42Analytics

Evidence

The validation record.

Every claim on this site has somewhere to go. Every result links to the methodology behind it. Every case study includes scope conditions — where the approach works and where it does not.

We do not distinguish between results and failures. We document both.

FishFacts · Updated daily

Peterhead demersal fish markets

76,579

forecasts verified

each tested against real auction prices, May 2019 – Jul 2026

67%

directional accuracy

correctly called price direction, H3–H10, vs 50% by chance

+13.5%

ahead of no-change

more accurate than assuming today's price holds, H5–H10

82.2%

price range reliability

actual prices landed within forecast range — target 80%

All figures are out-of-sample results across 110 species-grade combinations, tested against real Peterhead auction prices from May 2019 to present. No cherry-picking of species or horizons. The record updates automatically after every trading day.

Full species-level record, updated daily →

Validation principles

Hold-out validation

All performance claims are validated on data the model has never seen. In-sample fit is not evidence.

Scope conditions

Every result comes with explicit scope conditions — the market regimes, data volumes, and operational constraints under which it applies.

Published failure modes

When the methodology breaks down, we document it. The beef post-2021 regime degradation is in our research. The Q1 2022 lamb anomaly is in our case study.

Method

Understand the methodology

Decision Geometry — the retrieval-based approach underlying all 42Analytics products. How market states become geometric objects, how analogues are retrieved, and how forecasts are validated.

Read the method →