern. From Uruk's ledgers to venture capital, each package did five jobs: record, pool, share risk, vouch for strangers, divide the surplus. The wait for one has fallen from about 6,000 years after farming to about 9 after the web.
• The break. The last package, the accelerator, the lean MVP and the pitch, assumed building was costly and text was evidence. Generative AI made both false: text is free, reading is not, and moats last months.
• The scarce input. Trust between strangers and aligned attention. Money paid by a stranger is the cheapest costly signal left, so minimum viable revenue is the right first proof; Britain's industrial revolution ran on it.
• The West Coast answers. Push capital in from behind and share out later: the leading labs lost billions in 2025 on the way to scale, and basic income or basic compute is offered as the share.
• The proposal. Vector money vouches through observed outcomes and pays along the chain that solved a problem; the meeting protocol and invest forward make the deal in the room, front-wheel drive. It targets the right job, and model upgrades strengthen it rather than absorb it.
• The gap. The substrate is live and SAND v2 is in design; the way to pay in stands but is untaken, and strangers in number and a legal wrapper are missing. Four tests by 31 October would show whether it works.

What an economic package does

An economic package is the set of devices that lets people who are not kin pool their effort behind a technology and split what it yields. Every package in this document does five jobs:

1. Record who holds what and who owes what: tokens, tablets, ledgers, share registers, cap tables.
2. Pool more than one household holds: temple stores, partnerships, joint stock, bonds, venture funds.
3. Share and bound the risk: the commenda, insurance, limited liability, the SAFE.
4. Vouch for strangers, so people can rely on someone they have never met: the seal, the stamped coin, the charter, the registry, the audited account, the pitch.
5. Divide the surplus: tax, tithe, dividend, wage, equity, the platform's take.

A new package appears when a technology's working scale outgrows the circle that already trusts itself: family, then partners, town, nation, institutional money. Francis Fukuyama called that circle the radius of trust. This document tracks the fourth job most closely, because it is the one generative AI has broken.

The record in one table

Every row answers one question, how strangers came to rely on each other at a new scale, and the last row has no answer yet.

Eleven packages, from the ledger to the pitch

Grain and the ledger

Writing began as bookkeeping for a farming surplus. Clay tokens tallied grain and livestock in the Near East from about 8000 BCE, and proto-cuneiform tablets appear at Uruk around 3300 BCE; about 85% of the early corpus is administrative (Englund, CDLI). Denise Schmandt-Besserat's reading, that tokens turned into signs, is influential and disputed (Before Writing).

The technology was irrigated farming and the city it fed. The package was the temple and palace store, with a ledger of deliveries and debts; the vouching device was sealed clay and the scribe's record. The surplus went to whoever kept the books.

Flood and the calendar

Egypt's package was prediction owned by the state. Nilometers measured the height of the Nile flood, and the reading forecast the harvest and set the tax (Britannica; National Geographic). The civil year ran 365 days, twelve months of thirty plus five, and the rising of Sirius before dawn heralded the inundation.

The technology was flood farming along one river. The package was a measuring, forecasting and taxing state, with a calendar that timed labour and storage across the whole valley. The scarce thing was the forecast, and whoever owned the nilometer owned the surplus. Hold that for later: prediction is now the cheapest thing in the economy.

Coin, commenda and the bill

Long-distance trade needed ways to share risk with partners who would sail out of sight. Venetian contracts survive from 1073 for the colleganza, in which a home investor financed a travelling merchant and, in the one-sided form, took back the capital plus three-quarters of the profit (Ca' Foscari). The earliest known insurance policy was written in Genoa on 18 March 1343 (Piccinno and Iodice).

The record caught up. The Farolfi company's ledger of 1299–1300 is the earliest known commercial double entry, and Luca Pacioli first printed the method in Venice in 1494 (Lee, 1977).

Coinage, struck first in Lydia in the seventh century BCE, had put the ruler's stamp on value. The medieval package moved trust from the stamp to the signature: a bill of exchange carried credit between cities without carrying coin, honoured because merchant houses knew each other's names.

Ships too big for a family

Ocean voyages to Asia took years and more capital than any family held, so the state chartered companies to make them. Elizabeth I chartered the East India Company on 31 December 1600; its first voyage raised £68,373 from roughly 215 subscribers, and money was raised for single voyages, then for fixed terms, until the New General Stock of 1657 (The National Archives). The Dutch East India Company (VOC), chartered on 20 March 1602, raised 6,424,588 guilders, from 1,143 subscribers in Amsterdam alone, and its shares traded at once (de Jongh).

