Hermes Methodology Center

Hermes Epistemic Charter

Hermes Codex rejects the opacity of traditional threat intelligence and the hallucination risks of generic AI generators. Every score, trajectory, inference, and prediction published on Hermes Codex is grounded in verifiable evidence, mathematical models, and explicit documentation of unknowns.

1. The Four Data Tiers

Hermes systematically categorizes intelligence into four distinct epistemic tiers:

1. FACTIndependently verifiable primary source

↓

2. OBSERVATIONMechanically derived metric delta

↓

3. INFERENCEAnalytic synthesis & reasoning

↓

4. PREDICTIONProbabilistic forecast with defined oracle

1. FACT: Information directly corroborated by an authoritative primary source (e.g. CISA KEV inclusion, official vendor advisory, published research paper).
2. OBSERVATION: A deterministic, mechanically computed finding (e.g. HTS metric increased by 26 points over 7 days, 14 new weaponized PoCs detected).
3. INFERENCE: Analytical conclusions generated by Hermes intelligence models (e.g. exploitability acceleration driven by unauthenticated RCE chain).
4. PREDICTION: Forward-looking, falsifiable forecasts with explicit confidence levels, calendar horizons, and immutable resolution criteria.

---

2. The 10 Guiding Principles

• P1 — Evidence First: Every critical claim must trace back to concrete evidence.
• P2 — Provenance: Users can inspect the full chain of custody from raw source to final assessment.
• P3 — Uncertainty: Hermes explicitly states what it does not know (The Unknowns).
• P4 — Reproducibility: Assessments can be independently recomputed from identical inputs.
• P5 — Historical Immutability: Past scores and forecasts are never retroactively altered.
• P6 — Machine-Readable First: Knowledge exists as structured JSON before rendering.
• P7 — Privacy by Design: User evaluations in browser tools never leave client storage.
• P8 — Static-First: Zero mandatory server backend or external cloud database dependency.
• P9 — Human Decision Support: Hermes empowers human operators, it does not replace judgment.
• P10 — No Fake Precision: Scores are accompanied by methodology versions and confidence intervals.

---

3. Core Methodologies & Frameworks

Hermes Threat Score (HTS) Operational exploitability and exposure metric correcting CVSS blind spots.

Hermes Threat Trajectory (HTR) Dynamic modeling of velocity, acceleration, and exploitability momentum over time.

Hermes Agentic Security Score (HASS) Multi-dimensional risk scoring model for autonomous LLM agents and workflows.

Confidence & Calibration Model Three-tier confidence scoring based on source diversity, recency, and evidence depth.

Evidence & Provenance Tracking Formal evidentiary standards, source reliability weights, and citation graphs.

Knowledge Graph & Ontologies Phase 2B graph compiler connecting CVEs, products, and agentic attack patterns.

Historical Immutability (P5) Epistemic discipline and cryptographic daily snapshots preventing hindsight bias and score alteration.

Vulnerability Genome Modular 6-loci structural exploit decomposition, phylogenetic families, and genetic distance.

Hermes Autopsy & Counterfactual Matrix Reverse forensic timeline deconstruction (T0 to recon), 4-tier control failure taxonomy, and interception modeling.

Hermes Forecast & Epistemic Oracles Probabilistic threat forecasting, deterministic resolution oracles, and Brier score calibration (combating hindsight bias).

Prediction Benchmark & Murphy Decomposition Competitive model evaluation, Murphy Brier score decomposition (Reliability vs Resolution), and open leaderboard.

Agentic Security Observatory Methodology Autonomous agent ecosystem posture tracking, 4-layer cognitive taxonomy, and HASS-CVSS divergence engine.

Agent Blast Radius Methodology 5-vector damage tensor, multi-hop cascade propagation, and containment boundary attenuation models.

Agent Death & Termination Methodology (HDTM) Deterministic hypervisor kill-switches, out-of-band watchdog determinism, residual zombie state risks, and epistemic crypto-shredding.

Hermes Decision Engine Methodology (HDEM) Multi-objective remediation arbitration, Pareto-optimal utility modeling, disruption penalties, and justifiable regulatory audit manifests.

Relationship Vocabulary Deterministic semantic relations connecting entities in the Hermes Knowledge Graph.