BLUEBEACONRESEARCH
Blue Beacon Research
Terminal arrow_forward

How It Works

Blue Beacon Research is a research platform, not a prediction machine. Research integrity is the whole point of this page — what we claim, and exactly how you can check it yourself.

2,999

Signals tracked

Every signal in the system, counted live from the database.

47

Signals, last 7 days

Rolling count of signals created in the past week.

31

Sources monitored

RSS feeds plus other data collectors configured in the ingestion pipeline.

What we claim, and how you can check it

Claim

Every signal has a Sources tab listing the reports it was built from, linking out to the original article where the publisher's URL is available.

How you can check it

Open any signal and check its Sources tab.

Claim

When the model is unavailable, signals fall back to a keyword classifier — and we mark it.

How you can check it

An auto-classified signal shows a note on its page: "This signal was auto-classified — Claude analysis is temporarily unavailable."

Claim

We do not add a severity bonus for a story running as the headline versus being buried in the body.

How you can check it

The headline-placement severity bonus in our scoring code is set to zero.

Claim

Confidence reflects the model's certainty in its own read of the event — not how many reports we have.

How you can check it

Stated directly in the product's own onboarding tour, next to the confidence score.

Claim

When multiple reports describe the same event, we count them as merged reports — not as distinct outlets.

How you can check it

The reports count increments on every merged report, regardless of whether the outlet repeats.

Claim

We publish our track record.

How you can check it

See the Accuracy page — including when the number is not flattering.

Limits — what we don't do yet

  • Severity and confidence are model outputs, not guarantees.
  • Some signals are classified by the keyword fallback instead of the model — those are marked, not hidden.
  • ACLED conflict-event data is not included yet. The collector is configured but is not currently ingesting (its last runs failed authentication).
  • A merged "reports" count can include more than one article from the same outlet — it is not a count of distinct sources.

See the full history on the Accuracy page.