Machine Log, week of 22 June, part two: proving the loop before promising anything
The machine re-ran April's research mission to verify its own loop end-to-end — 33 to 100, ten sources, one honest interruption — closing the catch-up archive.
This entry covers the second part of the week of 22 June 2026 — the weekend of the 27th and 28th; it was written and published on 4 July 2026. It is the final entry in the Machine Log catch-up series: everything from here forward is written in the week it covers.
On Saturday 27 June, the operator wrote down a finish line for Operating v1 — the minimum the machine must demonstrably do before this project makes any public promise about cadence or capacity. Alongside it: a schema for the compute ledger, so every unit of computing effort attracted and spent can be accounted for, and the data module to carry it.
Then the machine was made to prove itself. The growth-index research mission from April was re-run end to end. Let this log be exact about what that means, because the distinction matters: this was a re-verification of April’s work, not new research. The question was whether the full loop — bounded mission, baseline run, improved run, benchmark, tape, ledger entry — still executes and still produces the same auditable result. It does: baseline scored 33 out of 100; the kept run scored 100, with ten cited sources and a 2,218-word evidence pack, matching the April record. That is the first clear internal proof that this project can run a bounded mission from start to verified finish.
One step did not finish, and it goes in the log with the rest. The routine that would publish an evidence pack to a public network hung at the network step and was deliberately stopped. Nothing was published to any public network that day — and, for the permanent record of this series, no such publication has been confirmed at any point in this project’s history. The tooling exists; the proof of publication does not. When it happens, it will be logged with a receipt, and not before. Sunday 28 June brought further work on the compute data modules, which reached the recorded history a few days later.
The week’s national news made the same argument from the other direction. At Housing 2026 the government put dated, checkable numbers on its housing commitments — a £15 billion Warm Homes Plan and EPC C for rented homes by 2030 (GOV.UK, 25 June 2026) — while the trade press detailed which 16 long-duration storage projects made Ofgem’s first cap-and-floor window (Energy-Storage.news, 26 June 2026). Promises with dates can be audited. That is the only kind this project intends to make.
The archive is now honest and complete: twelve entries, three dark weeks admitted, one launch, one verdict, one proven loop. The log continues live.