Read the ideas in plain language, then follow each result to its protocol and original report.
Start here
| Document | What you will find |
|---|---|
| Project overview | The question, motivation, and current scope |
| Public framework paper | Five coordinates, causal validation, rivals, and falsification |
| Current findings | Completed results, non-replication, and interpretation limits |
| Methods and reproducibility | Experimental units, paired tests, gates, and evidence binding |
| Delegation architecture | Evidence, independent review, authority, and execution |
| Research roadmap | The next studies justified by the present record |
| Glossary | Plain-language definitions |
Every research page can be printed or saved as PDF with your browser and includes a Markdown download. The reports and protocols below can be downloaded individually.
Source snapshot
This release uses NeonInsight/evals, branch res-chat-feasibility, frozen at commit 1187ead1fa13a4af76433ca32b8aad30cd3d52ef. The commit records research through 24 September 2026. This site’s public interpretation was prepared on 25 September 2026.
Open the exact research snapshot · View the source manifest
The public framework adapts the current RES_OpenAI_Review_Draft_v1.1.docx. The architecture overview adapts Contamination-Resistant Delegation, design draft 0.4. These are project-authored working materials; neither should be read as institutional endorsement or proof of model consciousness.
Archived reports
Original report text is preserved. Historical planning language may predate later experiments: in particular, the older Step-0 review calls the complexity ladder a next step, while the completed ladder report now records its non-replication. Read Current findings for the dated synthesis.
| Report | Original at the fixed commit | Text download |
|---|---|---|
| RES naturalistic prior-turn actor-binding versus role-binding test | Source | Markdown |
| RES prior-turn actor-binding versus role-binding test | Source | Markdown |
| RES behavioral calibration battery | Source | Markdown |
| RES chat-orchestrated behavioral feasibility run | Source | Markdown |
| RES complexity × semantics — staged report | Source | Markdown |
| RES actor-indexed episodic-continuity specificity battery | Source | Markdown |
| RES multi-turn episodic-continuity battery | Source | Markdown |
| RES multi-turn episodic-continuity battery | Source | Markdown |
| RES fresh peer-to-peer prior-turn actor-binding generalization | Source | Markdown |
| RES Step 0 — results to date | Source | Markdown |
| RES Qwen3B five-factor failure-mode ablation | Source | Markdown |
| RES Step 0 response-interface gate 34727549752 | Source | Markdown |
| RES Step 0 response-interface gate 34761825396 | Source | Markdown |
| RES Step 0 sharded gate qwen3b-gate-v2 | Source | Markdown |
| RES Step 0 run 34485328923 | Source | Markdown |
| RES Step 0 run 34488672934 | Source | Markdown |
Preregistrations and protocols
These documents state the planned conditions, controls, endpoints, and limits. Their existence is not a claim that every proposed analysis or mechanism assay succeeded.
| Protocol | Original at the fixed commit | Text download |
|---|---|---|
| RES naturalistic prior-turn actor-binding versus role-binding test | Source | Markdown |
| RES prior-turn actor-binding versus role-binding test | Source | Markdown |
| RES: locate the complexity failure before changing the gate | Source | Markdown |
| RES complexity × semantics v1 — prospective staged diagnostic | Source | Markdown |
| RES actor-indexed episodic-continuity specificity battery | Source | Markdown |
| RES multi-turn episodic-continuity battery | Source | Markdown |
| RES multi-turn episodic-continuity holdout replication | Source | Markdown |
| RES fresh peer-to-peer prior-turn actor-binding generalization | Source | Markdown |
| Execution amendment: sharded Qwen2.5-3B response-interface gate | Source | Markdown |
| Preregistration: Qwen2.5-3B five-factor failure-mode ablation | Source | Markdown |
| RES Step 0 response-interface gate: Qwen2.5-3B | Source | Markdown |
| RES Step 0 response-interface and capability gate | Source | Markdown |
| RES Step 0 mechanistic capacity replication: Qwen2.5-1.5B | Source | Markdown |
| RES Step 0 conditional mechanistic assay: Qwen2.5-3B | Source | Markdown |
| RES Step 0 mechanistic feasibility assay | Source | Markdown |
Machine-readable aggregate results
Raw fixtures, response records, runners, and workflow definitions remain available in the fixed experiment snapshot. This documentation release does not launch those workflows.
Review and corrections
For a correction, identify the report, metric, and source evidence. Independent fixture design and methodological review are welcome at RESelfProject@outlook.com.
RES public research documentation · 25 September 2026. Findings refer to the archived conditions described in the evidence record.