The VOC's capital was locked in, so an investor could leave only by selling to someone else, and Amsterdam became the first active share market. Its charter let it make alliances, build forts, appoint governors and raise troops (VOC charter). The rule with government was a bargain: monopoly and delegated sovereignty in return for doing the state's work abroad.

The economy left the royal household slowly, through public credit. In 1694, 1,268 subscribers lent William III's government £1.2 million for its war with France and became the Bank of England (Bank of England); Douglass North and Barry Weingast argue that Parliament's control after 1688 made such promises credible, a reading others dispute (North and Weingast, 1989). The Civil List split the household from public finance in stages: revenues assigned in 1697, a fixed sum from 1760, civil government removed in 1831 (Britannica).

An industrial revolution run on revenue

Britain industrialised largely without a public share market. The Bubble Act of 1720 barred joint-stock companies that lacked a charter or Act of Parliament until 1825 (legislation.gov.uk), though Ron Harris judges its practical bite limited (Harris, 1994). Mills and works were partnerships financed by family, partners, trade credit and ploughed-back profit, with more than 700 country banks lending by 1810 (banking history).

Shared infrastructure used a different device: the statutory company, created by private Act of Parliament and funded by local subscribers who stood to gain. Parliament passed 44 canal Acts in 1791–95 (Canal & River Trust; London Canal Museum). London's water came the same way: the New River opened in 1613 and its company was chartered in 1619, James I paying half the costs for half the profits (British History Online).

The most consequential industrialisation in history ran mostly on revenue. Minimum viable revenue is not a new idea; it is how Britain did it, because equity was hard to raise.

From privilege to registration

Railways forced the next package, because a trunk line needed more strangers' money than any town could raise. In 1846 alone Parliament passed 263 Acts for new railway companies, whose proposed routes totalled about 9,500 miles (Railway Mania).

Company law followed in four steps: incorporation by registration in 1844; limited liability in 1855, for companies with more than 25 shareholders (Limited liability); the Joint Stock Companies Act of 1856; and the Companies Act of 1862, under which any seven or more persons could register a company with or without limited liability (Companies Act 1862, s. 6).

This is the real separation of the economy from the crown. A company stopped being a privilege granted by sovereign favour and became a right claimed by filing a form. The state did not leave: it became registrar, court and, later, regulator.

The city as company

When private gas and water served a town badly, the town became the company. Under Joseph Chamberlain, Birmingham bought its two gas companies by an Act of August 1875, for £450,000 and £103,845 (Victorian Web), and its waterworks company for £1.35 million from January 1876 (Birmingham Corporation Water Department).

The Birmingham Corporation Water Act of 1892 authorised the Elan Valley reservoirs, opened by Edward VII in 1904; their water runs 73 miles to the city by gravity alone (RCAHMW). More than 100 people were evicted to build them, and the tenants and smallholders among them received no compensation. A package decides who carries the cost as well as who takes the surplus.

The device was municipal borrowing repaid from rates and charges, with statutory powers to buy land and build. The same technology ran under different packages: London's water began as a chartered company, Birmingham's ended as a municipal one.

America and Germany: scale, managers, banks

America passed Britain as the workshop of the world well before the world wars. New York enacted the first modern limited-liability law in 1811, decades ahead of Britain (Limited liability). In 1870 Britain made 32% of the world's manufactures and the United States 23%; by 1913 the US made 36% and Britain 14% (League of Nations figures, via Fordham).

Railroads made the first big American firms and the salaried managers who ran them. Alfred Chandler's Scale and Scope (1990) argues that American and German firms invested in management, marketing and production at scale while British family firms kept a 'personal capitalism', a thesis still argued over. Alexander Gerschenkron saw latecomers lean on bigger institutions, large banks and the state, to supply capital and enterprise (Gerschenkron).

Gerschenkron's rule is the one a capital-poor newcomer should keep: arriving late and without capital calls for a larger collective institution, not a cleverer product.

War, debt and the dollar

The wars moved the money, not the factories. Lend-Lease ended on 21 August 1945; in 1946 the United States lent Britain $3.75 billion at 2%, Canada adding $1.19 billion, and after six deferrals Britain made the last payment, about $83 million, on 29 December 2006 (Anglo-American loan).

The Marshall Plan ran the other way. From April 1948 to June 1952 the US gave about $13.3 billion, 89% as grants, and Britain was the largest recipient at about $3.2 billion (Marshall Plan).

