Consensus Logic

Build log

Built in public. Every step tested, every failure posted.

One post per package. What was attempted, what changed, the test output, the reviewer's verdict, regressions found, failed approaches, and what remains not run. This is where the brand proves it applies its own rules to itself. Rejections are posted too.

How the product is built

The way it works.

Consensus Logic is developed the way it asks you to develop: one AI creates, the other reviews, every package closes on a test, and a test that cannot run reports a fail, never a pass. The founder holds the authority. Each entry below is the record.

  • Objective

    The package and the outcome it was meant to reach.

  • Test output

    What the test actually printed. Pass, fail, or could not run.

  • Reviewer verdict

    What the reviewing AI found, including what it rejected.

  • What stays Not run

    Everything the test did not prove, named.

Entries

Newest first.

Draft: entries written from session notes; verify each against the test output before publishing

Key: M numbers are milestones, P numbers are work packages, CR is a review of a correction.

001 · 7 September 2026

Groundwork done: M0 closed, P01 Foundation built. The name and the brand landed.

The foundation milestone closed on the founder's acceptance, and the first package went in. The product that was AI Development Controller is now Consensus Logic: new repository under the consensus-logic slug, the earlier repository tagged as harvest source and left in place. The brand kit, the G3 braid mark, the tokens file and the icons, is in the repository, and the identity test now asserts the icon as well as the name.

From this package the roles are split the way the product splits them: Codex creates, Claude reviews.

Objective
Close M0 and build P01 Foundation under build plan v1.8, specification V2.3.
Test output
M0 closed. Mechanism evidence: identity, tokens and harness checks. No product evidence yet.
Reviewer verdict
The first correction under review CR-001 was rejected: the reviewer found two sibling paths that would have passed when they should have failed. The second correction was accepted and the reviewer recommends closing CR-001. That closure waits on the founder's acceptance before the next package starts.
What stays Not run
Every one of the 58 acceptance scenarios. Nothing runs through an installed application yet.
000 · 6 September 2026

The reset. A specification, a plan, and a refusal to give dates.

Two earlier attempts were harvested rather than carried. Specification V2.3 and build plan v1.8 were adopted as the only sources of truth, with the rule that no claim on this site is made until the acceptance scenario behind it has closed. The build plan gives no calendar dates because the last two were wrong. This log inherits that refusal.

Objective
P00 and R0: fix the specification, the plan and the harvest sources; start clean.
Test output
Sources identified and pinned. No test to run yet.
Reviewer verdict
Not applicable. The review role starts with P01.
What stays Not run
Everything.

Subscribe

One post per package. Nothing else.

By email, or by RSS.

Draft: subscribe endpoint and RSS feed not wired yet