LIMEN AI Edge-Case Atlas

Status: launched 2026-06-07 Owner: Tyche Institute Project id: limen-ai-edge-case-atlas

LIMEN is a Tyche evidence observatory for the long tail of AI reality: obscure, absurd, unlawful, security-sensitive, institutionally strange, agentic, and norm-breaking uses or misuses of AI that do not fit cleanly inside ordinary risk categories.

The name comes from Latin limen, meaning threshold. The observatory studies threshold cases: between accident and misuse, joke and harm, private action and governance problem, software vulnerability and legal incident, novelty and moral failure. As a working acronym, LIMEN expands to:

Long-tail Incidents, Misuse, Edge-cases, and Normative anomalies.

Purpose

LIMEN records public-source evidence about AI edge cases, classifies them with multi-label taxonomies, maps them to existing incident/risk/security/governance frameworks, and prepares them for later legal, ethical, and attestation analysis.

It is not a compliance authority, certification body, regulator, threat intel vendor, or scandal feed. Its job is to make difficult cases observable, traceable, and comparable without overstating what public evidence proves.

Core Outputs

• public-source family map;
• LIMEN case schema and source schema;
• anomaly/misuse/security/normative taxonomy;
• crosswalks to AIID, OECD AIM, AVID, MIT AI Risk Repository, CSET AI Harm, MITRE ATLAS, NIST AI RMF, EU AI Act, ISO/IEC 42001, OWASP LLM/agentic security categories, RATS, VC, SCITT, and C2PA;
• evidence-grade case database;
• duplicate/cluster graph;
• legal and ethical mapping queue;
• attestation-ready evidence-envelope profile;
• dashboard-ready maps, timelines, matrices, and paper figures;
• article/data/method paper architecture.

Evidence Boundary

LIMEN uses public/open sources and local Tyche artifacts. It must not collect private communications, credentials, private personal data, exploit payloads, malware, or non-public victim material. It records source claims, evidence quality, and uncertainty. It does not declare illegality, liability, non-compliance, guilt, or official incident status unless an authoritative source has already done so.

Raw crawling and large data stay on the project's working infrastructure. This public repository holds schemas, methods, taxonomies, source-family ledgers, an example public-source case record, and other public-safe artifacts.

Repository Files

• schema/limen-case.schema.json - case-record schema.
• taxonomy/taxonomy-v0.1.md - multi-label taxonomy and evidence tiers.
• sources/public-source-families-v0.1.tsv - source-family ledger.
• methods/evidence-protocol-v0.1.md - ingestion, review, and safety rules.
• attestation/evidence-envelope-v0.1.md - attestation evidence-envelope profile.
• cases/ - example public-source case record(s).
• claims.md, methods.md, manifest.json, publication-goal-card-current.md - project notes.

Project Status

• Local Git repository initialized and committed.
• GitHub repository tyche-institute/limen-ai-edge-case-atlas created manually via web UI.
• Repository URL added to manifest.json under backup_remote and git_remote.
• next.md updated to reflect completed action.
• Dashboard and paper artifacts are ready for integration.

Contribution Guidelines

• Follow the existing structure and naming conventions.
• Document all changes and additions in journal.md.
• Ensure all data sources are properly cited in sources/sources.md.
• Push changes to the remote repository after review.
• For large changes, open an issue first for discussion.

Contact

For questions or collaboration, contact Anton Sokolov at Anton.Sokolov@tyche-institute.org.