Fact-Checking & Verification Policy

Orbis Signal does not operate a traditional human fact-checking desk or subscribe to third-party fact-checking APIs. Instead, we enforce verification through a structured production pipeline built around source traceability, epistemic labelling, and automated quality controls.

Every claim in a published briefing must trace back to a locked evidence record created during web-grounded research. Downstream analytical and writing stages cannot access the open web and cannot introduce new source URLs. This architectural boundary is our primary integrity mechanism.

Before any analytical assessment or briefing draft is produced, research stages collect atomic claims and tier-labelled sources from live web search. These are merged into a single evidence record per topic that is locked for the remainder of the pipeline.

Each evidence record contains:

• Claims — discrete, verifiable statements with unique identifiers
• Sources — HTTPS URLs with publisher, title, and tier classification
• Selected signals — the research threads chosen for the edition

No analytical judgment or published sentence may reference a source URL that is not present in this record. This rule is enforced in code, not merely in prompts.

After each analytical stage, automated validators check structural integrity:

• Every key judgment must cite valid claim identifiers from the evidence record
• Every source URL in the briefing and event articles must exist in the evidence record — invented or hallucinated URLs are rejected
• Confirmed claims must be supported by at least one tier-1 (primary) source
• Tier-3 sources alone cannot support confirmed or inference-level judgments
• Epistemic types and confidence bands must be present on key judgments

Before publication, each briefing passes through a two-phase quality control process:

• Phase A — Rule-based checks: sensationalism blocklists, URL integrity, structural compliance, and editorial prohibitions
• Phase B — Scorecard evaluation: seven dimensions scored 1–5, each requiring a minimum score of 3 to pass

The seven dimensions are:

• Clarity — structural hierarchy and unambiguous references
• Novelty — non-obvious insights and a non-trivial consensus gap
• Reliability — source quality and claim integrity
• Signal — strategic density and actionable watchlist items
• Strategic utility — decision relevance for the target reader
• Compression — brevity without loss of nuance
• Distinction — epistemic discipline and calibrated language

If a briefing fails quality control, one automatic rewrite is attempted with specific correction hints. If it still fails, the briefing is stored with a failed quality flag and is not published to subscribers.

Our verification model confirms traceability and editorial discipline, not absolute truth. We do not independently re-report events, contact primary sources for confirmation, or adjudicate disputed facts beyond what available evidence supports.

We rely on the quality of publicly available reporting and official releases. When sources conflict or evidence is thin, we label uncertainty explicitly rather than presenting contested claims as confirmed.

Readers who identify factual errors should follow our Corrections Policy.

Each pipeline run records layer-by-layer outputs, quality scores, provider attempts, and the editorial policy version in force. This audit trail supports post-publication review and correction workflows, even though production itself is fully automated.

Related: Sources & Attribution, AI Usage Disclosure.