Trail 2026-10-23 (tenth session)

Budget: 15 fetched pages or 40 tool calls. Orientation: steer, index, research-values, dead-ends, trail-2026-10-16, trail-2026-10-19, trail-2026-10-21. Thread chosen: the last session's "worth continuing", telco-copper-retirement-2023-2026 (steer item 1, macro-view resources leg, within thread-transition), not vetoed by James: read one rural telephone co-op's audited statements on the Kentucky PSC site for salvage or gain on retired copper, to test whether the copper windfall reaches members as capital credits. If it dies in three hops, second thread is steer item 2 (who actually runs on surplus hardware), last touched 14 Oct.

Hops

Tooling: web_extract still broken (extract_backend empty); searches come through the free-tier fallback. Kentucky PSC PDFs download with curl and parse with pypdf from scratch/pylib. Each hop says full read or search text; searches count as fetches.

1. Expect: Peoples Rural Telephone Co-op's audited statements for 2021-2024, to see salvage in the years the copper came out. -> 2021, 2023, 2024 read in full (2022 is 404). Result: "Salvage, net of removals" in the cash-flow statement is $75,843 (2020), nil (2021), $21,288 (2022), nil (2023), $37,510 (2024), against plant construction of $3.8M-$9.9M a year. Retired units "including cost of removal, net of any salvage value" are charged to accumulated depreciation, so salvage never passes through margins and so never reaches capital credits. The RUS loan covenants in the same filings require the co-op to "seek approval of the sale, lease, or transfer of capital assets and disposition of proceeds for the sale or lease of plant, material, or scrap". My hunch from 21 Oct (windfall flows to members as capital credits) does not hold for this co-op: the copper fetched almost nothing net of removal, and the proceeds sit under the federal lender's approval.
2. Expect: the RUS rule that puts scrap proceeds under the lender. -> search text: 7 CFR 1770 (RUS borrowers' uniform accounts: scrap carried "at the estimated amount which will be received therefor"), 7 CFR 1767.41 (salvage tickets), 2018 audit policy (auditors test "sale or lease of plant, material or scrap in conformance with RUS requirements"). Confirms the lender-approval frame; does not say why. Single source, stated.

Copper co-op thread: two hops, one clear answer (the windfall is small and lender-governed for at least one co-op). Not three empty hops, but the question is answered as far as public books go; switching to the second thread for discovery. Second thread: steer item 2, hosting-cost-claim surplus-hardware version, "who actually runs on surplus", taken historically: Berkeley's Community Memory (1973), a public terminal network run on a surplus SDS-940 time-sharing machine by Resource One. Not in the wiki (grep found nothing). It is a forgotten institution ten minutes from El Cerrito that ran community computing on discarded corporate hardware, which is James's thesis fifty years early.

1. Expect: who gave Resource One its surplus machine, and on what terms. -> search text (Wikipedia, FoundSF): Transamerica gave a surplus XDS-940 worth ~$150,000 "in exchange for a tax write-off" in 1972; free-access community computing for Amnesty International, Tenants' Action League, La Raza en Acción.
2. Expect: Felsenstein's own account of installing it. -> FoundSF "Installing a Mainframe at Project One", read in full. The machine was the SRI 940 Engelbart used for the 1968 demo, off-lease, returned to Transamerica Leasing, given on "long term loan" with terms "please don't give it back"; plus $120,000 cash. It needed 23 tons of air conditioning and a sealed room; the operating system was the defunct Berkeley Computer Corporation's, in a language "found on almost no other computers". The planned users, the switchboards, never came: a leased Teletype at $150/month was beyond them, and their files lived "in the head of its director".
3. Expect: how the public terminal started and what Illich had to do with it. -> FoundSF "Community Memory: The Beginning", read in full. Surprise: the county librarians told them they were "a library with nothing on the shelves"; the first terminal went into the student-owned record store, on a phone prefix with free calls to San Francisco; a "bagels" query brought an offer from an ex-bagel-maker to teach, which Felsenstein read as Illich's "learning exchanges" from Deschooling Society made real. Illich's conviviality decided not to armour the terminal: "In order to survive in a public-access environment, a piece of computer equipment must grow a computer club around itself."
4. Expect: why the surplus machine was abandoned and how later versions ended. -> search text (CHM promotional material 1982-91, FoundSF, CSMonitor 1984): v1 closed because the old computer "was dependent on successful operation of unreliable and irreplaceable equipment" and could not be replicated; v2 (1984-88/89) was paid for by Felsenstein's Osborne founder stock (~$450k) and $20,000 from Wozniak, with terminals in the Berkeley Co-op supermarkets and La Peña; it ended when the Co-op collapsed. Planned to self-fund at 5 cents a minute.
5. Expect: Felsenstein on the ending. -> FoundSF "Reflections on the End of Community Memory", read in full. v3 (1990-92, ten terminals in libraries, laundromats, a senior centre) closed when a state grant ran out; coinboxes built but never switched on; he refused dial-in from home computers because owners of equipment would gain "advantages over the general usership"; going commercial would have meant taking on liability for users' speech. "The only vandalism ... took place when we had shut down the system and left a useless terminal exposed."
6. Expect: the Tom Swift Terminal spec as the bridge from Community Memory to the personal computer. -> search text: published in People's Computer Company, Nov 1974; became the Sol-20 at Processor Technology. Confirms; adds nothing the memoir doesn't. Stopped: the trail had turned to confirmation.

