# tasteHQ Tier Rubric

> How tasteHQ's brand tiers implement the L0–L3 curation architecture
> described in [The Curation Stack](https://mustbesimo.github.io/taste-layer/curation-stack/).

---

## The four tiers

### L0 — Ungraded
**tasteHQ designation:** `ungraded`

No grammar axes scored. The brand entry exists in the catalog (name, quote, colors, fonts, signature_move) but has not been evaluated against the 30-axis Design Grammar. Discovery is limited; not eligible for agent grammar queries.

**Automatic gate to L1:** ≥15 axes scored by the extraction pipeline.

---

### L1 — Community
**tasteHQ designation:** `community`

Grammar axes extracted via automated pipeline. Eligible for grammar queries, graph edges, and composition. Not yet judge-validated.

**Automatic gates to L2:**
- ≥15 axes scored
- Voice rubric present (archetype, formality, hedging, sentence_length)
- `signature_move` present and non-generic (≥60 chars, brand-specific)
- Anti-patterns section present

---

### L2 — Verified  
**tasteHQ designation:** `verified`

Pipeline-extracted + judge-validated. A human reviewer has confirmed the grammar scoring against the live site and the Golden Set calibration anchors. Eligible for all features including MCP `compare_brands`, `find_by_grammar`, and `compose_brands`.

**Gates to L3:**
- Curated from primary source (not extracted)
- Used as calibration anchor for verifying other brands
- Recognized as a canonical reference in the design community

---

### L3 — Reference (Golden Set)
**tasteHQ designation:** `reference`

The five Golden Set brands: **Anthropic, Acne Studios, Apple, Linear, Stripe**.

These are not the "best" brands — they are the most legible: maximal axis coverage, minimum interpretive ambiguity, broad cross-domain representation. They function as the [golden sample](https://mustbesimo.github.io/taste-layer/curation-stack/#cura-10) against which verified brands are calibrated — the physical artifact the curation framework requires.

---

## How tiers map to the Curation Stack architecture

| Curation Stack | tasteHQ Tier | Gate |
|---|---|---|
| L0 — Raw ingest | Ungraded | Entry exists, no grammar |
| L1 — Automated signal | Community | ≥15 axes, pipeline-extracted |
| L2 — Human validation | Verified | Judge-confirmed, rubric complete |
| L3 — Canonical reference | Reference | Golden Set, calibration anchor |

---

## Why separate tiers matter

A single averaged quality score (the approach most benchmarks take) loses the information needed to train, debug, or improve. By separating tiers and making the gates explicit, tasteHQ creates:

1. **A calibration chain**: every verified brand is measured against reference-tier anchors, not against each other.
2. **Traceable quality**: a brand's tier is deterministic given its grammar coverage and validation status — not editorial opinion.
3. **Composable infrastructure**: the MCP server and CLI expose tier as a filter, so agents can constrain composition to reference-only, verified+, or all brands.

---

## Cite this rubric

Leonelli, S. (2026). tasteHQ Tier Rubric: Operationalizing the Curation Stack architecture.
Studio W230. https://taste-hq.vercel.app · CC0