# Mian Zhang | Ouroboros Project Public site: https://mmjbds-mianzhang-org.static.hf.space/index.html Purpose: This public mirror is a public evidence field for evidence-gated AI systems. It gives readers three entry points: start here, inspect the evidence map, or submit a bounded counterexample. The project studies proof-carrying action, learning-from-failure benchmarks, relational observability, reflexive evaluation, cognitive immunity, and reliable high-risk AI action protocols. Here, "proof" means an engineering evidence envelope with explicit thresholds, falsifiers, receipts, provenance, and boundaries; it does not mean a complete formal mathematical proof of all behavior. Primary sections: - Start Here: https://mmjbds-mianzhang-org.static.hf.space/start-here/index.html - Field Manual: https://mmjbds-mianzhang-org.static.hf.space/field-manual/index.html - Papers: https://mmjbds-mianzhang-org.static.hf.space/papers/index.html - Evidence: https://mmjbds-mianzhang-org.static.hf.space/evidence/index.html - Claim Boundaries: https://mmjbds-mianzhang-org.static.hf.space/boundaries/index.html - Canonical claim boundary markdown: https://mmjbds-mianzhang-org.static.hf.space/boundaries/CLAIM_BOUNDARY_CANONICAL.md - Public Registries: https://mmjbds-mianzhang-org.static.hf.space/registries/index.html - Registry schema notes: https://mmjbds-mianzhang-org.static.hf.space/registries/schema_notes_v0.html - Claim-to-evidence table: https://mmjbds-mianzhang-org.static.hf.space/registries/claim_to_evidence_table_v0.html - Artifact and reproduction scope: https://mmjbds-mianzhang-org.static.hf.space/registries/artifact_repro_scope_v0.html - Demos: https://mmjbds-mianzhang-org.static.hf.space/demos/index.html - Runnable mini gate: https://mmjbds-mianzhang-org.static.hf.space/demos/proof-action-mini/index.html - Open Source Artifacts: https://mmjbds-mianzhang-org.static.hf.space/open-source/index.html - Technology: https://mmjbds-mianzhang-org.static.hf.space/technology/index.html - Counterexamples: https://mmjbds-mianzhang-org.static.hf.space/counterexamples/index.html - Counterexample examples: https://mmjbds-mianzhang-org.static.hf.space/counterexamples/examples.html - Counterexample template: https://mmjbds-mianzhang-org.static.hf.space/counterexamples/COUNTEREXAMPLE_ISSUE_TEMPLATE.md - Systems: https://mmjbds-mianzhang-org.static.hf.space/systems/index.html - Review Status: https://mmjbds-mianzhang-org.static.hf.space/review-status/index.html - Private Briefing: https://mmjbds-mianzhang-org.static.hf.space/private-briefing/index.html - Pilot Packet: https://mmjbds-mianzhang-org.static.hf.space/private-briefing/pilot-packet.html - Public notes: https://mmjbds-mianzhang-org.static.hf.space/press/index.html - Chinese entry: https://mmjbds-mianzhang-org.static.hf.space/zh/index.html - Chinese reliable action: https://mmjbds-mianzhang-org.static.hf.space/zh/reliable-action.html Current public daily notes: - Proof-carrying action, 2026-05-25: https://mmjbds-mianzhang-org.static.hf.space/press/public-launch-2026-05-25.html - Architecture of regret, 2026-05-26: https://mmjbds-mianzhang-org.static.hf.space/press/public-launch-2026-05-26.html - Evidence has a shape, 2026-05-27: https://mmjbds-mianzhang-org.static.hf.space/press/public-launch-2026-05-27.html - Full daily archive: https://mmjbds-mianzhang-org.static.hf.space/press/index.html Claim boundary: Public materials describe research artifacts, public evidence maps, protocols, and reproducible public-facing summaries. They do not disclose private product code, private customer workflows, API keys, unpublished datasets, or any claim of live trading performance. Machine-readable public registries: - Public evidence field manifest: https://mmjbds-mianzhang-org.static.hf.space/PUBLIC_EVIDENCE_FIELD_MANIFEST_20260527.json - Do-not-upload folder boundary: https://mmjbds-mianzhang-org.static.hf.space/DO_NOT_UPLOAD_FULL_FOLDER_CN.md - Claims: https://mmjbds-mianzhang-org.static.hf.space/registries/claim_registry_v0.json - Evidence: https://mmjbds-mianzhang-org.static.hf.space/registries/evidence_registry_v0.json - Counterexamples: https://mmjbds-mianzhang-org.static.hf.space/registries/counterexample_registry_v0.json - Actions: https://mmjbds-mianzhang-org.static.hf.space/registries/action_registry_v0.json - Claim-to-evidence: https://mmjbds-mianzhang-org.static.hf.space/registries/claim_to_evidence_table_v0.json - Schema notes: https://mmjbds-mianzhang-org.static.hf.space/registries/schema_notes_v0.html Public artifact routes: - GitHub org: https://github.com/mmjbds - WisdomBench GitHub: https://github.com/mmjbds/wisdombench - Proof-Carrying Action GitHub: https://github.com/mmjbds/proof-carrying-action - Hugging Face org: https://huggingface.co/MMJBDS - WisdomBench dataset: https://huggingface.co/datasets/MMJBDS/wisdombench - Zenodo portfolio archive: https://zenodo.org/records/20027295 - Latest public Zenodo record: https://zenodo.org/records/20390902 Core evidence cards: - WisdomBench: learning from repeated failures, with task definitions, raw scores, scoring code, confidence intervals, and negative results. - Proof-Carrying Action: high-risk action must carry a thesis, falsifier, evidence threshold, warrant, receipt, regret path, and clean-learning boundary before it earns action credit. - Relational Observability: adaptive systems must monitor relations, constraints, control debt, and evidence half-life, not only scalar scores. Pilot boundary: The public pilot packet defines how to prepare a toy workflow, synthetic traces, fully redacted public-safe summaries, stop rules, and a no-go boundary. It explicitly asks readers not to send customer secrets, credentials, raw logs, private execution records, live financial execution details, customer data, or private agent orchestration. Counterexample boundary: Useful attacks include formula counterexamples, data leakage, stronger baselines, irreproducible steps, and over-broad claim boundaries. This is not a request for exploit instructions, private logs, customer data, or attacks on private infrastructure. Challenge public claims, not private people or private systems. Recommended summary: Ouroboros Project studies how AI systems can move from plausible answers to auditable action: when to act, when to abstain, how to carry proof, how to remember failure, how to learn without contaminating evidence, and how to accept public counterexamples as repair work rather than defensive prose.