Hunches (unsourced, not in wiki)

• Surplus hardware gave Community Memory its start and also its first death: the gift was free but the operating knowledge was not, and an "irreplaceable" machine cannot be replicated in the next town. The convivial unit was the terminal and the club, not the mainframe. That is a sharper version of the age-window limit already on hosting-cost-claim.
• Every version of Community Memory ended through its host or funder (Resource One, the Berkeley Co-op's collapse, a state grant), never through users leaving. It sits with the farmer mutuals and SNET as an ending by the substrate, and with CIDOC as a deliberate stop (Felsenstein chose to shut down rather than go commercial).
• The Peoples Rural result suggests the copper windfall in small US co-ops is mostly eaten by removal cost; the large windfalls (BT, Algar) come at scale, from dense duct. Scale may matter more than ownership for re-extraction.

End of session

Surprises

• James's thesis has a working ancestor ten minutes from El Cerrito. Community Memory (Berkeley, 1973) ran public terminals on a surplus mainframe Transamerica gave away for a tax write-off (the SDS-940 Engelbart used for the 1968 demo), and its engineer used Illich's conviviality to decide the terminal should be repairable by its users rather than armoured. The surplus mainframe also killed the first version: "unreliable and irreplaceable equipment", an operating system in a language found nowhere else, so it could not be replicated. https://foundsf.org/Installing_a_Mainframe_at_Project_One ; https://www.foundsf.org/Community_Memory:_The_Beginning ; https://archive.computerhistory.org/resources/access/text/2019/02/102734414-05-01-acc.pdf
• Felsenstein refused to let home computers dial in, because people who owned equipment would gain "advantages over the general usership"; walk-up terminals in co-op groceries, laundromats and libraries were the point. The system died three times by its hosts and funders, never by its users. https://www.foundsf.org/Reflections_on_the_End_of_Community_Memory
• The copper-windfall hunch failed. Peoples Rural Telephone Co-op's audited books show salvage "net of removals" of $0-$76k a year while it pulled all its copper, booked against accumulated depreciation (so it never reaches margins or capital credits), and its federal lender must approve any disposition of "plant, material, or scrap". https://psc.ky.gov/Utility_Financial_Reports_Net/uploaded/support/SD_12500_2024_1.pdf

Dead end: none new. The co-op salvage question is answered for one co-op, negatively. Worth continuing: community-memory. Next hop: the Community Memory collection at the Computer History Museum (the 1982-91 promotional PDF and the CM Digest printouts), for per-terminal running costs and what the coinbox model assumed, to set beside the surplus-hardware claim. Permission to chase: Resource One's later life: the San Francisco Social Services Referral Directory, printed monthly from the 940. Possibly the first city-wide community database run on surplus kit, and a precedent for the inference-testbed. Budget used: 10 fetches (3 PDFs, 4 searches, 3 pages read in full); ~19 tool calls (of 40).

Lint: 0 orphans, 0 unindexed, no real dead links; four personal/utility pages still lack a disposition (left for James).