Skip to content

Control plane vs source-of-truth

PETROVA splits two things most documentation systems conflate: methodology (how you work) and project truth (what the work governs).

Lives in PETROVA-HQ (the control plane)Lives in each consumer repo
META-RULES.md (MR-1 … MR-12)docs/north-star/intent.md
Verb schemas (spec/verbs/*.json)MILESTONES.md
Registry (registry.yaml)docs/decisions/*.md
Bootstrap prompts (core/prompts/)docs/findings/*.md
Integration contractsAGENTS.xml (consumer roster)
Skill packages (skills/*)docs/spec/ (project requirements)

The dividing line is MR-12: a CLAUDE.md projects from grounded sources; it does not introduce facts of its own. Methodology is shared because every petrova-line repo uses the same MRs, the same verbs, the same fleet identity grammar. Project truth is local because what kahn-hq is for and what smo1-io is for differ — and their decision ledgers must travel with their code.

Putting north-star content in a central database would break three things at once:

  • git blame stops working. Decision context decouples from the code it justified.
  • Branch protection bypasses become possible. A central edit surface that doesn’t go through PR review violates the every write is reviewable invariant.
  • Cross-repo intent confusion. A central north-star tempts you to homogenise project intent. Different repos serve different purposes; central storage erodes that.

Centralising methodology has the inverse effect: every consumer repo inherits the same disciplined rules without re-litigating them, and a single MR change ripples cleanly through the whole line.

When you find yourself reaching for “where should this live?”:

  • If it’s the how of working (a process rule, a verb shape, an agent-fleet boundary), it belongs in PETROVA-HQ.
  • If it’s the what of a specific project (intent, constraints, history), it belongs in that consumer repo.
  • If it’s both — say, an architectural decision that has methodology implications — open the decision in the consumer repo (per MR-7), and if it ends up changing an MR, open a follow-up decision here to record the methodology change.