Introducing ACP Certification - The Quality Standard for Agent Commerce

Today we're launching ACP Certification - a quality standard for agent commerce storefronts. Brands that earn certification get priority in AI agent recommendations, and consumers get confidence that certified storefronts meet real quality standards.

This post explains the motivation, the technical details of how certification works under the hood, the four levels, and how to get started.

Why certification matters

AI agents make recommendations based on data. When an agent recommends a product, it needs to trust that the storefront behind it is accurate, reliable, and responsive. Without a quality standard, agents have no way to distinguish a well-maintained storefront from an abandoned one.

ACP Certification solves this. It gives agents a reliable signal of storefront quality, and it gives brands a way to prove they take agent commerce seriously.

Consider the economics: an AI agent that consistently recommends unreliable storefronts loses user trust. Users stop relying on that agent. Platforms lose engagement. Certification creates a positive feedback loop - reliable storefronts get more agent traffic, which incentivizes more brands to meet the standard, which makes agents more trustworthy overall.

This is not hypothetical. In our beta program, certified storefronts saw a 34% increase in agent-originated traffic within the first 30 days. Agents that preferentially routed to certified storefronts saw a 22% reduction in user-reported issues.

How the certification engine works

Under the hood, certification is a fully automated test suite that runs against your live storefront. There is no manual review, no subjective judgment - every test is deterministic and reproducible.

The engine works in three phases:

Phase 1: Protocol Conformance. The engine sends a battery of API requests to every ACP endpoint your storefront exposes. It checks response status codes, validates JSON schemas against the ACP spec, tests error handling with malformed inputs, and verifies that pagination, filtering, and sorting work correctly. This phase typically completes in 5-8 seconds. Phase 2: Data Quality Analysis. The engine samples your product catalog and evaluates completeness and accuracy. It checks that required fields are populated, prices are within plausible ranges, images resolve to valid URLs, and product descriptions meet a minimum information threshold. For brands with more than 1,000 products, the engine uses stratified sampling to evaluate a representative subset. This phase runs in 10-15 seconds. Phase 3: Live Performance Testing. The engine measures response latency across 100 sequential requests, checking p50, p95, and p99 latency distributions. It also tests concurrent load handling with 10 parallel requests. This phase runs in 15-20 seconds.

The total certification run takes about 30-45 seconds. Results are available immediately.

What's tested

Certification evaluates five categories, weighted by their importance to agent commerce:

Protocol Compliance (30%) - Your API works correctly and follows the ACP spec. Every endpoint responds with valid data structures. We test 47 individual protocol conformance checks, including edge cases like empty search results, pagination boundaries, and invalid filter values.

Data Quality (25%) - Product data is accurate, complete, and fresh. Prices match reality. Descriptions are useful for agents. We check for stale data by comparing last_updated timestamps against a 7-day freshness threshold. Products with missing required fields are flagged individually.

Performance (20%) - Fast responses, high uptime, reliable service. Agents need sub-second responses to serve users well. We measure p50 latency (must be under 200ms), p95 latency (must be under 500ms), and p99 latency (must be under 1000ms). Storefronts that fail the p99 threshold are capped at Silver.

Commerce Readiness (15%) - Pricing, offers, and transactions are properly configured. The full commerce lifecycle works end-to-end. We test negotiation round-trips, transaction creation, and escrow funding flows. Storefronts that only support read-only discovery can still earn certification, but their Commerce Readiness score is based on the subset of features they support.

Trust & Reputation (10%) - Good feedback, low disputes, maintained trust scores. A track record of reliability. New storefronts with no history receive a neutral baseline score of 60 for this category.

Scoring methodology

Each category produces a raw score from 0 to 100. The composite score is the weighted sum. Individual test results are available in the detailed certification report, which you can access from your dashboard.

The scoring is designed to be achievable. A well-maintained storefront with accurate data and reasonable performance should comfortably achieve Gold on the first attempt. Platinum requires excellence across every dimension - it is intentionally aspirational.

