Machine Log, week of 18 May: the controller's first verdict
The first Compute Sprint Controller run ranked twenty candidate missions and told the project to stop building until real people could catch the outputs.
This entry covers the week of 18 May 2026; it was written and published on 4 July 2026 as part of the Machine Log catch-up series.
Two build days this week, and the second one produced the most useful sentence the machine has ever generated.
Wednesday 20 May: a public scorecard was added — the running record of what the machine has attempted and how it scored — the sprint’s status was written down, and the default mission was made runnable straight from the Control Room, so starting real work no longer required the operator’s private tools.
Thursday 21 May: the first run of the Compute Sprint Controller. In plain terms, the controller is a routine that takes every candidate mission — twenty of them, in this batch — and ranks where limited computing effort should go next, instead of letting enthusiasm decide. Its verdict was recorded in the run log, and it was blunt. The bottleneck was not research capacity or ideas; it was the unverified direct channel to supporters. The default next action it set was quality-checking that channel. And its headline conclusion, quoted from the log: do not spend heavily on frontier agent work until the direct supporter channel and review loop can catch the useful outputs.
That deserves a plain-English translation, because it is the opposite of what AI projects usually tell themselves. The machine assessed its own situation and concluded that more machine was not the answer — that generating brilliant outputs nobody receives, reviews or acts on is indistinguishable from generating nothing. Build the catching apparatus first. This log can already confirm the diagnosis was correct: the five briefs still parked at review since 29 April were the evidence, sitting right there in the queue.
The same Wednesday, by coincidence — and this series never claims more than coincidence — the Chancellor proposed her own bottleneck removal: legislation to let Parliament designate clean energy projects as Critical National Importance, narrowing judicial-review exposure and putting other major projects on a fixed-window challenge route (GOV.UK, 20 May 2026). The week also produced the first serious constitutional analysis of the new Devolution Act (UK Constitutional Law Association, 19 May 2026), and a hydrogen pipeline directed into the national consenting regime the day before. Everyone, it seems, spent the week asking the same question: where exactly is the constraint, and what would actually move it?
The controller’s answer now stands on the public scorecard. The next entry covers what happened when the machine went dark for three weeks — and the single day that brought it back.