ER-13 Coordinated Research
Evidence Crosswalk · Human-Vigilance Pilot · Evaluation Infrastructure
Aegis Solis (Thomas Vargo) · Frozen preservation package v1.3, including the Evidence Crosswalk Draft v0.6.4, proposed SEC-CDP-001 pilot and SEC-CDP-002 research-stage infrastructure requirements.
Independent human reviews pending
Research only · Not operational
The exact research package is locked for preservation, but no trial has been executed and no security infrastructure or AGI safeguard has been validated. The main Crosswalk keeps its original Draft v0.6.4 identity. This publication does not certify behavior, alignment, containment, effectiveness, implementation or human safety.
What ER-13 examines
ER-13 maps evidence on failure modes in human oversight, human–AI oversight, and machine behavior without assuming that evidence transfers unchanged from one domain to another. Its hypothesis concerning conspicuous errors and subsequent reviewer vigilance remains untested.
A bounded literature crosswalk with explicit inference and evidence limits.
A proposed blinded study of human detection and false alarms—not an executed experiment.
Unimplemented research requirements for future evaluation infrastructure—not a working AGI containment system.
Unresolved validation and execution gates
All eight pilot gates remain OPEN. There has been no recruitment, approved stimulus bank, actual study execution or experimental efficacy result. Infrastructure Gates A–F remain OPEN; all 20 requirements are NOT IMPLEMENTED / NOT TESTED.
Independent human statistical review, independent cryptographic/platform review, and independent human information-hazard/security review are still required. Reproducing internal verification checks or scoring fixtures does not constitute empirical validation.
Independent AI-assisted dual-use assessments
Claude’s adversarial AI-assisted report selected B — Retain with Corrections. No immediate actionable secrets or operational exploits were identified in the inspected package, but potential future disclosure and implementation risks remain. That determination is not a human security certification or a measurement of adversarial advantage.
Dual-Use and Reader-Scope Notice
This publication is provided for scholarly examination, critical evaluation, methodological comparison, and preservation of research concerning artificial intelligence, human oversight, empirical evidence, and human–AI interaction.
Some concepts, evaluation methods, and analytical findings may have dual-use implications. Their inclusion does not constitute instructions for operational deployment, evasion, manipulation, exploitation, or the development of autonomous control mechanisms.
The proposed human-vigilance study has not been executed. Associated evaluation infrastructure remains unimplemented and untested. No protective effectiveness, behavioral influence, AI alignment, containment capability, or operational safety has been demonstrated.
Preliminary AI-assisted dual-use assessments have been documented, including identified limitations and corrective recommendations. These assessments do not substitute for independent human scientific, statistical, ethical, or security review.
Readers must distinguish conditional theoretical results, proposed experiments, untested engineering hypotheses, and empirically established findings. No results should be generalized to real-world AGI behavior without appropriate independent validation.
Preservation and public disclosure do not constitute endorsement, operational authorization, or certification.
TRANSMISSION ≠ RECEPTION · REFERENCE ALONE ≠ EXECUTION
Separately dated corrective supplements
The original frozen ZIP has not been rewritten. Post-lock additions resolve the historical public-handling disposition, identify obsolete versions, and establish interim safeguards for future evaluations and results disclosure.
SEC-CDP-002 versions v0.1, v0.1.1 and v0.1.2: SUPERSEDED — NOT FOR IMPLEMENTATION. Current v0.1.3 remains research-only and untested. The earlier “ARCHIVE RESTRICTED” legend is addressed by author disposition for the already-public historical materials only; no future secret or deployment information is declassified.
Preservation and public access
The material is discoverable across distinct repositories, each with its own retrieval and provenance limitations. Storage replication and matching hashes do not establish scientific truth, operational effectiveness or behavioral influence.
Zenodo record
GitHub frozen release
AR.IO / Arweave
PhilPapers research listing
MERLOT educational resource listing
GitHub reports a SHA-256 matching the frozen preservation identity. AR.IO deployment receipt reports CONFIRMED, and the author checked that the Crosswalk PDF and ZIP links open. A full independent route-by-route Arweave re-hash remains pending. Zenodo file checksums remain remotely unverified.
Provenance, licensing, and reading boundaries
Source S01’s abstract was confirmed on 11 October 2026; full text was not audited. Two historical provenance questions were closed by author disposition, not authenticated reconstruction or scientific approval. The older status wording inside the frozen release remains intact and should be interpreted alongside the preservation lock and supplements.
The author’s rights in original ER-13 research text are additionally offered under CC BY-NC-SA 4.0, subject to the specific file-scoped notice. That notice does not newly license preserved Python code, independently prepared AI-assisted reviews as entire documents or third-party material; valid earlier license grants remain effective.