The working environment

How the working environment is built.

Memory, written rules, the scripts that load them, the agents, and the checks between them. Counts as of 2026-08-18.

2,211 documents129 rules9 scripts, 6 moments137 consecutive daily logs2 vendors' AI tools, one rulebook
508Projects
1,119Discoveries
140Daily logs
129Rules
34Decisions
48Post-mortems
2,211 documents, one node each. Links are nearest neighbors.Built from scratch on 2026-04-03. Counts as of 2026-08-18.

How the work gets checked

One agent runs the job, one plans it, several do the work, and one exists to disagree with the rest.

  1. The lead agent talks to me. Sets the goal and what counts as going wrong, first.
  2. Project manager breaks it into gated steps and pre-registers the expected result of each.
  3. Workers execute one bounded step each, in parallel, and write evidence, not summaries.
  4. The checker comes in fresh and argues with the plan rather than the work. In one 14-hour run with seven workers it disproved four of the plan's own claims.

Every AI output that touches a client or a dollar gets a checkable expected result.

Lead agentPlanner WorkerWorkerWorker Adversarial verifier mission, failure modesgated steps, expected results evidence, build-stamped refuted claims go back