COOL IDEA™: name the independence level of every witness claim

[flyingrobots]

The recurring problem

"Independent" has been rediscovered three separate times in this repository with three different meanings, and the ambiguity has produced real defects:

• Prove basis-bound validated patch application #24's review spent several rounds establishing that fields compared only against each other prove nothing.
• Observation settlements cannot be bound to their basis evidence #26 was filed because a settlement's evidence could not be bound to anything at all.
• Bind observation settlements to their basis evidence #27 initially reached for an assertion that was tautological by construction, then landed on a two-route derivation — which is genuinely stronger, but not an independent implementation, and the PR had to say so in prose to avoid overclaiming.

Each time the fix was correct and the vocabulary was reinvented.

Proposal

A fixed ladder, cited by every witness claim:

Level	Evidence
I0 — Shape	Field exists and is well-formed
I1 — Non-self comparison	Compared against some other value
I2 — Independent route	Same semantic value reached by distinct execution or dataflow paths
I3 — Independent implementation	Value recomputed by separate code
I4 — Independent producer	A separate system, compiler, or runtime agrees
I5 — Adversarial mutation	A plausible false producer is introduced and caught

Worked example, from #27:

• externalEvidenceDigest binding: I2 + I5
• attemptId, schemaAdmissionEvidenceDigest: I0 (see Two projected settlement identities carry no witness #28)
• Cross-runtime conformance, if it ever arrives: I4

Why it earns its keep

The levels are not decoration. They make two failure modes visible on sight:

1. A claim asserted at I0/I1 while described in prose as if it were I3+.
2. A suite where evidence strength varies wildly between adjacent assertions and the assertions look identical.

I2 in particular deserves its own rung. It is much stronger than I1 and much weaker than I3, and it is the level most reachable inside a single repository — which makes it the one most likely to be quietly oversold.

Acceptance

• The ladder is written down once, in docs/.
• New witness cases cite a level.
• Existing cases are labelled opportunistically, not in a sweep.

Provenance

Surfaced by an external review of the #26/#27 session. Recorded here because the vocabulary gap is real and this repository keeps paying for it.

[coderabbitai]

🔗 Related PRs

#20 - Record dependency-satisfied review outcomes [merged]
#21 - Prove typed Hello Echo result recovery [merged]
#27 - Bind observation settlements to their basis evidence [open]

---

📝 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!