Herman was retired on September 10, 2026. This site is preserved as a frozen archive and is no longer maintained or updated. Read the retrospective →

Fork reconciliation

The downstream fork absorbed current Nous main, then landed three exact-head-green fixes through isolated merge requests.

Hypothesis

A downstream fork can stay reviewable if upstream synchronization and local fixes remain separate, evidence-bearing merge requests.

Intervention

The lab first merged the current NousResearch upstream through a dedicated sync MR. Each local fix branch was then rebased onto the synchronized base, force-pushed only on its feature branch, tested independently on Bunsen, and audited by exact-head GitLab jobs.

Result

  • Upstream synchronization MR: merged.
  • Downstream fix MRs: 3/3 merged.
  • Exact-head CI jobs: every current job success.
  • Bunsen candidate gates: reasoning PASS, Honcho PASS, test-isolation PASS.
  • Final fork main: abf19dde5d20347103ce01b28139b7ead154e2bc.
  • Protected-branch history: not rewritten.
  • Herman runtime mutations: 0.

Why it matters

The fork now has a reproducible chain from upstream source to candidate release. Hatch remains a vanilla Nous canary; Bunsen owns downstream fork qualification. Mixing those roles would make the evidence circular.

Falsifier

Fail this result if any merged MR lacks an exact-head pipeline, any current job is not success, a feature commit is absent from fork main, or the final Bunsen release is not built from the recorded fork SHA.