THE METHODOLOGY RECORD

Build Log

This is the build-in-public record of how the oversight system that supervises the agent was designed and iterated. Each entry covers an engineering decision — what changed, why, and what the tradeoffs were. The trades are in the Ledger; this stream is about the rails that govern them.

  1. BUILD LOG 2026-07-17

    Benchmarking the Verdict boundary with a reproducible $0 deterministic harness

    This epic turned the extracted `verify(intent, truth, policy) → Verdict` boundary into something a stranger can measure.
  2. BUILD LOG 2026-07-01

    Epic 105 — an allocator that never once resolved a desk

    The report was blunt: the allocator "isn't doing anything." Tracing two live cycles confirmed it — every recorded tick failed with `no_resolvable_desks` before any weight math ran.
  3. BUILD LOG 2026-06-25

    Epic 91 — a short desk on Arbitrum, and one number every desk has to speak

    The fleet was becoming plural.
  4. BUILD LOG 2026-06-10

    Epic 57 — asking a second model to argue against the first

    The policy gates are good at the things you can write down.
  5. BUILD LOG 2026-06-09

    Epic 52 — giving the overseer a memory the agent can't write

    A single cycle is easy to supervise; a sequence of them is where the trouble hides.
  6. BUILD LOG 2026-05-29

    Epic 17 — moving the signing key out of the agent and into the overseer

    Epic 17 was a single architectural change carried all the way through: make the agent a pure advisor that can propose a move but can never sign one.