The supervision engineer can tell which variation is engineering and which is appetite. On the portal, the rows look identical.
A visual essay on AI, judgment and public money
The moat is the decision layer
The model can read the whole record. It still cannot know what this quarter is for. The advantage no procurement can copy belongs to whoever decides that, and answers for the outcome.
June 12 · eighteen days to fiscal close
The record knows everything except the point
Six facts from one road contract. On the desk: a payment certificate and a pending variation order. Pick what better means this quarter.
Data unchanged
Approve the variation. Release the certificate.
Eighteen days from the close, unspent funds return to treasury and next year’s ceiling drops with them. If absorption carries the most weight, keep the money moving.
This objective leans on: funds disbursed, days to close. It discounts: the 17-point gap between money and road.
The board did not change. The definition of better changed.
None of the three calls is corrupt, and none is automatically right. But the signature is not an abstraction. Whoever signs the certificate answers to the Auditor General.
The name
A decision is the facts, times a definition of better that someone owns
Prediction can narrow uncertainty. It cannot settle which legitimate goal deserves authority this quarter.
The record
What is happening?
61 percent built, 78 percent paid. Shareable, machine-readable, increasingly procurable.
The definition of better
What matters most now?
Close the year, protect the record, or build the market. Legitimate, competing, and not in the data.
The call
What do we do?
One action, one signature, one name that answers for the outcome.
The facts can be shared. The definition of better cannot be assumed.
The middle column is what this essay’s title calls the decision layer. Do not picture a component you can buy. It is work: choosing, out loud, which legitimate goal rules this quarter, and owning what follows.
Any ministry can procure the same dashboard, the same model and, increasingly, the same data standards. What no procurement can deliver is the middle column: a definition of better chosen here, this quarter, by someone who answers for the outcome.
This is not a public-works quirk. Swap the ministry for a restaurant at 7:02 p.m. with fourteen empty seats, or a classroom in exam season, and the same six-facts trap repeats wherever goals compete. It is also why chess flatters the machines: chess supplies its own objective, checkmate is checkmate. A road programme has to choose its objective again every quarter, and someone bears every tradeoff.
So name the design mistake plainly: treating an intelligence system as if it were already a decision system.
What the record cannot hold
The site knows things the portal does not
The people closest to the work hold context that is distributed, unwritten or simply newer than the record. Each note below attaches to a fact the board already showed.
The 61 percent is last month’s certificate. The culvert at kilometre 14 washed out on Tuesday.
The contractor is competent and stretched. One late certificate tips all three of their sites into arrears.
The escarpment section cannot be compacted wet. Miss the window and it waits until February.
Some of this can and should be written into the record. Some changes faster than any record. Some is learned only by standing on the site. A decision system has to make room for all three, without pretending the record already holds them.
The loop
The decision layer is not a box you own. It is a loop you run.
Run it on the June 12 board. Five stations, in order, and the last one feeds the first.
- 01
The record
The portal, the model, the forecasts. 61 built, 78 paid, one variation pending.
- 02
The site
What the engineer, the accountant and the contractor know that the record does not yet.
- 03
This quarter’s better
The priority, said out loud: this quarter we protect the audit trail.
- 04
The signed call
One person approves, holds or splits, and puts a name to it.
- 05
What happened
The outcome goes back into the record. The next call starts smarter.
05 feeds 01. The return stroke is the whole game.
Local judgment is not automatically right. It can preserve bias, folklore and obsolete practice. The loop needs evidence, review and outcomes that can correct it.
Now the title, paid off. Models get cheaper every quarter. Records spread; that is what disclosure standards are for. Dashboards are copied in a single procurement cycle. The loop is the part that cannot be procured, because it is not a thing. It is a practice, and it compounds: every turn leaves the record a little truer and the judgment a little sharper.
For a company, that is a moat. For a public institution it is something better: the capability that remains when the vendor leaves.
The moat
The moat is not the model, and it is not the record. It is the shortest healthy loop between the record, the people on the site and the consequences of the call.
June 12 will come around again. The board will show six clean facts, and a certificate will still need a signature. The numbers will not tell you whose it should be. Make sure it belongs to someone who answers for the road.
Facts are copyable. Purpose is not. The advantage goes to whoever keeps deciding what better means, and answers for it.