source-grading

# Source Grading: The Admiralty System Every claim in research or investigation output must be graded on two independent axes. ## Axis 1: Source Reliability (Who provided this?) | Grade | Label | Criteria | Examples | |-------|-------|----------|----------| | **A** | Completely Reliable | Proven track record; legally accountable for falsehood | Court records, SEC filings, government audit reports, enforcement actions | | **B** | Usually Reliable | Institutional reputation at stake; editorial/peer review | Academic peer-reviewed studies, major investigative journalism, GAO/OIG reports | | **C** | Fairly Reliable | Domain expertise but potential bias; less rigorous review | Trade press, industry reports, state agency press releases | | **D** | Not Usually Reliable | Self-interested; no independent verification | Company press releases, PR, marketing, self-reported data | | **E** | Unreliable | Known bias, history of inaccuracy, no accountability | Social media, anonymous forums, unverified tips | | **F** | Cannot Be Judged | Source reliability cannot be assessed | New/unknown source, automated data with unknown provenance | ## Axis 2: Information Credibility (Is this claim true?) | Grade | Label | Criteria | |-------|-------|----------| | **1** | Confirmed | Independently verified by 2+ sources from different domains | | **2** | Probably True | Consistent with known data; one independent confirmation | | **3** | Possibly True | No confirmation, no contradiction; plausible but unverified | | **4** | Doubtful | Inconsistent with some known data; requires assumptions | | **5** | Improbable | Contradicted by known data or independent sources | | **6** | Cannot Be Judged | Not enough information to assess truth value | ## Combined Grade Format Write grades as `[Grade: X#]` inline with claims: - "ABI owes $14.3M in wage theft" **[A1]** — Crain's (A), confirmed by DOL records (1) - "The entity has no web presence" **[F3]** — absence of evidence (F), possibly true (3) - "$1.68B in Medicaid billing" **[A1]** — CMS official data (A), cross-verified (1) ## Rules 1. **Grade source and information independently.** A completely reliable source (A) can report wrong information (4-5). An unreliable source (E) can provide independently confirmed information (1). 2. **Upgrade credibility when sources converge.** Three C-grade sources independently reporting the same fact → credibility upgrades toward 1-2. 3. **Downgrade credibility when sources share upstream.** Three sources citing the same original report = ONE source, not three. 4. **Absence of evidence is not evidence of absence.** "No enforcement action found" is [B3] at best. 5. **Self-interested sources get automatic D.** Company statements about their own compliance, self-attestation, marketing — never above D unless independently confirmed. 6. **Data from our own analysis is [DATA].** A-reliability (official source), credibility depends on query correctness. ## Integration with Provenance Tags | Old Tag | Admiralty Equivalent | |---------|---------------------| | [DATA] | [DATA] (keep — indicates own analysis) | | [SOURCE: url] | [Grade: X#] with the URL | | [INFERENCE] | [INFERENCE: X# + X#] — cite grades of premises | | [UNCONFIRMED] | Any claim graded 3-6 on credibility | ## When to Grade - **Always** during investigation or OSINT research (the `research` skill routes it here) - **Always** in analysis documents for handoff (attorneys, journalists, government) - **Optional** in informal conversation or brainstorming - **Required** before any lead enters the evidence matrix of the `/analyze` hypotheses lens