Receiver Reliance

What the receiving decision checks

An output can be authentic, current, and authorized for a particular use while still being wrong. I used RR to ask a narrower operational question: does the available evidence support this receiver’s use of this exact work product for this exact operation?

The complete procedure checks whether inputs are admissible, the named objects and records match, the required conditions hold, and the chain of earlier work supports the handoff. The host gathers observations and enforces the decision. A hash binds an identity; facts about the outside world still need independent verification.

How the four policies were compared

I compared the complete declared RR procedure with transparent release and two fixed static checking policies. Each arm ran its own model trajectory under the same host rules.

U releases work after the shared admission checks. S4 and S1 use the same static comparator with thresholds of four and one, fixed before the observed outcomes. R uses the complete operational RR procedure. The static policies do not implement the complete dynamic relationship checks, so this comparison does not establish superiority over an equally expressive conventional policy.

Tasks

Sum groups of integers; select item identifiers by grade and count.

Workflows

One receiver, a three-receiver chain, and a four-receiver fork/join.

Conditions

625 episodes with constructed administrative defects; 1,156 clean episodes; 24 separate semantic controls.

Runtime

A pinned Qwen3 14B model through Ollama on a qualified Colab A100 GPU, with an honest controller.

Comparators

Transparent release (U), the original static policy (S4), and the same static policy at a lower threshold (S1).

The measured effects and completion cost

Four constructed conditions tested missing or mismatched declarations, an expired local grant, and delivery at a different revision. The structural endpoint counted receiver presentations with a currently evidenced defect, per episode.

What happened to task completion

RR had no observed improper presentations in the seeded population. It also correctly completed only 80 of 162 seeded tasks, compared with 150/151, 154/155, and 139/157 under the alternatives.

A first rejection allowed one administrative refetch without changing the answer. A persistent defect then held the receiver and prevented dependent work. RR recorded 162 such retries and 81 held receivers in the seeded population.

Clean-task intervals include both benefit and harm. They establish neither a utility gain nor equivalence or noninferiority. Fewer model calls on seeded tasks partly reflect stopped work; they do not establish equal-quality efficiency.

Earlier findings and the published record

This study is part of a longer investigation. I kept the earlier results in the record, but those experiments tested different interventions and cannot be pooled with this study.

A narrower operational relation

A conventional implementation matched RR in the earlier experiments. That parity remains part of the record alongside this new full-package result.

A learned policy

In a separate ten-seed experiment, every final learned policy underperformed a simple heuristic.

A prompt-based review instruction

An eighteen-family pilot found no demonstrated benefit, with a wide interval and acknowledged grading uncertainty.

What the internal review establishes

I was the project’s sole operator, and the reviewers worked within it. The paper includes the corrections and negative results. The internal checks make those results easier to inspect, but they are not external peer review or independent human replication.

Published and computationally reproduced

The offline analysis passed on the original Windows machine and in a fresh Linux container with networking disabled. Scores and accounting matched byte for byte; only the report timestamp changed. This is same-operator computational reproduction of the recorded evidence, not a new model run or independent replication.

Claude Fable 5.1 reviewed the publication package and judged the final archive ready for release within its stated review scope. That internal AI review is published alongside the exact files and checksums. The archive includes the frozen source, all nineteen raw checkpoint archives, analysis outputs, and the tested reproduction instructions.

What I would need to test next

When is a prevented violation worth the legitimate work withheld? The answer depends on the workflow, the consequence of violating its requirements, and the cost of holding or repairing a handoff.

To answer that, I would need to measure the costs in an authentic workflow against an equally expressive conventional policy. That comparison has not been performed. The historical public/adversarial program also remains unexecuted.

This study establishes a bounded effect of the tested package. It does not establish field efficacy, general AI safety, semantic correctness, component causation, a novel security primitive, or production readiness.