Every fact ingested by MeritSKU is tagged with an evidence tier and a SHA-256 cryptographic provenance digest. This provenance tracking guarantees auditability and prevents speculative LLM hallucinations from poisoning financial ledgers.
1. The 5-Tier Evidence Hierarchy
Our architecture strictly partitions evidence into five discrete tiers with decreasing confidence weights:
| Tier | Classification | Examples | Confidence Weight |
|---|---|---|---|
| Tier 1 | Observed Fact | Pro-forma factory invoice, carrier API tracking, audited bank receipt | 1.0 (Highest) |
| Tier 2 | Platform Analytics | Direct Shopify order history, Google Search Console telemetry | 0.85 |
| Tier 3 | Market Aggregation | Aggregated 3PL rate sheets, commercial freight index, HTS tariff book | 0.70 |
| Tier 4 | Heuristic Models | Competitor review velocity trends, category return rate regression | 0.50 |
| Tier 5 | Merchant Assumption | Unverified user estimates, verbal supplier promises | 0.20 (Lowest) |
2. Cryptographic Payload Digests (SHA-256)
Whenever an external quotation or document is ingested, MeritSKU computes the raw SHA-256 hash of the payload and records it in the evidence ledger.
This establishes a permanent, immutable chain of custody: any alteration to supplier terms or pricing immediately invalidates the upstream evaluation.
3. Zero-Hallucination Policy & Missing Data Rules
If a critical attribute (such as package height or tariff code) is absent from the source document, MeritSKU strictly fails closed to typed null.
The AI extraction engine is prohibited from guessing or fabricating realistic-sounding defaults. Missing attributes are surfaced as explicit blocker warnings in the decision report.
Was this guide helpful?
Your feedback trains our editorial documentation standards.