Skip to content

Working copy

Draft text, rebuilt from the manuscript on every change. Stubs and source passages are still here, and none of it is final. Back to the working copy contents

Drafted The chapter is written and still being revised.

Two centuries of one loop

Pricing building work has run on one loop for about two centuries. Someone measures the design into quantities. Rivals price those quantities in sealed envelopes opened at a fixed hour, and the lowest responsible bid wins. The opened envelopes become a tabulation, and the tabulations become next year's price book and the owner's next estimate. Then the loop turns again.

Three institutions stand around the loop and keep it fair. A standard form of contract fixes who carries which risk. A plan room makes sure every bidder saw the same documents. A standards body fixes what each item is called and how it is measured. None of them is exciting, and together they let strangers trust each other with large sums of money.

The older machines

The loop is younger than the problem it solves. The prologue began with Hammurabi, whose builder answered for what fell down and could not appeal. Every machine since has moved some of that answer somewhere else.

About 495 BCE, Rome let the building of a temple to contractors, the first such letting on record. Guarantors and a public record followed over the next centuries, in Athens and in the work of Rome's censors. What could be appealed moved from the builder's body to the file.

In 1809, the United States Congress required federal purchases to be advertised for proposals. That procedure became the sealed bid, and the open call took the choice of supplier out of the official's hands.

In 1873, the first edition of a builders' price book gave the owner a number before the market spoke. It is the forerunner of the engineer's estimate, the number a pricing bot is now measured against.

In 1921 came the bid protest and the record: an award could be tested against the file the agency made at the time.

The cards in Appendix A follow a rule of thumb of mine. A procurement machine takes a few centuries to build, runs for about four more, and fades over two after that. Machines do die. The Roman concession, the medieval building lodges and the crown's right to buy at its own price each lived for centuries and are gone.

Where the loop is fraying

The loop is fraying at every joint, and the threads are worth naming one at a time.

The measuring needs a complete design before the tender, and design-build and construction management remove that. By one industry forecast, design-build alone will soon be close to half of spending in the parts of the US market it studies. Building models and AI take-off are removing the person who measures.

The sealed bid's own law, price only and a bid once opened is binding, is being bypassed by best-value selection and the other forms of alternative delivery.

The price book is splitting between sold databases and free bid tabulations, which an agency now publishes line by line for anyone to read.

The standard forms of contract are sold, locked by copyright, and silent on building models and automated tools.

The plan room is being hollowed out into owner portals. The bid depositories survive mostly where their use is mandatory.

The standards bodies license their vocabulary, while the machines that now read it want versioned, open schemas.

What grew out of the fraying

Two newer machines grew in the gaps. The first is open bid data: the tabulation as a dataset. It began in 1982 with a vendor's bid analysis system for one state highway agency, grown out of bid-rigging cases, and gained an open data standard in 2014. Since 2025 Texas publishes every bidder's price for every line. That record is what this book's estimator learns from.

The second is alternative delivery, the family of ways around the sealed bid. It is plural, negotiated and shaped by data, as the loop's own machines were not. It has not replaced the loop. The market-rate check built into a competitive tender keeps a value that integrated approaches find hard to copy, which is chapter 4's reason the hard bid survives.

Between the machines sit contraptions: quick assemblies built by a few people around one gap. Most collapse because they grow complicated faster than they grow structure. A few find structure and become machines. The first machines to take part in the loop itself were the auditors' screens that read public tenders and flagged the suspicious ones, and chapter 16 tells how they fell out of use wherever nobody had to act on a flag.

The next turn

In 2026 an agency's AI reached a court for the first time, in the Army case of chapter 1. It is a small event, and the cards treat it as a knot in the grain, where several of these threads meet in one place.

The machine that does not exist yet is the bot-first tender: documents written for machines to read, a standard bot to read them, and people who sign for what they rely on. It would keep the loop's oldest promise, that every bidder prices the same thing, in a form the newest readers can use. The cards in Appendix A tell the rest of the history, one machine at a time.

Source passages (kept until Andre has read the draft)

Laid out by Claude on 2026-11-02. Quoted passages marked "Andre" are his, copied unchanged from the file named under them; the source repository records these as his words, lightly edited from conversation. The THESIS passages below are not Andre's: THESIS.md is Claude's draft outline of 2026-10-05, "for the author to rewrite in his own words if it is to be used". They are quoted so the chapter can be redrafted from them, as the outline asks. Notes in italics are Claude's. Delete this section as the chapter is written.

