By 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:
- Source reliability (A-F) — historical trustworthiness of the provider
- Information credibility (1-6) — believability of this specific claim
- 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)
| Grade | Meaning | Business Examples |
|---|---|---|
| A | Completely Reliable | Government filings, ERP/payment APIs, signed contracts |
| B | Usually Reliable | Premium data vendors (D&B, ZoomInfo), reputable news, corporate websites |
| C | Fairly Reliable | CRM data (rep-entered), LinkedIn profiles, trade publications |
| D | Not Usually Reliable | Job aggregators, unverified reviews, general web scraping |
| E | Unreliable | Lead farms, known disinformation, clickbait content |
| F | Cannot Be Judged | New sources with no track record (not "bad" — unknown) |
Information Credibility (1-6)
| Grade | Meaning | Business Application |
|---|---|---|
| 1 | Confirmed | Corroborated by 2+ independent sources |
| 2 | Probably True | Logical, consistent with known context, unconfirmed |
| 3 | Possibly True | Plausible but lacks corroboration |
| 4 | Doubtful | Possible but illogical or unsupported |
| 5 | Improbable | Contradicted by other information |
| 6 | Cannot Be Judged | No basis for evaluation |
Important: F and 6 mean "unknown," not "bad."
Lead Scoring Weight Matrix
| Code | Interpretation | Weight | Action |
|---|---|---|---|
| A1 | Confirmed Fact | 1.0 | High-priority alert |
| B1/A2 | High Confidence | 0.9 | Push to CRM |
| B2/C1 | Actionable Intel | 0.75 | Pipeline + verification |
| C3 | Weak Signal | 0.4 | Watchlist only |
| D4/E5 | Noise/Conflict | 0.0 or negative | Suppress |
| F6 | Unknown | 0.0 (neutral) | Sandbox for pattern matching |
Confidence Framework
Display two separate outputs for each assessment:
Probability (Likelihood)
| Term | Probability |
|---|---|
| Remote Chance | ≤5% |
| Highly Unlikely | 10-20% |
| Unlikely | 25-35% |
| Realistic Possibility | 40-50% |
| Likely / Probable | 55-75% |
| Highly Likely | 80-90% |
| Almost Certain | ≥95% |
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):
- Tag as F6 initially
- Zero-trust weighting (neutral, not negative)
- Shadow score in background
- Graduate to C/B after N validated signals