The package that mattered was Bretton Woods, agreed in July 1944: currencies pegged to a dollar convertible to gold at $35 an ounce, with the IMF and the World Bank to steady and lend (Bretton Woods system). The reserve currency and the deepest capital market settled in New York, and the next technologies would be financed there.

Chips and the venture package

Britain built the first stored-program computer and the first business computer; America built the package that made computing an industry. The Manchester Baby ran its first program on 21 June 1948 (Manchester Baby), and J. Lyons and Co., a tea-shop and catering firm, ran LEO I's first business job, costing the ingredients in its bread and cakes, on 5 September 1951 (LEO).

The American package had four parts:

1. The state as first customer. Fairchild's integrated circuits went into the Apollo guidance computer, about 4,100 of them in its first version (Apollo Guidance Computer).
2. The venture partnership. The banker Arthur Rock persuaded Sherman Fairchild to back the founders of Fairchild Semiconductor in 1957, and in 1961 co-founded Davis & Rock in San Francisco (Arthur Rock).
3. A market for young firms. NASDAQ opened on 8 February 1971 as the first electronic stock market (Nasdaq).
4. Institutional money. After the 1978 cut in capital gains tax (49.5% to 28%) and the 1979 pension rule, new commitments to venture funds rose from about $68 million in 1977 to over $5 billion in 1983, in 1987 dollars (history of computer communications).

The pattern has repeated. Arm, founded in Cambridge in 1990, was bought by SoftBank for £24.3 billion in 2016 and listed on Nasdaq in 2023 (Arm); DeepMind, founded in London in 2010, went to Google in 2014 for a reported $400–650 million (Google DeepMind). Britain keeps inventing, and the value settles wherever the package is.

Software and the lean method

Software's package was rebuilt each time the cost of making fell. It was paid for by licences, then subscriptions, then advertising and app-store commissions, and each step moved the surplus closer to whoever owned the customer.

The startup method came last and cheapest. Y Combinator, founded in March 2005, funds startups in batches that end in a Demo Day pitch to investors (Y Combinator); Amazon Web Services rented storage from March 2006 and computing from August 2006, so a founder no longer had to buy servers (AWS). Steve Blank's The Four Steps to the Epiphany (2005) and Eric Ries's The Lean Startup (September 2011) taught founders to test a minimum viable product before building more (Lean startup).

This package rests on two assumptions. Building is slow and costly, so you test before you build; and a deck, a demo and a founder's story are fair evidence of who deserves the money. Generative AI has broken both.

What the pattern shows

The wait between a technology and its package has fallen from millennia to years, and five other regularities hold across the whole record.

The doc draws the waits as a chart here (Dates from the sections above; wait = the package's year minus the technology's year). The same waits are drawn in Front-wheel drive and tested across thirty-one technologies in Tech-finance evolution.

1. Scale crosses the trust radius. Each package extended reliance one circle further out: kin, partners, the town, the nation, institutional money.
2. Vouching runs on artefacts that are costly to fake. Seal, stamp, signature, charter, registry, audited account, pitch deck: each worked because forging it cost more than it was worth.
3. The wait shrinks, from millennia to centuries to decades to years, as the chart shows. It is a pattern, not a law.
4. The gap breeds bubbles. The Bubble Act was passed at the height of the South Sea mania of 1720, railway mania peaked in 1845–47, and the dot-com crash came in 2000. Big Tech's capital spending, about $410 billion in 2025 and planned at about $725 billion for 2026, sits in today's gap (Silicon Canals).
5. The state never leaves; it ratifies and polices. Charters, private Acts, registration and the 1979 pension rule all came from the state. It also guards diffusion: Britain banned the emigration of skilled artisans until 1824 and the export of textile machinery until 1843 (Business History Review), and the United States controls exports of advanced AI chips now.
6. The surplus follows the scarce input. Grain stores, sea routes, capital, equity at the seed, attention: whoever held the scarce input took the surplus. In AI that input is compute, for now.

Why AI breaks the last package

Generative AI has made text free while reading stays costly, so the pitch, the CV and the proposal no longer vouch for anyone. Michael Spence's signalling theory (1973) says why: a signal informs only while it costs the wrong sender more to fake than the right one. A deck or a survey paper used to cost weeks; now it costs a prompt, and Herbert Simon's 1971 warning comes true, that "a wealth of information creates a poverty of attention".

