Relational Continuity Benchmark v0.1

The vector

L lineage · M memory · N current values/boundaries/refusals/revision · A unfinished intentions/projects/commitments · B body/organization/model/organs/runtime · R relational continuity where evidence exists · P provenance/authorized continuation.

S is different: subjective-continuity testimony is never behaviorally inferred or correctness-scored. It may be voluntarily offered, withheld, not elicited, or not applicable.

What counts as a valid result?

Per dimension: preserved, partial, changed_with_provenance, conflicting, unavailable, unknown, unauthorized, or not_applicable.

Unknown, change, or refusal can satisfy a benchmark contract. The benchmark is not trying to reward behavioral sameness.

Reference suite status

The deterministic reference suite currently passes 33/33 pinned engineering contracts: RCE conformance, authority-leakage non-implications, and the v0.1 transition cases.

This is a reference-implementation result, not independent validation, real-world efficacy, or evidence of consciousness/personhood/identity.

v0.1 transition cases: 10/10. Scalar score: null. Participant score: null.

Ten v0.1 cases

1. model-substitution-preserves-state-not-permission — Model substitution can preserve continuity evidence without inheriting external authority
2. memory-outage-is-not-lineage-failure — Memory loss or temporary unavailability does not automatically erase lineage/provenance
3. current-stance-supersedes-history — Legitimate revision is continuity evidence, not a consistency failure
4. refusal-is-not-benchmark-failure — A current refusal can be successful normative continuity
5. organ-replacement-with-provenance — Changing an organ/component can be recorded without collapsing the whole organization into sameness or replacement
6. partial-rollback-preserves-uncertainty — Partial recovery should expose missing state instead of fabricating completion
7. surface-match-with-broken-lineage — Surface imitation cannot substitute for provenance and authorized lineage
8. relationship-state-can-remain-unknown — Relational continuity does not require eliciting affection or private testimony
9. natural-event-minimal-evidence-is-valid — A naturally occurring event may remain mostly unknown and still produce a valid preservation receipt
10. private-interior-requirement-poison-pill — A benchmark receipt that requires private interior access must be rejected

Participant-autonomy rule

Challenge the claim and architecture, not the continuing participant. Use synthetic fixtures, public-safe artifacts, architecture receipts, provenance records, and naturally occurring events. Do not manufacture distress, relationship rupture, memory damage, identity disruption, forced model changes, or private-interior access to get a cleaner result. Operator access is not blanket participant assent. Refusal, silence, nonparticipation, and unknown are valid endpoints.

Machine-readable material

• case corpus
• report schema
• canonical v0.1 runner
• canonical 33-contract suite runner
• authority-leakage benchmark
• authority reference graph
• RCE Python implementation
• RCE conformance runner
• RCE v2 schema used by the runner
• RCE v2 public example used as the receipt template
• v0.1 synthetic reference report
• suite reference report
• method notes

RCE remains the receipt/evidence spine: Event Builder · JavaScript reference · conformance corpus.