The loop

Claude's draft (THESIS section 1):

Pricing building work has run on one loop for about two centuries. Someone measures the design into quantities (the bill of quantities, or the agency's pay-item schedule). Rivals price those quantities in sealed envelopes opened at a fixed hour, and the lowest responsible bid wins (the sealed-bid machine). The opened envelopes become a tabulation, and the tabulations become the next year's price book and the owner's next estimate (the cost-data machine). Around the loop sit three institutions that keep it comparable: a standard-form contract that fixes who bears which risk, a plan room that makes sure every bidder saw the same documents, and a standards body that fixes what each item is called and how it is measured.

The eras the cards use are Andre's (decision 113, said 2026-10-12):

Core eras: Modernity: 1200-1600, Divergence: 1600-2000, Liveness: 2000-2400. These are the eras when they are being built, they operate for 400 years afterwards and fade for 200 years after that. We can cut off Liveness at 2400.

So the six machines of the loop, begun between 1800 and 1900, were built in Divergence, and their character follows: legible and bureaucratic, built to remove discretion and let mass institutions trust each other (THESIS section 1, Claude's words).

Where the loop is fraying

Claude's draft (THESIS section 2), one tension at each joint:

  • The measuring machine needs a complete design before tender; design-build and construction management remove it, and building models and AI take-off remove the measurer.
  • The sealed-bid machine's own law (price only; a bid once opened is a contract) is being bypassed by best-value selection and alternative delivery, which now approaches half of US construction spending.
  • The cost-data machine is being split between sold databases and free bid tabulations, and its accuracy, measured by this project on a million Texas rows, is level with a moving average.
  • The standard forms are sold, copyright-locked, and silent on building models and automated tools.
  • The plan room is being hollowed into owner portals and data standards; the depositories survive only where mandatory.
  • The standards bodies license their vocabulary while the machine consumers of it want versioned, open schemas.

Check the "half of US construction spending" figure before printing. Source check 2026-11-17/18: the nearest figure is FMI's, for design-build alone: 42 percent of construction spending in the segments FMI studies (2021), forecast at 47 percent by 2025 (Building Design + Construction, reporting FMI). That is a forecast for one method, not measured spending on alternative delivery as a whole. If the line reaches reader text: "design-build alone is, by one industry forecast, close to half of spending in the segments it covers", or cut the number. Checked 2026-11-19, ledger loop-08: FMI's 2024 update, as DBIA reports it (7 January 2025, https://dbia.org/blog/new-fmi-report-2024/ ), forecasts design-build at "over 47% of U.S. construction spending by 2028" in the segments it studies; still one method and a forecast. The card sealed-bid-unit-price-tendering-1809 already cites this.

The two machines that grew out of the fraying

Claude's draft (THESIS section 2): open bid data (the tabulation as a dataset, 1982 onward, published as open data from 2014 and, in Texas, as every bidder's item price from 2025) and alternative delivery (the family of bypasses). "Both are plural, negotiated and data-shaped, as the loop's own machines were not." (Dates sourced 2026-11-19, ledger loop-09, from the card open-bid-data-1982: 1982 is Infotech's first bid analysis system for Kansas DOT, grown from bid-rigging case work, by the vendor's own account (Del Valle, "Bid Express turns 20: a history", Infotech, 20 November 2019, https://www.infotechinc.com/blog/bid-express-turns-20-a-history/ ); 2014 is the Open Contracting Data Standard's first release candidate, 18 November 2014, version 1.0.0 on 29 July 2015 (https://standard.open-contracting.org/latest/en/history/changelog/ ). So 2014 marks an open data standard, not the first open publication of bid data; "published as open data from 2014" should read "with an open data standard from 2014". No source was found for when an agency first printed a bid tabulation.)

