Separate generic graph writes from text MutationPlan evidence

[flyingrobots]

Classification

Native design debt and misleading authority vocabulary.

Problem

CheckpointPlan currently contains a member named mutation: MutationPlan, even though checkpoint declaration is explicitly required not to mutate text.

MutationPlan also bundles two different concerns:

1. generic Echo graph read/write footprint plus delta emission; and
2. text-result evidence such as buffer ID, head ID, byte length, line count, and version.

A checkpoint uses the first concern to write one Jim proposition, but fills the second with a retained basis head and a current buffer version. The name and shape imply a text mutation where none exists and make mixed-coordinate responses easier to construct.

Evidence

• MutationPlan definition and graph emission
• CheckpointPlan { mutation: MutationPlan }
• Checkpoint planning writes only CheckpointFact

Why it matters

Names are part of the authority boundary. Calling every admitted fact write a text mutation weakens the invariant that checkpoint declaration cannot mint a head, rewrite, diff, or text change. The combined structure also encourages unrelated coordinate evidence to travel together.

Constraints

• This is a structural refactor, not permission to deepen a handwritten inverse or mutation protocol.
• Preserve generated operation identity, footprints, deltas, receipts, ticks, and WAL behavior.
• Preserve exact content-addressed fact identities.
• Keep checkpoint planning visibly non-text-mutating.
• Prefer a small split over a generalized framework.

Acceptance criteria

• Extract a neutrally named graph footprint/delta plan containing only graph reads and pending writes.
• Keep text-mutation result evidence in a type used only by operations that actually create/replace text state.
• Make CheckpointPlan contain explicit checkpoint-basis evidence and the neutral graph plan.
• No checkpoint field, method, or test describes the operation as a text mutation.
• Existing create/replace/checkpoint native and integration witnesses remain green.
• Add or retain a negative witness proving checkpoint emission contains no RopeHead, RopeRewrite, RopeDiff, or text-node write.
• No protocol or wire-shape expansion is introduced merely to complete this refactor.

Related

• Route checkpoint declarations through Echo #282 introduced the first installed non-text-mutating Jim proposition.
• The response-coordinate follow-up should be solved with explicit evidence types, not another overloaded plan.

[coderabbitai]

🔗 Related PRs

#23 - feat(echo): add Echo release gate witness [closed]
#205 - Document graph-backed rope runtime gate [merged]
#265 - CR-01: Separate checkpoints from Echo anchor authority [merged]
#280 - Add runtime-backed causal range evidence [merged]
#282 - Route checkpoint declarations through Echo [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!