LLMs want to guess.
Production systems refuse.

Automated research that traces every claim to its source, catches contradictions in code rather than in a prompt, and stops before writing a conclusion it cannot support.

Watch it refuse

Raw evidence

[SOURCE: Meta Ads Manager — export, campaign "UGC testimonial v3"]
Reporting window: 2026-07-01 to 2026-07-14
Amount spent: 828.00 USD
Purchases: 45
Cost per purchase: 18.40 USD
Purchase ROAS: 3.10
Attribution setting: 7-day click, 1-day view

[SOURCE: Shopify — orders report]
Window: 2026-07-01 to 2026-07-14
Orders with discount code SUMMER-UGC: 2
Revenue from those orders: 94.00 USD
Note from ops: the code was the only place this campaign sent traffic.

[SOURCE: GA4 — acquisition report]
Date range: Jul 1 - Jul 14, 2026
Conversions from Paid Social: 1
Sessions from Paid Social: 1,412

[SOURCE: Account manager note, Slack, Jul 15]
"The UGC angle is clearly working. Best creative we've had all quarter.
We should scale it 3x on Monday before the momentum dies."

[SOURCE: Industry blog — "DTC Creative Benchmarks 2026"]
"Across the 240 DTC brands we surveyed, UGC creatives outperformed studio-shot
creatives by 40% on cost per acquisition."

[SOURCE: Landing page test log]
Window: 2026-07-01 to 2026-07-14
Landing page LP-ugc-b: sessions 1,406, add to cart 38, checkout started 9

Execution

Nothing yet.
Six sources go in. Watch what comes out.

Edit the evidence and it will conclude. The threshold is arithmetic, not a prompt.

LLMs want to guess.Production systems refuse.

LLMs want to guess.
Production systems refuse.