---

name: intelligence-source-grading description: Apply formal intelligence tradecraft to business data systems. Separate source reliability from information credibility, handle uncertainty explicitly, and communicate confidence appropriately. Covers the Admiralty Code, confidence frameworks, and signal triangulation.

Intelligence Source Grading

Apply formal intelligence tradecraft to business data systems. Separate source reliability from information credibility, handle uncertainty explicitly, and communicate confidence appropriately.

Core Principle

Never treat all data as epistemologically equal. A verified government contract and an unverified job posting require different handling in scoring algorithms.

Always separate three layers:

1. Source reliability (A-F) — historical trustworthiness of the provider
2. Information credibility (1-6) — believability of this specific claim
3. Analytical confidence — soundness of your final judgment

Quick Reference: The Admiralty Code

Combine source grade (letter) with credibility grade (number) to form codes like B2 or C3.

Source Reliability (A-F)

Information Credibility (1-6)

Important: F and 6 mean "unknown," not "bad."

Lead Scoring Weight Matrix

Confidence Framework

Display two separate outputs for each assessment:

Probability (Likelihood)

Analytical Confidence (High/Moderate/Low)

High: Multiple Grade A/B sources, minimal conflict, stable topic Moderate: Credible sources but limited corroboration or some bias risk Low: Single source, Grade D/E quality, high conflict, or volatile situation

Critical rule: Never combine probability and confidence in the same sentence.

✅ "It is likely Account X is entering a buying cycle. We have moderate confidence based on hiring signals and procurement activity."

❌ "We are highly confident this is likely to happen."

Signal Triangulation

Corroboration vs. Repetition

Corroboration: Independent collection paths converge on the same fact Repetition: Same claim copied across dependent sources (not confirmation)

Ten news articles citing one press release = 1 signal, not 10.

The Rule of Threes

Signals corroborated across three different categories achieve Grade 1:

• HUMINT: CRM notes, conversations, social media
• TECHINT: DNS, technographics, product usage
• OSINT: Press, news, job boards
• FININT: Filings, funding, procurement

Signals corroborated within one category achieve Grade 2 at best.

Handling Contradictions

When signals conflict (e.g., "hiring aggressively" + "layoff announcement"), do not average to neutral. Flag as divergence requiring investigation.

Display: "Market Divergence Detected. Manual Review Recommended."

Implementation Checklist

Data Model: Store each signal as an Observation with:

• Source type + instance
• Original URL/reference
• Timestamp + observed_at
• A-F reliability grade
• 1-6 credibility grade
• Independence cluster ID (for deduplication)

Scoring Engine:

• Convert grades to weights
• Apply recency decay
• Discount duplicate sources (don't stack linearly)
• Apply conflict penalty
• Output: score, likelihood term, confidence level

New Source Handling (The "F" Protocol):

1. Tag as F6 initially
2. Zero-trust weighting (neutral, not negative)
3. Shadow score in background
4. Graduate to C/B after N validated signals