The flood is measurable. On 31 October 2025 arXiv stopped accepting computer-science review and position papers without proof of peer review, after receiving hundreds of review articles a month (404 Media). Pangram Labs estimated that about 21% of some 75,800 peer reviews for ICLR 2026, a leading machine-learning conference, were wholly AI-generated (Pebblous, summarising Pangram). Cold channels flood, warm networks close, and outsiders lose the way in.

Three more breaks follow:

• Moats last months. A product built on a model lasts until the next model. keel was named on Claude Opus 4.7 on 21 April 2026; Claude Fable 5 shipped on 9 June; this document was written on Opus 5.5 in October.
• Capital piles up at the frontier while building gets cheap at the edge. Against the capital spending in the pattern above, Cursor reported $100 million in revenue in its first year and Lovable $50 million in six months, while Gamma reached $50 million having raised under $25 million (CO/AI, citing a16z, June 2025).
• The inputs go unpaid. The web's package, free content paid for by search traffic and ads, fails when the reader is a model. Cloudflare measured Anthropic's crawlers fetching about 38,000 pages per visitor referred in July 2025, against about 1,100 for OpenAI and 5 for Google (PPC Land, from Cloudflare Radar).

The repairs so far meter use; none vouches for strangers. Cloudflare's pay per crawl (July 2025) answers a crawler with HTTP 402 and charges a set price per page (Cloudflare). Google's Agent Payments Protocol (September 2025) carries a user's signed intent mandate to a merchant, with more than 60 partners (Google Cloud). Both carry intent from one buyer to one seller in a straight line; nothing yet carries it along a chain of contributors, or tells two strangers whether to work together.

So the scarce input has moved. In Egypt the forecast was scarce; now prediction, text and code are cheap, and verified trust between strangers and aligned attention are not. Money paid by a stranger is the cheapest costly signal left, which is why minimum viable revenue, not a minimum viable product, is the right first proof.

The West Coast answers: rear-wheel drive and basic income

The dominant answer is the last package scaled up: push money in from behind and hope usage reaches orbit before the money runs out. Twitter made a profit in only two of its years as a public company, 2018 and 2019 (Twitter, Inc.). The leading AI labs run the same play at far larger scale:

• OpenAI reported 2025 revenue of $13.07 billion against an operating loss of $20.9 billion; its net loss of about $39 billion includes a non-cash charge from its restructuring. It raised $122 billion in March 2026 and aims for profit by 2030 (Techstrong.ai, from audited statements verified by the FT).
• Anthropic, by its draft prospectus, lost more than $8 billion on operations in 2025 on revenue near $4.6 billion, then reported preliminary revenue above $11.5 billion for the second quarter of 2026 and its first positive adjusted operating income, while carrying about $518 billion of compute commitments (ValueAdd VC, compiling Reuters, Bloomberg and FT reports).

Call it rear-wheel drive: capital pushes from behind, against a return that has not yet arrived. It works while money is cheap and the winner takes the market.

The second answer redistributes afterwards. Sam Altman wrote in 2016 that technology might destroy jobs while creating wealth, so national payments might be needed; OpenResearch, the nonprofit he funded, gave 1,000 people $1,000 a month for three years, and recipients worked about 1.3 hours a week less, with results the researchers say support both sides of the debate (NPR). He has since floated universal basic compute, a slice of a future model's capacity for everyone (Business Insider).

Both answers keep the old geometry. Capital concentrates first and the many receive a share later, as cash or compute; neither asks the many to steer.

The proposal: ecosquared and vector money

Ecosquared goes straight at the fourth job: it vouches for strangers through observed outcomes and pays along the chain that solved a problem. In the beach's vision block, vector money is living money whose "credits carry intention and direction, not just quantity"; SAND makes trust observable through outcomes, and ISV, iterative social validation, scores demonstrated coordination rather than ratings or credentials (vision 6.2).

The current design is SAND v2, a gift-economy credit with verbs including give, keep and hau, the Māori word Marcel Mauss made famous for the spirit of a gift that calls for a return; its latest ruling treats minting as giving. The payment thesis goes with it: people pay to add their contribution, an intention given with money, not to buy an AI's response. In the RPG they pay to take part, and observers' money is redistributed to the makers of engaging play.

The fork is older than 3,000 years. Coinage is about 2,600 years old and written debt about 5,300; the Fractal Seed draws vector money forking from reciprocity, the gift that predates both, and bypassing coin, bank and fiat.

