Document Protocol Theta recursive-feedback provenance

[Steake]

What changed

• adds a commit-level provenance report for GödelOS recursive output feedback and Protocol Theta
• adds a publication-style technical note comparing the historical architecture with Hasan and Hossain's stateless self-feeding protocol
• recovers and distinguishes the formal (C_n), implementation proxies, archived lowercase c, (\Phi), global-accessibility, surprise, stability and suspension measurements
• adds a publication and archival plan plus a collegial authors' note
• audits three distinct historical whitepaper PDF blobs across four supplied filenames, recording SHA-256/Git identities, mutable internal dates, the duplicate Version 3/Version 4 payload, and the unsupported Version 3 result claims
• adds a standalone three-condition experiment:
  • independent same-seed baseline
  • exact stateless output-to-input recurrence
  • source-faithful GödelOS recurrence with the recovered metacognitive wrapper
• adds a reproducible retrospective analyser for the surviving Grok and DeepSeek JSONL

Why

The repository contains September 2025 implementations and recorded trajectories that predate arXiv:2608.11348v1, but the surviving evidence has material gaps. This change preserves the useful historical result without laundering uncertainty: complete prior-output reinjection is strongly established, while historical execution of the paper's exact raw-output-as-sole-input protocol is not.

The work is framed around GödelOS's original programme of bounded recursive cognition, self-models, continuity and falsifiable consciousness-correlate hypotheses. It makes no allegation of copying or derivation and does not claim that recursive feedback demonstrates phenomenal consciousness.

Impact

Researchers can now inspect the precise provenance boundary, rerun three explicitly separated protocols, and reproduce the modern lexical analysis without confusing it with historical metrics. Historical artefacts remain in Git history rather than being silently restored or modernised at HEAD.

Validation

• python -m py_compile experiments/recursive-feedback/*.py
• python -m unittest discover -s experiments/recursive-feedback -p 'test_*.py' -v — 5 passed
• deterministic smoke run — 18 JSONL records across 3 conditions, 3 depths and 2 repetitions
• historical analyser reproduced the reported 1,248 non-synthetic recursive/shuffled records
• local Markdown-link validation passed
• git diff --check passed
• remote Git blob SHAs match all nine local files

Known provenance limits

• the exact historical JSONL-generating runner is missing
• archived per-depth request inputs and literal seed prompts are absent
• provider RNG seeds and immutable model revisions were not recorded
• old Git commit dates do not independently establish first public push time
• the Version 3 PDF's numerical result table has no recovered raw data or generating analysis