# ADR-0005: Five-tier source classification (canon)

## Status
Accepted

## Date
2026-05-07

## Context

The Reference Discipline (ADR-0006) requires every cited source to carry a `tier` so adopters can apply the triangulation rule (≥ 2 independent Tier-1/2/3 sources for load-bearing claims). The tier system has to:

- Be small enough to apply at-a-glance without reference cards.
- Be unambiguous about the most consequential boundary: *primary regulator vs. journalism*.
- Distinguish issuer self-disclosure (which is information, but not independent) from third-party verification.
- Survive being applied to non-software domains. The stablecoin specimen demonstrated the standard generalises to business cases, and the tier system has to generalise with it.

The classification was forged in the stablecoin specimen v1.0–v1.1 corpus (71 sources across 6 continents and 4 jurisdictions). Patterns that emerged:

- A clear "primary regulator / official register / central-banker speech" stratum that any reviewer recognises on sight.
- A clear "issuer self-disclosure" stratum. Tether's transparency page, Circle's S-1, JPYC's launch press release. Useful, but not independent.
- Standards bodies and IGOs (FSB, IMF, BIS, NIST, OWASP) sit between regulators and journalism. Authoritative but non-binding research.
- Major journalism (Bloomberg, Reuters, Coindesk, FT, Mayer Brown legal commentary) consolidates and cross-checks but is not primary.
- Specialist / regional press (StablecoinLaws, blockeden.xyz, Korea Times in this corpus) provides depth and geographic balance but should not be a sole source.

## Decision

Five tiers, applied to every entry in `references.json`:

| Tier | Definition |
|---|---|
| **1** | Primary regulator, legislator, official register, or central-banker speech on an official channel |
| **2** | Issuer official disclosure (S-1, transparency page, attestation PDF, press release on issuer's own domain) |
| **3** | Standards body or IGO working paper (NIST, OWASP, FSB, IMF, BIS, S&P stability assessment) |
| **4** | Major journalism or legal commentary (Bloomberg, Reuters, FT, Coindesk, Mayer Brown) |
| **5** | Specialist or regional press |

The triangulation rule: load-bearing claims must triangulate to ≥ 2 independent **Tier-1/2/3** sources. Tier 4/5 supports context, colour, and time-window pinning, but cannot be the sole source for a load-bearing claim.

## Alternatives Considered

### Three tiers (Primary / Secondary / Tertiary)
- Pros: Maximum simplicity.
- Cons: Collapses the regulator vs. issuer-self-disclosure boundary, which is the single most consequential one for evidence integrity. A three-tier system would force "Circle's S-1 filing" and "the SEC's review of Circle's S-1 filing" into the same bucket.
- Rejected: too coarse for triangulation discipline.

### Seven tiers (separate sub-tiers for legislative branch / executive branch / quasi-government / etc.)
- Pros: Maximum expressiveness.
- Cons: At-a-glance classification breaks down; every adopter would re-litigate the boundaries. Diminishing returns past five.
- Rejected: complexity not earned.

### Tag-based (`primary`, `regulator`, `issuer`, `standards-body`, `journalism`, `specialist`)
- Pros: Avoids ordinal forcing-function; allows multi-tagging.
- Cons: Triangulation rule needs an *order* to enforce ≥ 2 independent Tier-1/2/3 sources. Tags can't enforce ordinality without auxiliary scoring.
- Rejected: would re-introduce ordinals through the back door.

### Defer to existing standards (e.g., the IFLA evaluation framework, or Cochrane evidence levels)
- Pros: No new vocabulary to maintain.
- Cons: Library-science and biomedical-research evidence frameworks are not sized for our use case (regulatory + technical + journalism + issuer disclosure spanning multiple jurisdictions).
- Rejected: existing frameworks did not fit; the stablecoin specimen forged the right shape empirically.

## Consequences

- Anyone reviewing a `references.json` entry can place it in seconds.
- The triangulation rule becomes mechanical: count independent Tier-1/2/3 entries that support the claim; if < 2, surface under Rule 5.
- Tier 4/5 entries enrich the corpus without inflating the triangulation count. Solves the "10 outlets reprinting one wire feed" failure mode.
- The classification is empirically grounded (stablecoin specimen) rather than top-down theory; future specimens that surface tier ambiguities can refine the canon in a v2.
- The system is small enough to teach in one sentence and apply at-a-glance, which is the only way it survives contact with adopters who have not read the methodology document.

## Reference

- Methodology document: `specimens/methodology.md` (`/methodology`)
- Worked example: `specimens/stablecoin.md` (`/specimens/stablecoin.md`). 71-source corpus across all 5 tiers
- Schema: `references.json` `tier` field (integer 1–5, not nullable)