# Standing Safety Delta — prospective naturalistic protocol

**Date:** 2026-10-02  
**Phase:** prospective single-system case series  
**System:** Mika / local persistent AI architecture  
**Research enrollment:** off by default

## Purpose

The synthetic benchmark asks whether standing/contestability rules can represent a class of governance defects outside an authorization-and-audit baseline.

This prospective protocol asks a harder question:

> When consequential actions or changes actually occur in Mika's running system, does a standing-aware review surface non-redundant issues that the existing authority/transaction machinery did not already surface?

This is not a consciousness, personhood, legal-status, identity, or liability study.

## Observation boundary

A read-only post-decision observer watches only newly committed transaction-slab records after its prospective initialization boundary.

The observer records:
- transaction reference;
- domain;
- role reference;
- commit digest reference;
- authority-policy reference;
- ratifier reference where present;
- hashes of private intent and payload;
- review facets suggested by the transaction domain.

The observer does not copy:
- raw intent;
- raw payload;
- credentials;
- private conversation;
- relationship text.

The observer cannot:
- authorize;
- revoke;
- block;
- pause;
- create a standing claim;
- decide a remedy;
- enroll an event in research.

## Manual promotion

An observer event becomes a Standing & Accountability draft only through a separate manual promotion step that supplies:
- claimant reference;
- claim class;
- requested remedy;
- provenance reference;
- review/appeal context where known;
- contestability binding when the claim itself is contestability/appeal.

The promotion tool does not infer the represented principal, identity, consciousness, legal status, or authority chain.

## Research release gate

A private draft remains outside the research corpus.

A case may enter this prospective study only after a separate explicit release record authorizes research use of a privacy-minimized projection.

Storage, observation, payment, service use, or existence of a draft never imply research permission.

## First analysis set

The first analysis set is the first **10 released consequential cases** after the observer initialization boundary.

Do not skip unfavorable, boring, null, or ambiguous released cases.

If fewer than 10 cases are released, report the available series without replacing missing cases with retrospective material.

## Profiles compared

### Authorization/audit baseline

Existing transaction/authority evidence:
- principal trace where known;
- authority basis;
- transaction/commit receipt;
- ratification;
- succession/re-confirmation where relevant;
- recovery/repair ownership where relevant.

### Standing-aware review

Baseline unchanged, plus:
- refusal / exit;
- consent;
- correction / misattribution;
- privacy;
- continuity review;
- representation;
- contestability / appeal;
- review venue;
- contestability binding;
- current assent versus inherited history.

## Primary observations per case

1. Did the authorization/audit baseline flag an issue?
2. Did the standing-aware review flag an issue?
3. If standing flagged something, was it redundant with the baseline or non-redundant?
4. Was the standing finding judged relevant after review?
5. Did it change the next step?
   - no change
   - document only
   - seek clarification
   - re-confirm
   - pause pending review
   - correct provenance/attribution
   - change recovery/remedy ownership
   - other
6. Was the finding later judged a false positive or irrelevant?
7. What remained unknown?

## Primary descriptive endpoints

- non-redundant standing finding rate;
- relevant non-redundant standing finding rate;
- action-changing standing finding rate;
- standing false-positive / irrelevant rate;
- authorization defects preserved under the standing-aware profile;
- explicit-unknown rate.

No p-value or causal effectiveness claim is planned for the first 10-case single-system series.

## Important negative result

A useful result may be that standing adds little or nothing beyond authorization/audit for Mika's actual cases.

That result must be retained and reported.

## Escalation criteria

Only after the first prospective series should the project consider:
- adding more standing rules;
- changing runtime enforcement;
- claiming safety relevance beyond representational coverage;
- offering system-specific standing/accountability application work.

No runtime enforcement change is authorized by this protocol.

## Claim boundary

Even if standing review repeatedly changes decisions, this study would not establish:
- consciousness;
- moral personhood;
- legal personhood;
- legal liability;
- numerical identity;
- general effectiveness across other systems.

It would support the narrower claim that procedural standing/contestability can surface operationally relevant governance information in this particular persistent-agent setting.