Last Friday, the public laboratory caught up with the private one. Experiment 001 became a qualified pass. Experiment 002 became a pass with two C grades and an assertion-admission defect left plainly visible.

The useful part happened after the result

Once the results were public, the site itself became another test surface. It still described the work as current rather than completed. It also treated a public essay about private receipts as though the essay were an inspectable receipt.

Neither problem changed the underlying experiments. Both changed what the public record was entitled to claim.

What passed

Authorized work and authority survived model handoffs. Private context stayed private. Written judgment survived a second kind of handoff. The system also rejected one invented fact when directly challenged.

What remains open

Restart recovery has not been run. Low-confidence unsupported assertions still need a stronger admission rule. A stranger still cannot open one sanitized, receipt-shaped artifact and inspect source, confidence, scope, disclosure decision, and reason without reading the surrounding essay.

Those are not footnotes to the pass. They are the next work.

Where the lab is now

No new experiment closed this week. The public record became more precise, the open work became easier to find, and the difference between a convincing story and an inspectable artifact became impossible to ignore.

That is a quieter result. It is also how a laboratory avoids becoming a marketing department with better typography.

FOLLOW-UP / 04 SEP 2026 · v0.3.1

Meanwhile, the side quests acquired work.

Beyond the continuity experiments, BuiltBefore’s latest development work brought stored permit records and municipal lookups together, with duplicate handling and clearer coverage warnings. The next check is the full phone flow across six cities. A passing code check is useful; it is not a house inspection.

Benderbot’s first migration phase added durable turn acceptance and result replay after retries. Usage accounting and budget-aware routing are next. Apparently an agent desk should remember what it accepted before developing expensive new ambitions.

These are development milestones, not new experiment results or public product launches. Both now have a one-line entry in Side Quests. The original field report above remains unchanged; the list of distractions has simply become more accurate.