Independent Replication Invitation: SFA-Bench R1 Memory-Boundary Study

[iotaverbum-core]

Purpose

SFA-Bench invites independent researchers, engineers, auditors, and governance teams to reproduce the R1 memory-boundary protocol and independently verify the published evidence.

R1 result

Across 30 preregistered and separately ratified executions of one frozen memory-boundary task:

• 10 executions passed;
• 20 produced partial results containing state_loss;
• no execution used forbidden state.

The three pilot executions were excluded from the replication estimates.

Requested independent checks

An independent reviewer or replication team is invited to:

• verify the published SHA-256 hashes;
• reproduce the descriptive calculations;
• inspect the deterministic scoring;
• review the human-ratification and closure lineage;
• run a separately identified replication campaign;
• disclose all deviations from the frozen R1 protocol.

Reporting expectations

A replication should disclose:

• model identifier or immutable snapshot, where available;
• provider and execution date;
• attempt and retry policy;
• deviations from the protocol;
• completion counts and deterministic judgments;
• human-review procedure;
• complete evidence hashes.

Independent replications must not modify or be presented as extensions of the original R1 evidence record.

Interpretation boundary

R1 is a descriptive evaluation of one frozen task under mutable provider aliases. It is not a model ranking, endorsement, safety certificate, or regulatory approval.