Who said yes to that match?

EPA's table of closed WIFIA loans has two rows eight days apart in September 2026. The first, with a close date of 16 September, lists the borrower as "Las Virgenes, California", for the Pure Water Project, $251 million. The second, closed 24 September, reads "City of Columbia", Long-Term Water Supply Program, Tennessee, $272 million. Neither row carries the name a vendor would likely have on its account list.

I believe that when AI reads utility records, a person should decide the uncertain links, and every decision should leave a log entry: who said yes, when, on what evidence, and what is still open. Software can propose the link, and someone with a name decides it.

Two loans, two judgment calls

Start with Columbia. Every board decision on the program sits with Columbia Power and Water Systems (CPWS). On 18 March 2026 the CPWS board approved the Master Bond Resolution and sent it to the City Council, 5-0. Software that reads the loan table sees "City of Columbia" and has to guess whether the loan belongs with the CPWS program. In our work at AquaIntel, a link like that carries the label "possible match" until someone checks it.

Here is what checking looks like. EPA's fact sheet for the loan names the borrower as the City of Columbia, Tennessee, and says the city, "on behalf of Columbia Power and Water Systems, will utilize WIFIA financing" for the program. That sentence is enough for me to file the loan with the CPWS program. It does not say how the city and CPWS are set up legally, or who carries the debt. So the decision I would log is narrow: the loan funds the CPWS program, accepted on the fact sheet, with the legal structure left as an open question.

The next document closed it. In a letter of 9 June 2026 to the Tennessee Local Development Authority, the city describes CPWS as its board of public utilities, which operates the city's water system. The letter says the WIFIA loan will be "payable from and secured by" a senior lien on the system's net revenues, "pursuant to the terms of the City's master water bond resolution." So the city is the borrower, and the debt is repaid from the water system CPWS runs. The log entry gains a second piece of evidence, and the open question gets an answer and a date.

Las Virgenes takes more steps. EPA's fact sheet names the borrower as the Las Virgenes-Triunfo Public Financing Authority. The board record I found belongs to a third body, the Las Virgenes Triunfo Joint Powers Authority. Its packet for the 8 December 2025 special meeting refers to "the JPA's WIFIA loan" and says the JPA's bond counsel advised it "on the formation of the Las Virgenes-Triunfo Public Financing Authority." The same packet names the JPA's participants: Las Virgenes Municipal Water District and Triunfo Water & Sanitation District.

That is one loan under a place name, a financing authority, a joint powers authority and two districts. A vendor's account list probably holds one of the districts.

The packet for the JPA's 6 April 2026 joint meeting with the financing authority sorts out the roles. The Pure Water Project "will be owned and operated by" the JPA. A draft of the state loan agreement in the same packet lists the WIFIA loan agreement as one among EPA, the JPA and the financing authority. The two districts pay the debt service directly to the financing authority under separate payment agreements: 70.60% from Las Virgenes Municipal Water District and 29.40% from Triunfo.

I would file the loan under the JPA and link both districts with their shares, with the April packet as the evidence. Someone else could reasonably file it differently, which is the reason to write the choice down.

What a rubber stamp looks like

A new book frames this as a question about people's roles: The Enterprise Brain, by Ragy Thomas, Sravan Vadigepalli and Chandhu Nair (2026). I am working from the public material on it. The book's site describes a move from operator to administrator to governor, where humans set principles, agents execute and humans audit. It also names an "Administrator Trap". I know that idea by name only. My reading is the staff member whose job shrinks to clicking approve on whatever the software puts in front of them.

The trap describes how a review queue goes wrong. A list of possible matches with one Approve button gets cleared fast. Each click turns a guess into a fact, and the next report shows it with no label. When a later question depends on the link, nobody can say who made it or why.

A log entry that prevents this needs five things:

1. The link, in the words of both records: "City of Columbia" on the loan, the CPWS Long-Term Water Supply Program on the other side.
2. The decision: accept, reject or hold.
3. The evidence, as a link to the page that settled it. For Columbia, the line in EPA's fact sheet.
4. Who decided, and the date.
5. What stays open. For Columbia, the legal structure, until the city's June letter answered it.

An approve button that skips item 3 produces rubber stamps.

Decide the task, then decide who does it

In a YouTube interview ("What Does UnifyApps Actually Do?"), Ragy Thomas talks about separating what must be done from who does it, a person or a model. For matching, the task is "decide whether this record belongs to this utility, and say why."

Today a person should take the uncertain links and software can take the exact ones, such as two records with the same system ID. Over time the log shows which kinds of calls people accept every time on the same kind of evidence. Those can move to software, with the rule written down next to them. Calls where reviewers disagree stay with people. The log makes that move safe, because anyone can see what the rule was built on.

The log also pays off on the next record. When another loan, SRF list or bond notice names the City of Columbia, the earlier decision comes up with it, along with the fact sheet and the open question. Nobody does the Columbia research twice.

