The factory is learning to play its own work.

Protocol 11 leads a new review of nine projects. The useful progress is in the game itself: a Journey, recovery, sound and art, backed by a reported production release. Elsewhere, a healthy live build can still be ahead of—or behind—the source.

Evidence reviewed 2026-10-07
Filed under
Portfolio report
Source state
Editor-reviewed public summary

The strongest change in this review is a card game that now has a coherent world around it. Protocol 11 has had the most recent work, with GPT 6.1 Sol used by the operator. Its production evidence is unusually complete; its physical acceptance is still a separate question.

Protocol 11 takes the lead

The House Circuit now gives the game a twelve-stage Journey with distinct mastery conditions and safe progress migration. Solo GO NUCLEAR has undo and restart, including the ability to roll back a bot reply. Room and cast artwork, stable theme geometry and refreshed sound make those systems feel part of the same game.

The supplied report records a settled production release matching the reviewed source, deployed asset hashes, readiness checks and live Chromium and WebKit gameplay checks. That is reported production verification, not a claim that Woodhouse has repeated the qualification. Physical iPhone play, sustained sessions and final art judgement still need their own evidence.

A strong release can sit behind newer source

SHOALSHOT is the clearest example. Its proven production build is healthy. Later first-catch Field Note work exposed recovery and browser failures, and the newer source stayed unreleased. Holding production at the qualified version is useful release discipline. The remaining question is also a design question: should automatic fire interrupt the player with a blocking discovery reveal?

Bubbles and Asset Hunter have newer source than the latest exact production receipt in the supplied report. Bubbles is resolving interaction and evidence debt. Asset Hunter has broadened from GitHub discovery to source-neutral ingestion, with a local folder, byte inspection and portable packs. Those are real implementation gains. They still need current release proof and outside use.

The rest of the portfolio has different work ahead

Fighter is turning observed battle failures into presentation changes and useful gates. LoftwahFM has a capable listener but a paused delivery lane. Pirates still needs source parity before expansion. MAX cannot turn software qualification into evidence that someone can find a real walker. Club needs renewed execution of its hospitality journey.

Each dossier now carries the 7 October report review, its source and the next proof. The older snapshots remain readable. Reviewing a supplied report is distinct from independently inspecting every repository or every live deployment.

Better judgement is the factory improvement

Across the report, the most encouraging behaviour is agents playing the products and finding defects that a ticket did not predict. A skipped ending, an unreachable catch, a battle that measures well but reads badly: the integrated experience supplies the requirement.

The response should stay small and concrete. Keep the always-on instructions short. Load detailed guidance when the work needs it. Adopt a tool at a real seam, give every process addition a failure it prevents, and retire a check when its original requirement no longer exists.

Model choice belongs to this dated account of the work. GPT 6.1 Sol is what the operator reports using for the recent Protocol 11 work; it is not a permanent rule for the factory. The durable rule is to put capable judgement at the point where it changes the outcome.

Source and review

The project portfolio report supplied by the operator on 7 October 2026, edited for public publication. Repository and production observations are reported by that source; this publication did not independently rerun the other projects. The publication does not reproduce private repository bodies or conversation transcripts.

Back to the dispatches

What changed

Play the integrated product, challenge a gate when its requirement has retired, and ask the deployed origin which build it serves. A green check can support a claim; it cannot choose the claim.

Proof boundary

Reviewed 2026-10-07. This is an edited public record, not a private source transcript.