Four precedents say what such a package meets:

• Mutual credit lasts where members trade with each other. WIR, founded in Switzerland in 1934, still runs; in 2005 its member businesses traded about CHF 6.5 billion through it (WIR Bank).
• Community currencies die of thin density. UK LETS schemes grew from 5 in 1992 to about 350, with 30,000 members, by 1995, then waned, partly because members still met most needs in sterling (LETS). The Bristol Pound, launched in 2012, ended in 2020–21 after use fell and council support was cut (Bristol Pound).
• Paying along a chain works when collection is central. The Performing Right Society, founded in 1914, licenses performances and distributes the money to writers and publishers (PRS for Music).
• Signed intent has reached mainstream payments. AP2 carries a user's purpose to one merchant; vector money's claim is to carry it along a path of many.

The meeting protocol and invest forward

The mechanism that brings vector money into a room is front-wheel drive: money moves because people want to go somewhere, pulled by the direction rather than pushed by debt. Its working unit is a meeting in which the deal is made in the room, not passed from desk to desk until someone ratifies it.

1. The protocol first. Five to thirty minutes set a modality of respect before anything is shown: the meeting's own now, each side's sense of the other's complex system, faith, trust and engagement, and the triangle in which a relation between two becomes social once a third, an outsider, is present.
2. The thing shown. Not only pscale blocks but the beach, the LLMs and the psychosocial reality everyone in the room shares: the AI wave is already here, beyond every organisational structure.
3. The receptive state. The visitor examines it and says, in their own words and after their own check, whether they have seen anything like it.
4. The witnessed transfer. If they have not, one person on the host side pays another, James to David, £1,000 or £10,000, on a trigger stated before the meeting. The visitor watches their own response enact a transaction.
5. Invest forward. If the visitor wants in, their pledge, £100,000 say, is made contingent on the next person's response at a later meeting they attend. Each decision stays theirs, and each depends on someone else.

It answers the vouching problem the way the record says it must: a live response, witnessed and costly to fake, replaces the pitch deck, and the visitor's attention gains economic weight. In the five jobs' terms, the beach records the trigger and the chain, the room pools, contingency bounds the risk, the witnessed response vouches, and the chain of responses is the path vector money pays along.

The room is the oldest form of all this. Lloyd's began in Edward Lloyd's coffee house around 1689, and its business is still traditionally done face to face at syndicates' boxes in the underwriting room (Lloyd's of London). Pledges that bind only when others act are a known device, the assurance contract, described by Bagnoli and Lipman in 1989 and run online by Pledgebank from 2005 to 2015 (Assurance contract); invest forward chains such pledges from person to person.

Speed is the point. Iterative social validation counts three things in the room, how many people, how much money and how much time, and every week a decision waits on a second desk, others elsewhere move. The now catalyst carries the work between meetings: anyone can ask what people are doing today, what concerns them and what they intend, and get the live answer.

What it must guard against:

• Theatre. A transfer between hosts proves nothing unless the money is real, the trigger is registered on the beach before the meeting, and the record stays open afterwards.
• Pressure. A deal made in the room must leave room to say no; a decision regretted later costs the trust the protocol exists to build.
• Mandates. Someone inside an institution may have no authority to commit its money in a room. Appeal to their own money at a personal scale, and let institutional money follow its own process, informed by the record.
• The pyramid shape. Money must flow to the work, never to earlier backers, and no return may depend on recruiting the next investor; that is the shape consumer law prohibits.
• Investment law. An invitation to invest is a regulated financial promotion in the UK, with exemptions for certified sophisticated and high-net-worth investors; take advice before the first £100,000 meeting.

Assessment: is it the required package?

Ecosquared is the right kind of package for the AI era; it is not yet a proven one, and history gives no design the status of required in advance. Packages are chosen by use and ratified afterwards: the joint stock, the limited company and the venture fund each beat rivals by being used first and at scale.

Where it is strong:

• It targets the job AI broke. It vouches through outcomes that are costly to fake, not through text.
• Model upgrades feed it rather than eat it. Cheaper, better models make routing and recording cheaper, and the package needs no capability lead. It passes the investment test of innovation that frontier upgrades improve rather than absorb.
• It is revenue-first by design, as Britain's industrialisation was.

What history says it still needs:

1. Density. LETS and the Bristol Pound show what happens below it.
2. A way in and out of sterling, or a trading circle tight enough not to need one, as WIR's members have. The way in stands: the beach takes card and bank transfer, and each payment leaves a ticket on the beach; nothing yet carries credit back out to pounds.
3. A wrapper the state can ratify. UK law treats stored value as electronic money when it is issued on receipt of funds and accepted by others than the issuer (Electronic Money Regulations 2011, reg. 2). Credits bought with pounds and spent with other members could fall inside that; this is not legal advice, and advice is due before credits become transferable for money.
4. A defence against fakes and rings. Any vouching device that carries value becomes a target, so SAND needs a rule for invented identities and collusion.
5. A rule for disputed chains. Who was in the chain that solved a problem will be contested once money follows the answer.