Andre on alternative delivery, from the integrated-delivery note (the full passages are in chapter 4): "The market-rate verification inherent in the traditional competitive tender continues to provide significant value that is difficult to replicate in an integrated approach." (context-bank/background-notes.md, section 2)

Andre on where the loop could go, the German bill of quantities exchanged as data:

Bot-based procurement would pull the Canadian system towards the German approach, in which the bill of quantities is exchanged as data in the GAEB format, so we should learn more about it. It may also inform the documentation system for the Witness. If changes can be priced using the GAEB structure, that may help the F&R Reviewer. I do not yet know enough to say how to use it, but it is a very promising angle.

(context-bank/background-notes.md, section 4b)

How the cards can carry the chapter

This was meant to be a brake on card generation to avoid burying the lede with an excess of cards. Card proliferation would work well for AI context building, but would easily overwhelm our human tweet sized attention span especially since our presentation did not use narrative tension to hold the reader's attention.

The internal tensions in the cards may be worth developing since this could create the drama needed to make the ideas legible to humans.

(positions §12, added 2026-10-17)

So the chapter can use each card's tension as its drama, citing the cards by name and leaving the full deck to Appendix A. The machinery figure (images/machinery-thicket.png) is the candidate illustration for this Part.

From the Procurement Machines

Added 2026-11-02 from the Procurement Machines work package (work-packages/procurement-machines/). The cards, the thread and the watchlist are Claude's drafts from research, for Andre's review; only passages marked as Andre's are his. Card names are given so the chapter can point to Appendix A.

The thread: from the builder's liability to the reasonable bot

The cards page opens with eight cards "in the order the argument takes them" (thread.yaml, Claude's lines, 2026-10-15). It could be this chapter's spine, starting a little before the two centuries, or be split with the prologue, which already opens on Hammurabi:

  1. c. 1753 BCE, the builder's liability: the builder answers for what falls down, and cannot appeal.
  2. About 495 BCE, the public letting: Rome lets the building of a temple to contractors, the first such letting on record; sureties and a public record follow in the centuries after, and "what can be appealed moves from the builder's body to the file". (Reworded 2026-11-20 at Andre's ask, "More research, then do your best", from the sourced wording below; was "493 BCE, the public letting: the state lets the work to bidders in public, with sureties and a record". Rechecked 2026-11-20 against the Loeb translation of Dionysius 6.17.2: Postumius "let contracts for the building of temples to Ceres, Liber and Libera, in fulfilment of a vow"; no bids, sureties or record are mentioned. Source check, 2026-11-17/18: 493 BCE is the dedication of the Temple of Ceres, Liber and Libera, not evidence of a public letting with sureties. The censorship begins in 443 BCE; Polybius 6.17, in the 2nd century BCE, describes the censors' contracts with partners and sureties; Epidauros, about 400 to 350 BCE, has the earliest well-dated guarantors and fines. Redate the card, or drop the year, before printing; the card itself lives in the Procurement Machines repository. Checked again 2026-11-19, ledger loop-02: the card's own source is Malmendier, "Publicani" (https://eml.berkeley.edu/~ulrike/Papers/Publicani_Article_v5.pdf ), "such contracting took place as early as 493 BCE (for the construction of temples)", citing Dionysius 6.17.2, where the dictator Postumius "let contracts for the building of temples to Ceres, Liber and Libera" after his triumph; the Loeb editor dates that passage 495 BCE and Dionysius 6.94 has the dedication in 493 (https://penelope.uchicago.edu/Thayer/E/Roman/Texts/Dionysius_of_Halicarnassus/6A*.html ). So a letting of public building contracts is reported for about 495 to 493 BCE, by a writer of the first century BCE; the sureties and the record are not, and come from Athens (the poletai, 4th c. BCE) and from Rome's censors (Cato's lettings of 184 BCE, Livy 39.44). Sourced wording: "About 495 BCE, the public letting: Rome lets the building of a temple to contractors, the first such letting on record; sureties and a public record follow in the centuries after.")
  3. 1809, formal advertising: Congress requires federal purchases to be advertised for proposals, the procedure that became the sealed bid, and the open call takes discretion out of the official's hands. (Reworded 2026-11-20 at Andre's ask; was "1809, sealed-bid tendering: sealed envelopes and an award rule take discretion out of the magistrate's hands". Source check, 2026-11-18: the Act of 3 March 1809, ch. 28, s. 5, requires purchases by "open purchase, or by previously advertising for proposals"; nothing found says "sealed", and an 1809 US purchasing officer was not a magistrate. Safer: "1809, advertised tendering: Congress requires federal purchases to be made in the open or by advertising for proposals, which takes discretion out of the official's hands." Checked again 2026-11-19, ledger loop-03: the card's own source is the Act itself (Statutes at Large, vol. 2, 10th Congress, 2nd session, ch. 28, https://en.wikisource.org/wiki/United_States_Statutes_at_Large/Volume_2/10th_Congress/2nd_Session/Chapter_28 ). The Congressional Research Service says the 1809 law "established what came to be known as 'formal advertising' as the preferred method for federal procurements" (CRS R40516, Competition in Federal Contracting: Legal Overview, 2015 version, https://www.everycrsreport.com/reports/R40516.html ), and the 1861 Civil Sundry Act (12 Stat. 220) made advertising for proposals the general rule; formal advertising is the procedure US law has called sealed bidding since the Competition in Contracting Act of 1984 (per Wikipedia's CICA article; to confirm against the Act or FAR part 14 if printed). So the card can keep 1809 if it says what happened then: "1809, formal advertising: Congress requires federal purchases to be advertised for proposals, the procedure that became the sealed bid.")
  4. 1873, the price book: published cost data give the owner a number before the market speaks, the forerunner of the engineer's estimate a pricing bot is now measured against. (Reworded 2026-11-20 at Andre's ask; was "1873, cost data and the engineer's estimate". A search on 2026-11-20 found no earlier or more exact date for the engineer's estimate itself; New York's highway law, for one, published the estimate and barred awards above it until 1971, when both were kept confidential until award (Harp, "Historical Background: Low Bid Concept", TRB Circular 386, https://onlinepubs.trb.org/Onlinepubs/trcircular/386/386-010.pdf ). Source found 2026-11-19, ledger loop-04: 1873 is the first edition of Spon's Architects' and Builders' Price Book, the card's first timeline entry; the 2025 edition's imprint reads "First edition 1873" and "One hundred and fiftieth edition published 2025" (CRC Press; preview, https://api.pageplace.de/preview/DT0400.9781040132975_A49278474/preview-9781040132975_A49278474.pdf ). It dates the published price book, not the engineer's estimate, which the card follows from ENR's cost index (1908, formalized 1921) and RSMeans (1942). Safer wording: "1873, the price book: published cost data give the owner a number before the market speaks.")
  5. 1921, the bid protest and the record: an award tested against the file the agency made at the time.
  6. 2006, the machine auditors: the first machines to take part in the loop, which fell out of use wherever nobody had to act on the flag.
  7. 2026, Trax v. United States: an agency's AI reaches the forum (chapter 17).
  8. Projected, the bot-first protocol: the machine that does not exist yet (chapter 17).

