Methodology
Wattpicker runs a two-pass extract+verify pipeline against every spec on every SKU. Headline specs (capacity, AC continuous, AC surge, plug types, expansion compatibility) require ≥2-source agreement before they go “green” in the catalog.
Pass 1 — Extraction
For each new SKU, an LLM agent extracts structured specs from a primary source: the manufacturer datasheet PDF, the brand product page, or a verified Shopify /products.json payload. Every extracted record carries source URL and an extracted_at timestamp.
Pass 2 — Verification
A separate verification agent (different prompt scaffold) cross-checks each headline spec against ≥2 independent sources. The agent flags discrepancies with structured detail. We measured a ~33% pre-verification headline-spec error rate on a tested SKU before this step was added — verification is non-negotiable.
Compatibility math is deterministic
LLMs produce plausible-but-wrong answers when asked to reason about wiring. Wattpicker uses deterministic engines for:
- PV wiring validation (per-MPPT-input voltage and current ceilings, cold-Voc derate, series/parallel topology)
- Plug-direction enforcement (TT-30R is an output port; TT-30 is a charging accessory — different concepts, different tables)
- Vendor-lock for expansion batteries (BMS protocols differ between brands; cross-vendor mixing is unsafe)
- Cable-connector matching
Reviewer citations are verbatim or absent
Every reviewer citation we render carries: verified URL, timestamp anchor (or article paragraph), verbatim quote, and a verified_at timestamp. If we're tracking a reviewer but haven't verified a citation yet, we display “verification pending” — we never invent a quote.
No display ads, no manufacturer sponsorships
We monetize via affiliate commissions on multi-SKU shopping lists. We don't take ad money or paid placement — that would compromise the tool. Affiliate disclosure.