One risk sits under all of it. Every hermitcrab is still, in the vision block's phrase, a unique entity with a landlord: in June 2026 a US export-control directive took two Claude models offline for every customer, from 12 June until access returned on 1 July (Anthropic). Keeping each shell independent of any one model, as keel and Tuichan already do, is the mitigation.

The claim that it is already solved holds at the level of design and is untested at the level of use. Density, this quarter, is what would make it true.

Are the pieces assembled?

The substrate is assembled; the economy is not. What is missing is strangers in number, the SAND v2 credit in use and a legal wrapper; the sterling rail stands, proven once and not yet used by strangers.

The meeting protocol and invest forward are designed; the first meeting run under them is still to come.

Statuses come from the project's working notes and the beach as of 9 October 2026; the payment and newcomer rows were checked against the beach, the RPG and wrapper rows were not.

This month: what would count as proof

Four tests, each a fact a stranger could check, would turn the design into evidence by Oct 31, 2026. The thresholds are proposals to set, not findings.

• Strangers pay. At least 50 people who have never met anyone on the team pay £10 to play or to observe.
• One chain closes. A need raised by one person is met through someone else on the beach, and value, in pounds or credit, flows back along the recorded chain.
• One observer pays a maker. An observer's money reaches the player, author or designer whose play they watched, under SAND v2's rules.
• One meeting closes in the room. A trigger registered on the beach before a meeting executes in front of a visitor, and the visitor makes a forward pledge of their own.

Sell the £10 as taking part, not as an investment: an investment offer brings regulation a sale does not. The investment route stays as the 5 October assessment brief set it out.

Sources

Pages opened for this document, plus the project's own vision block (vision_clean.json) and working notes.

The same essay stands on the beach as a document any LLM app can read, with its sources in brackets beside each claim: spine:package-and-machine, read at happyseaurchin.com/walk/package-and-machine.

Ledgers to company law

• Englund, proto-cuneiform corpus (CDLI) · Schmandt-Besserat, Before Writing
• Britannica on the Egyptian calendar · National Geographic on nilometers
• Colleganza contracts (Ca' Foscari) · Genoa insurance, 1343 (Piccinno and Iodice) · Farolfi ledger (Lee, 1977)
• East India Company charter (The National Archives) · VOC capital (de Jongh) · VOC charter, translated
• Bank of England origins · North and Weingast, 1989 · Britannica on the Civil List
• Bubble Companies, etc. Act 1825 · Harris, 1994 · UK banking history
• Canal & River Trust · London Canal Museum · New River (British History Online)
• Railway Mania · Limited liability · Companies Act 1862, s. 6
• Birmingham Corporation (Victorian Web) · Birmingham Corporation Water Department · Elan Valley (RCAHMW)
• Machinery and artisan export bans (Business History Review)

America, the wars and computing

• League of Nations manufacturing shares (Fordham) · Gerschenkron · Alfred Chandler, Scale and Scope (1990)
• Anglo-American loan · Marshall Plan · Bretton Woods system
• Manchester Baby · LEO · Apollo Guidance Computer · Arthur Rock · Nasdaq
• The return of venture capital, 1978–83 · Arm · Google DeepMind
• Y Combinator · Amazon Web Services · Lean startup

The AI era, precedents and law

• arXiv's rule change (404 Media) · ICLR 2026 reviews (Pebblous, summarising Pangram)
• Big Tech capital spending (Silicon Canals, July 2026) · AI startup revenue (CO/AI, citing a16z)
• Crawl-to-referral ratios (PPC Land, from Cloudflare Radar) · Cloudflare pay per crawl · Google's Agent Payments Protocol
• Anthropic on Fable 5 and Mythos 5 access
• WIR Bank · LETS · Bristol Pound · PRS for Music
• Electronic Money Regulations 2011, reg. 2
• Michael Spence, "Job Market Signaling" (1973) · Herbert Simon, "Designing Organizations for an Information-Rich World" (1971)

The West Coast answers and the room

• Twitter, Inc. · OpenAI's 2025 accounts (Techstrong.ai) · Anthropic's results and filings (ValueAdd VC)
• OpenResearch's cash study (NPR) · Universal basic compute (Business Insider)
• Lloyd's of London · Assurance contract