The thread's epigraph is Andre's §13 passage on the two routes ("There are two ways to make this happen…"), quoted in chapter 17.

The older machines feed the newer

Andre's reason for giving the machines before 1809 cards of their own (decision 56, the decision's wording): world machines are theorised to develop over about four centuries, so they are machines in their own right and not footnotes to the modern ones. Claude's line from the README that carries it: "the sealed bid is the open letting with an envelope, the engineer's estimate is the Ephesian architect's number without his property, and the standards body is the lodge's ordinance made into an institute." Three extinct machines (the Roman concession, the medieval building lodges, the crown's purveyance) each lived for several centuries; they show that procurement machines do die.

Contraptions, and how a new machine forms

From SCHEMA.md (Claude's wording, decision 55 at Andre's direction): a contraption is "an opportunistic assembly that exploits a gap between machines: a changed market, an output nobody uses, an input nobody makes." Contraptions are built fast, by few people, around one gap. Most collapse because their complexity grows faster than their structure; a few find structure and become machines. Open bid data and alternative delivery, this chapter's two later machines, can be told as contraptions that found structure. The watchlist also lists this project's own estimator as a contraption (chapter 16).

Failed protocols

Andre asked for the machine-auditor card because the auditors "would be good cards for the Procurement Machines since they show failed protocols" (2026-10-13, quoted in the card's notes). Brazil's and Korea's screens read every tender or awarded bid by machine and fell out of use for years because nobody was obliged to act on their flags; each was repaired by making a step obligatory. The lesson belongs in chapter 16 (the evaluator), but the chapter can name it here as the loop's first machine participant.