# Research context

PFABC is a proposed formal framework and experimental software contribution.
This package does not establish peer review, empirical validation of its bridge
registry, or improved human reasoning. Its code can implement a stated rule
correctly while the scientific usefulness of that rule remains open.

## Relevant primary literature

- Patrick Butlin and colleagues, [Consciousness in Artificial Intelligence:
  Insights from the Science of Consciousness](https://arxiv.org/abs/2308.08708)
  (2023). This research report derives indicators from several scientific
  theories and discusses assessment of AI systems. Its claims about systems
  available at that time must not be presented as observations of current models.
- Patrick Butlin and colleagues, [Identifying indicators of consciousness in
  AI systems](https://doi.org/10.1016/j.tics.2025.10.011) (2025). This article
  develops a theory-based indicator approach while discussing uncertainty and
  further research. Its indicator methodology is related context, not validation
  or endorsement of PFABC's specific rules or registry.

Primary sources checked September 7, 2026. These references are an entry point,
not an exhaustive review or a claim of scientific consensus on PFABC.

## Assumptions to keep visible

The supplied formal specification makes operating modes and theory choices
explicit. Its target-realist mode adopts a non-eliminativist prior concerning
protected presence-for. The prior is not evidence that this target exists or
that a particular system has it. Other modes frame the target differently.
The conversation exercise can be used without endorsing that metaphysical prior.

The registry's indicators, thresholds, and links are part of this proposed
implementation. A configured threshold is not automatically an empirically
calibrated threshold. A bridge declared as passed is not a test conducted by the
software. The strength and relevance of the underlying observations still need
scientific assessment.

The learning task is symmetrical: preserve justified bounded capabilities,
identify unsupported stronger attributions, and avoid treating missing evidence
or an AI's denial as proof of absence. Testimony can be relevant; its credibility
for the particular inference requires an argument.

## Contribution and open questions

This release makes records, interpretations, assumptions, revisions, and
conditional evaluations inspectable. A useful next evaluation is whether people
can transfer those distinctions to a new case without being coached. The
[proposed pilot](PILOT_PROTOCOL.md) measures that separately from software tests.
Neither a passing test suite nor a favorable AI review establishes that effect.