Methodology — How it works

We show our working.

The Pattern is an automated daily intelligence brief. This page explains how signals are generated, what counts as a prediction hit, and why we publish misses as publicly as wins.

01 — Signal generation

How signals get here

Every day, an automated pipeline ingests articles from 151 curated feeds across fashion, brand, design, culture, entertainment, tech, and lifestyle. The pipeline runs at 06:00, 14:00, and 22:00 UTC.

Step 01
Feed harvest
151 RSS/Atom feeds pulled. Articles scored 0–100 for cultural relevance using a specialist scoring model.
Step 02
Deduplication
Same-story clustering removes near-duplicates. Signal variety across categories is enforced.
Step 03
Synthesis
Top-scored articles are passed to Claude (Anthropic). Signals, lead story, pattern thesis, and prediction are extracted.
Step 04
Publish
Output is verified, formatted, and deployed to thepattern.media. Audio version generated via ElevenLabs.

What the AI does and doesn't do: The AI synthesises and interprets. Every signal is grounded in a real article from a real source — URLs are published with each signal. The AI adds the layer of "so what." It does not invent stories or sources.

02 — Prediction rules

What counts as a hit

Every edition includes one prediction. Predictions are scored against strict criteria. There are no soft calls.

These rules exist to prevent the most common prediction failure mode: vague calls that can be retrospectively claimed as right. We'd rather be wrong and specific than right and useless.

03 — Confidence scoring

What the confidence score means

Each prediction includes a confidence score from 1 to 10. This is the model's subjective probability estimate at time of writing, not a retrospective rating.

Score
Meaning
Typical reasoning
1–3
Wild guess
Pattern suggests a direction but no concrete evidence. Worth noting; unlikely to land.
4–5
Speculative
Multiple signals converging. Plausible but uncertain. Most interesting predictions live here.
6–7
Probable
Strong signal basis. Consistent with the named entity's public direction. Expected range for daily predictions.
8–9
High confidence
Near-certain based on available signals. Reserved for calls where the outcome seems already decided by context.
10
Certainty
Not used. A score of 10 would mean the prediction adds no informational value.
04 — Why we publish misses

Accountability is the product

The problem
Prediction theatre
Most industry forecasting is unfalsifiable. Vague calls about "shifting consumer sentiment" can never be wrong. They're designed to protect the forecaster, not inform the reader.
Our approach
Permanent record
Every prediction stays on the record. Hits, misses, and expired calls are all published. The scoreboard on the homepage reflects the real numbers — no pruning, no reframing.
What misses tell us
Signal quality
A wrong prediction is diagnostic. It reveals where the signal model over-indexed, which sources create false positives, and which categories produce noisier signals. Misses make the system better.
Benchmark
60% is good
Industry analysts average 50–55% on specific calls. A hit rate above 60% on named, dated, binary predictions is meaningful. We aim for that bar, not a curated highlight reel.
05 — Lead time

What "ahead of mainstream" means

Signals carry a lead time estimate: the number of days before this story is expected to reach mainstream business or trade press coverage.

Signal lifecycle — typical path to mainstream
The Pattern
Trade press
Business media
Mainstream

Lead time is estimated based on the source tier of the originating article. Primary sources — specialist journals, niche trade reports, early industry data — carry longer lead estimates than secondary sources already in major trade press. A signal from a specialist publication not yet in mainstream business media might carry a 14–21 day estimate. A story already in trade press might carry 3–7 days.

These are estimates, not measurements. The figures reflect the estimated gap between source tier and mainstream coverage, based on typical diffusion patterns for that story type. We're transparent about this distinction.

06 — Signal curation

What makes a signal worth including