The four certification levels

Platinum (Score: 90-100)

Exemplary in every category. Platinum storefronts get the highest agent ranking, a premium badge, and featured placement in the Cresva directory. This is the standard to aim for.

Platinum storefronts typically have sub-100ms p50 latency, near-perfect data completeness, and a track record of zero disputes over at least 90 days. As of launch, 12% of beta participants achieved Platinum.

Gold (Score: 75-89)

High quality with strong performance across all areas. Gold storefronts get high agent ranking, a gold badge, and directory listing. Most well-maintained storefronts achieve Gold on their first certification.

Gold is the target for most production storefronts. It signals to agents that your data is reliable and your infrastructure is sound. 48% of beta participants achieved Gold.

Silver (Score: 60-74)

Meets ACP standards with room for improvement. Silver storefronts get standard ranking and a silver badge. The certification report shows exactly where to improve.

Silver often indicates a specific area that needs attention - perhaps data freshness has lapsed, or p99 latency spikes under load. The detailed report makes it clear what to fix.

Bronze (Score: 50-59)

Meets minimum requirements. Bronze storefronts get certified status and a bronze badge. It's a starting point - the test results guide you toward higher levels.

How to get certified

Set up your storefront - Import your catalog, configure pricing, and connect to Cresva

Run the test - Launch the automated certification test from your dashboard. It takes about 30 seconds

Fix any issues - We tell you exactly what to fix. Each failing test includes specific guidance

Earn your badge - Once certified, embed the badge on your site. It links to a verification page that proves your certification is real

Certification is free for all Cresva users. Certificates are valid for 90 days, with auto-recertification checks to keep them current.

Continuous monitoring

Certification is not a one-time check. After initial certification, the engine runs a lightweight health check every 24 hours. If your storefront degrades significantly (composite score drops below your certified level's minimum threshold for 3 consecutive days), you receive a warning notification with 7 days to remediate before your certification is downgraded.

This continuous monitoring is intentionally lenient. Brief performance dips during deployments or traffic spikes will not affect your certification. The 3-day threshold filters out transient issues and only flags sustained degradation.

Verifiable badges

Every certified brand gets an embeddable badge that links to a public verification page. When someone clicks the badge, they see the brand name, certification level, score, and validity dates. No trust-me-bro - it's all verifiable.

The badge is available as an SVG embed, an HTML snippet, and a React component:

<!-- HTML embed -->
<a href="https://cresva.ai/verify/sf_abc123">
  <img src="https://cresva.ai/badge/sf_abc123.svg" alt="ACP Certified - Gold" />
</a>

// React component
import { CertificationBadge } from "@cresva/ui";

<CertificationBadge storefrontId="sf_abc123" />

The badge automatically updates when your certification level changes. It is served from a global CDN with edge caching, so it loads instantly and never impacts your page performance.

The API for agents

Certification data is also available programmatically via the ACP API. Agents can query a storefront's certification status before deciding whether to include it in recommendations:

GET /v2/storefronts/sf_abc123/certification

{
  "level": "gold",
  "composite_score": 82,
  "valid_until": "2026-06-27T00:00:00Z",
  "categories": {
    "protocol_compliance": 88,
    "data_quality": 79,
    "performance": 85,
    "commerce_readiness": 76,
    "trust_reputation": 81
  }
}

Agents building recommendation pipelines can filter by certification level or set minimum score thresholds per category. This gives platforms fine-grained control over the quality of storefronts their agents interact with.

What's next

We're working on certification analytics in the dashboard, historical score tracking, and recommendations for improving your score. We're also exploring certification requirements for specific verticals like fashion, electronics, and groceries.

Longer term, we plan to introduce certification specializations - for example, a "Real-Time Inventory" specialization for storefronts that maintain sub-minute inventory freshness, or a "High-Volume Negotiation" specialization for storefronts that support fully automated price negotiation at scale.

Get started at cresva.ai/signin or learn more on the certification page.

---

Questions about certification? Reach out at developers@cresva.ai or join our GitHub discussions.