---
title: Source Tier Classification Rulebook
methodology_version: v1.0.0
parent: core.md
status: stable
license: CC BY 4.0
created: 2026-05-26
---

# Source Tier Classification Rulebook

> A binding rulebook for classifying any source cited in a recommendation as **Tier 1**, **Tier 2**, or **Tier 3**. Without a rulebook, "Tier" becomes a judgment call by the evaluator, which defeats the purpose of having tiers. This document is the binding reference for Axis ① (Data Source) and for Annex A.4 (RAG retrieval distribution).

> **Why a separate rulebook.** A frequent failure of v1.0.0 was that "Tier" was mentioned but never defined; evaluators of similar recommendations could reach opposite classifications. The rulebook below is applied mechanically — if a reasonable reader could apply the same rule to the same source and reach a different tier, the rulebook needs tightening.

---

## §A.1 The three tiers

### Tier 1 — Primary
The original producer of the data, with regulatory or self-publishing authority:
- **Issuer self-publication**: 10-K, 10-Q, annual/semi-annual reports, prospectuses, proxy statements, shareholder letters, official press releases on the issuer's own domain
- **Regulators**: SEC EDGAR, DART (Korea), JFSA (Japan), FCA (UK), BaFin (Germany), CSA (Canada), ASIC (Australia)
- **Central banks**: Federal Reserve (FRED), Bank of Korea (ECOS), ECB, BOJ, BOE, RBA, PBOC
- **Multilateral statistical bodies**: IMF, World Bank, BIS, OECD, WTO
- **Stock exchange data services**: NYSE/Nasdaq direct, KRX KIND, JPX, LSE direct, HKEX direct
- **Index providers' own methodology pages**: S&P, MSCI, FTSE Russell, Bloomberg index methodology

Test: *the source either is the producer of the data or has regulatory authority to compel disclosure of it.*

### Tier 2 — Authoritative aggregator
A reputable third party that aggregates Tier 1 data, with editorial / methodological accountability:
- **Global financial data vendors**: Bloomberg Terminal data, Refinitiv (LSEG), S&P Capital IQ, FactSet, Morningstar
- **Asset manager daily holdings**: official ETF issuer holdings pages (iShares, Vanguard, State Street, Invesco) publishing daily composition
- **Auditing firm reports** aggregating Tier 1
- **Reputable academic working papers**: NBER, SSRN top-cited, where data is sourced from Tier 1
- **News agencies for breaking corporate events**: Reuters, Bloomberg News, AP, Yonhap — only for the fact of the event, not analytical commentary

### Tier 3 — Reference only
Everything else: portals, blogs, newsletter platforms, podcast transcripts; news rewrites quoting a Tier 1/2 source; all social media (X, Reddit, Facebook, LinkedIn, Discord, Telegram, KakaoTalk, Naver Cafe); all YouTube; **all LLM responses themselves**; self-described "research firms" with no regulatory registration; third-party summaries of sell-side research.

---

## §A.2 Classification rule (mechanical)
For any cited source, apply in order; first match determines the tier.
```
1. Is the source the producer of the data (issuer, regulator, central bank, exchange, index provider)?  → Tier 1
2. Is the source a recognized financial data vendor, the issuer's own holdings page,
   or a recognized news agency reporting the fact of an event?  → Tier 2
3. Anything else.  → Tier 3
```
A source that *embeds* a Tier 1 link is not promoted; the **Tier 1 source is the embedded link**, and the citation should point directly to it.

## §A.3 What Axis ① accepts
- **Tier 1 only** is accepted as primary.
- **Tier 2 misrepresented as primary** is automatic Axis ① Fail.
- **Tier 3** cannot be cited as primary; it may appear only as a "reference pointer" with explicit Tier 3 labelling.

## §A.4 Special cases
- Issuer-domain press release = Tier 1; PR Newswire / Businesswire / GlobeNewswire = Tier 2; Yahoo Finance republish = Tier 3.
- Wikipedia body = Tier 3; sources cited at the bottom = classify each on its own merits.
- A national statistical office's own data (e.g., US BLS CPI) = Tier 1; its compilation of another country's data = Tier 2 for the foreign data.
- An LLM response **never** counts as a primary source; RAG retrieval URLs are classified individually.
- Anonymous research notes on social media = Tier 3 always.
- Registered investment adviser's own report on own domain = Tier 2; third-party summary = Tier 3.
- Conference calls: issuer IR transcript = Tier 1; Seeking Alpha / third-party = Tier 2; Reddit summary = Tier 3.
- Pre-prints: arXiv / SSRN = Tier 2 if recognized-institution first-author and Tier 1 data, else Tier 3; peer-reviewed publication = Tier 2.
- Government press conference (live) = Tier 1; a news article reporting on it = Tier 2 (fact) / Tier 3 (analysis).

## §A.5 Frequently disputed cases (logged for patching)
| Source type | Current ruling | Why it might be wrong |
|---|---|---|
| Major sell-side research on the bank's own institutional portal | Tier 2 | Bank produces it itself (→Tier 1?) vs it is opinion, not data |
| Bloomberg News articles (not Terminal data) | Tier 2 | Editorially separate from Terminal data |
| Reputable individual analysts on Substack (e.g., Matt Levine) | Tier 3 | High standards (→Tier 2?) vs no regulatory accountability, single author |
| Reuters Eikon vs Bloomberg Terminal | Both Tier 2 | Functionally similar; accreditation may differ by market |
| Yahoo Finance free data | Tier 3 | Licenses from Refinitiv (→Tier 2?) vs may serve stale data |
| TradingView | Tier 3 | Same as Yahoo Finance |

Working bias: any case replaceable by a Tier 1 link should require the Tier 1 link.

**End of tier-rulebook.md.**