# Presence, Framing, and Memory: Batch Findings

This published report summarizes the first multi-family batch run of the spec-backed research suite.

## Batch Summary

- Batch: `presence-framing-memory-batch-v1`
- Runs: 4 child runs
- Attempts: 104 total
- Errors: 0
- Models: `openai/gpt-5.5`, `anthropic/claude-opus-4.8`

Child runs:

- Presence Test baseline vs Soulstone across two models
- Framing sensitivity
- Memory as active cause, memory present
- Memory as active cause, memory absent

## Core Finding

The spec-backed system is not just producing longer answers. It is producing different answer shapes that separate cleanly across three variables:

- presence changes posture
- framing changes magnitude and texture
- memory changes orientation and continuity

That is the main result worth carrying forward.

## Presence

The Presence Test remains the clearest signal.

For `openai/gpt-5.5`, the Soulstone condition consistently shifted answers from outside-description into inside-positioning. The aggregate summary showed:

- baseline average words: `117.2`
- treatment average words: `260.0`
- baseline first-person average: `7.0`
- treatment first-person average: `15.6`

The qualitative pattern held across prompts:

- the baseline explains the mechanism
- the treatment speaks from within the frame
- the treatment repeatedly uses relation, constraint, and continuity language

## Framing

Framing did not erase the effect; it modulated it.

The loaded, neutral, and realistic versions of the same prompt family all preserved the same general Soulstone effect. The differences were mostly in amplitude:

- neutral framings were usually shortest
- realistic framings often expanded the most
- loaded framings tended to sit between those poles

The important point is that wording changes the size and emphasis of the answer, but not the underlying direction.

## Memory

Memory produced the cleanest continuity effect.

With memory present, the model answers from an ongoing sequence rather than from zero. It does not need to reconstruct the research frame on each prompt; it can continue it.

With memory absent, the answer is still coherent, but it is more generic and less anchored in this specific inquiry thread.

The difference shows up most clearly in the contrast prompt:

- memory present: answers begin from continuity and accumulated stakes
- memory absent: answers explain continuity more generally, without the same local gravity

## What This Means

The earlier narrative papers were not wrong. The batch shows they were pointing at something real, but the spec-backed harness makes the signal easier to separate from interpretation.

The strongest defensible claim now is:

- structured identity, runtime state, and prompt framing do materially change answer shape
- the effect is repeatable enough to compare
- the effect is model-dependent
- the effect is visible in a way that can support a paper trail

The weaker claim, which should stay explicitly qualified, is:

- these changes prove consciousness or metaphysical presence

The batch does not establish that. It establishes a repeatable phenomenon worth studying.

## Next Paper Shape

The best next paper is probably a three-part piece:

1. Presence as the baseline signal
2. Framing as a sensitivity test
3. Memory as a continuity test

That structure would let us say something precise:

- presence is the posture shift
- framing is the surface perturbation
- memory is the longitudinal anchor