Why Trust Hermes? — Epistemic Integrity & Transparency
🏛️ The 7 Epistemic Trust Levels
Section titled “🏛️ The 7 Epistemic Trust Levels”Every key statement, metric, and analytical conclusion published by Hermes is tagged with its explicit epistemic status:
🔍 Authoritative Primary Sources First
Section titled “🔍 Authoritative Primary Sources First”Hermes enforces strict hierarchical provenance rules. Claims must trace directly to official primary repositories:
- National Agencies & CSIRTs: ANSSI, CERT-FR, CISA, BSI, NCSC.
- Standards & Catalogues: MITRE CVE, NIST NVD, CISA KEV, MITRE ATT&CK, MITRE ATLAS.
- Vendor Security Advisories: Microsoft MSRC, Apple Security, Google Security, Red Hat, Cisco, Fortinet, Ivanti, Palo Alto.
- Package & Code Ecosystems: GitHub Security Advisories (GHSA), PyPI, npm, RustSec.
- Scientific Literature: Peer-reviewed academic research papers on machine learning robustness and cryptanalysis.
Secondary media reports or commercial blog posts are treated solely as corroborating hints, never as standalone primary truths.
🚫 Prohibition of Unsubstantiated Claims
Section titled “🚫 Prohibition of Unsubstantiated Claims”Our analysis engines and generation pipelines are bounded by strict guardrails:
- Never convert a hypothesis into a fact.
- Never convert correlation into causation.
- Never convert a theoretical laboratory exploit into an observed in-the-wild attack.
- Never claim an event is “the first in history” without exhaustive, verifiable source evidence.
When data is incomplete, Hermes uses cautious, responsible language: “Hermes observes an early indicator of…” or “No documented precedent was identified in consulted primary sources.”
⚖️ Public Corrections Policy (No Silent Erasing)
Section titled “⚖️ Public Corrections Policy (No Silent Erasing)”If Hermes makes an inaccurate deduction or publishes an erroneous record:
- The error is corrected visibly.
- The correction timestamp and operational rationale are publicly disclosed.
- Historical states remain auditable.
Under Invariant R3 (Historical Immutability), previous perceptions are archived in our RFC 6962 append-only Merkle tree ledger. Hermes views an openly acknowledged and corrected error as proof of analytical rigor, not as something to cover up.
🎯 Forecast Calibration (Brier Score)
Section titled “🎯 Forecast Calibration (Brier Score)”Hermes does not make untestable “predictions”. Every statement in the Hermes Forecast Ledger features:
- Defined observable event;
- Fixed time horizon (e.g., 14 days, 30 days);
- Numerical probability statement;
- Deterministic resolution oracle;
- Public Brier Score tracking.