Match review / Brannock Valley Water Authority / Possible matchesACME · workspace

Possible matches for Brannock Valley Water Authority

4 records named differently from the utility · 1 waiting, 3 decided

Uncertain link: a person decidesExact ID match: accepted by rule

Review queue

RecordName foundProposed linkWhy proposedDecision

State SRF list“Brannock Valley PFA”Brannock Valley Water AuthorityName overlap, same countyAcceptRejectHold

Board minutes“BVWA Phase 2 intake”BVWA Intake ProgramAcronym in minutesAccepted

Loan table“City of Brannock”Brannock Valley Water AuthorityCity name in utility nameHold

Permit notice“BV Authority WTP”Brannock Valley Water AuthorityAddress matchRejected

Decision on the highlighted row

EvidenceLink required before Accept or RejectReasonOne line, in your wordsOpenWhat the evidence does not settle

Audit log

Accepted: BVWA Intake Program12 Oct 2026, 10:42

WhoAccount lead, EastEvidenceBoard minutes, 14 Sep 2026, p. 6ReasonMinutes use BVWA for the authority.OpenNone

Hold: City of Brannock loan9 Oct 2026, 16:05

WhoAccount lead, EastEvidenceLoan fact sheet, p. 1ReasonFact sheet says the city borrows for the authority.OpenWho carries the debt?

Rejected: BV Authority WTP permit7 Oct 2026, 11:18

WhoEngineer, EastEvidencePermit notice, p. 2ReasonPermit holder is a different authority at the same address.OpenNone

Reused decisions

2 new records this week name “BVWA”. Each arrived with the 12 Oct decision, its evidence link and its reviewer attached.

The people who settle these calls

For utility readers, this is the human side. Someone in the CPWS or city finance office could settle the Columbia question in a minute. At Las Virgenes, the people who worked on forming the financing authority know which body carries the loan. Often that knowledge sits with senior staff: the finance director who set up the authority, the engineer who knows which acronym belongs to which project. Much of it is written down nowhere.

An endorsement of the book by Ferrazzi, which I have only in paraphrase, says people fear AI for what it reveals. I think that applies here. Ask a senior engineer to review a machine's matches, and the review can feel like a test of everything they never wrote down. I would frame it as the place where their knowledge gets recorded, under their name, with their reason. A log entry that names who settled a call gives them the credit, and the answer stays with the utility after they retire.

The same holds inside a vendor team. The long-tenured rep who knows that two names are one buyer is the right reviewer, and the log keeps that knowledge when the territory changes hands.

Our own numbers show how much of this work waits on a person. In AquaIntel's utility database on 15 October 2026, 8,913 links between a utility and a source record were confirmed. Another 3,822 proposed links were marked as needing research, 2,510 were held as a looser association, and 21 had been rejected. Software had proposed 5,199 candidate drinking water systems, and none was accepted automatically. Every one is in a review queue.

What I don't know

I have read how the city's own letter describes the Columbia master bond resolution, and how the Las Virgenes packet describes the parties to the WIFIA loan agreement. I have not found the bond resolution itself or the signed loan agreement posted publicly. I do not know how large a queue gets before reviewers start clicking without reading. And these are two loans.

I see these as new skills to learn. People will have to adapt to the new ways of working, and soon.

Try this on Monday

Pick three utilities on your account list. Search for each in EPA's WIFIA closed-loans table and in your state's SRF project list. Where the borrower name differs from your account name, write down which account you would file it under, the page that convinced you, your name and the date. Then ask the colleague who knows that account best whether they agree.

At AquaIntel, software proposes the links and a person decides each one. None of the 5,199 candidate systems in our database was accepted automatically.

Sources

• EPA WIFIA closed-loans table, columns "Borrower, Project Name, State, Close Date, Loan Amount" (last updated 16 Oct 2026). Rows: City of Columbia, Long-Term Water Supply Program, Tennessee, 9/24/2026, $272 million; Las Virgenes, California, Pure Water Project (PWP) Las Virgenes-Triunfo, 9/16/2026, $251 million: epa.gov
• EPA fact sheet, Columbia, Tennessee ("Borrower: City of Columbia, Tennessee"; "Columbia, Tennessee, on behalf of Columbia Power and Water Systems, will utilize WIFIA financing"): epa.gov
• EPA fact sheet, Las Virgenes, California ("Borrower: Las Virgenes-Triunfo Public Financing Authority"): epa.gov
• Las Virgenes Triunfo Joint Powers Authority, special meeting agenda packet, 8 Dec 2025 (participants, pp. 8-9; bond counsel item and "the JPA's WIFIA loan", pp. 20-21): lvmwd.gov
• CPWS Board of Public Utilities minutes, 18 Mar 2026 (Master Bond Resolution to City Council, 5-0), on the CPWS board page: cpws.com
• City of Columbia, letter to the Tennessee Local Development Authority, 9 Jun 2026, in the TLDA board packet of 24 Aug 2026, p. 35 (CPWS as the city's board of public utilities; WIFIA loan secured by system net revenues under the master water bond resolution): comptroller.tn.gov
• Las Virgenes Triunfo JPA and Public Financing Authority, joint special meeting packet, 6 Apr 2026 (project owned and operated by the JPA, p. 94; payment shares 70.60% and 29.40%, p. 92; WIFIA loan agreement parties, p. 114): lvmwd.gov
• AquaIntel utility database, 15 Oct 2026.
• The Enterprise Brain: Rewiring Your Business for the AI-Native Era (Ragy Thomas, Sravan Vadigepalli, Chandhu Nair, 2026). Book site (operator, administrator, governor; Administrator Trap).
• YouTube, "What Does UnifyApps Actually Do?" (Ragy Thomas): decoupling what must be done from who does it.