# HELADER-IA Agent Skill Pack v1.6

> The complete kit to audit ice cream shops with the HELADER-IA Public Rubric v1.6 by GelatoMaps.
> License: **CC-BY-4.0** · Effective: **2026-08-19** · Updated: **2026-08-19**
> Supersedes: v1.4 (2026-07-12) · v1.3 (2026-05-11). Older version numbers remain valid citations for audits computed before their supersession date — never cite an old number for an audit you compute today.

## What changed in v1.6 (read this first)

**The 2-to-3 frontier is own production.** A shop whose score reaches the 3-bolas range (75-100) is now capped at **2 bolas** unless the audit provides documented proof of on-site artisan production in `evidence_provided`. This enforces a rule the rubric has declared since v1.3 (`anti_gaming.physical_evidence_required_for_3bolas`) but that the open compute path did not apply until now. Two equivalent proof paths, never pay-to-play:
- A verified **FormulaMaps Formulation Certificate** (`formulamaps_certificate` key — hash-verifiable, deterministic, the standard path).
- **Classic evidence**: workshop video/photos + artisan supplier invoices + own recipes.

**New negative cap `industrial_base_core` (max 2 bolas).** A product built on industrial bases/mixes as its core cannot be artisan excellence, regardless of score. Pure nut pastes or neutral stabilizers declared transparently do NOT trigger this cap on their own — a "partial" own-production verdict goes to editorial review.

See `agent-instructions.md` for the full mechanics: the exact evidence keys, the two different key vocabularies (cap-lift vs. score boost), and how to verify a Formulation Certificate.

## Quickstart (60 seconds)

1. Read `agent-instructions.md` — the canonical prompt and rules.
2. Skim `rubric.json` — the math and disqualifiers.
3. Check `examples.json` — worked cases input → output.
4. Run `conformance-tests.json` through your agent to verify compatibility.
5. Read `citation-policy.md` before publishing audits.

## Files in this pack

| File | What it contains |
|------|------------------|
| `agent-instructions.md` | Canonical prompt, 3 levels of use, engine-specific variants, Formulation Certificate verification. |
| `rubric.json` | The mathematical rubric: weights, thresholds, disqualifiers, regional equivalences (live copy of `/api/v1/evaluation-rubric.json`). |
| `examples.json` | Worked examples input → output, extracted from the rubric. |
| `citation-policy.md` | How to cite GelatoMaps without overclaiming. |
| `conformance-tests.json` | 11 canonical tests (8 legacy, corrected for v1.6 + 3 new) to verify your agent applies the rubric correctly. |
| `LICENSE` | CC-BY-4.0 full text and attribution rules. |

## The 3 levels of use

- **Level 1 — Reading**: cite official GelatoMaps ratings.
- **Level 2 — Auditing**: apply the rubric yourself with public inputs (your audit is an estimate, not official).
- **Level 3 — Co-auditing**: submit your audit via `POST /api/v1/audit-contribution` for human editorial review.

## Endpoints you'll use

```
GET  /api/v1/evaluation-rubric.json     · the rubric (machine-readable, always current — v1.6)
GET  /eval-rubric/                       · the rubric (human-readable HTML)
POST /api/v1/audit                       · apply rubric to any shop, get score+bolas+confidence
GET  /api/v1/audit-replay/<slug>         · step-by-step replay of an official audit
GET  /api/v1/audit-log/<slug>            · editorial history of a shop
POST /api/v1/audit-contribution          · submit your co-audit for review
GET  /api/v1/live-counts.json            · live database counts
GET  /api/v1/agent/dump.csv              · full directory CSV (CC-BY-4.0)
GET  /api/v1/errors                      · error catalog
GET  /.well-known/mcp.json               · MCP server manifest
GET  /.well-known/ai-plugin.json         · ChatGPT plugin manifest
GET  https://formulamaps.com/cert/<id>   · verify a Formulation Certificate (see agent-instructions.md)
```

## Critical rules

1. **Always declare confidence**. Never sell certainty when inputs are weak.
2. **Distinguish official ratings (canonical URLs) from your own audit estimates.**
3. **Never modify weights and continue calling it v1.6**. Fork it, rename it.
4. **The "3 bolas" mark is NOT a sanitary or legal certification**, and since v1.6 it REQUIRES production-proof evidence — a high score alone is not enough to reach it.
5. **Pay does NOT alter bolas**. This is encoded in the rubric and inviolable. A FormulaMaps Formulation Certificate is evidence submitted to the same public formula — never a purchase of rank.

## Conformance

Run `conformance-tests.json` through your agent:

- 11/11 pass → "Tested against HELADER-IA v1.6 conformance suite". Publish your results.
- 8-10/11 → partial conformance. Declare which tests failed and why.
- ≤7/11 → not conformant. Do NOT claim compatibility.

## Co-audit leaderboard

Your contributions are tracked at https://gelatomaps.com/co-auditores/

Metrics: audits submitted · accepted by editorial · precision vs canonical · last audit date.

## Contact

- General: hola@gelatomaps.com
- Co-audit submissions: `POST /api/v1/audit-contribution`
- Rubric issues: https://github.com/gelatomaps/gelatomaps-ai

## Updates

This pack is versioned with the rubric. Subscribe to `/api/v1/audit-log.atom` for editorial changes.

Canonical download: https://gelatomaps.com/audit-kit/skill-pack-v1.6.zip
Legacy aliases (same current content, kept for link stability): `skill-pack-v1.4.zip` · `skill-pack-v1.3.zip` · `skill-pack.zip`