01 · From source to signal

One article arrives. What happens next?

One article arrives. We classify it, map it to a topic, score its lean, and embed it. Click any signal to see the data it became.

What is a signal? Every RSS item, post, paper, or release we ingest is fetched, deduplicated, classified, embedded, and stored. That is what we call a signal.

Live demo · pick a signal

1 · Fetch

—

2 · Classify

—

3 · Topic

—

4 · Lean

—

5 · Embed

—

See the data

These are real signals from our database. Source, tier, and topic mapping are stored exactly as shown.

Signal id	Outlet	Tier	Title	Topic

Scroll for step 02 → From signals to consensus

02 · From signals to consensus

Now do that for thousands of signals.

Now we have thousands of signals on the same contested question. Toggle outlets on and off — watch how much the reading shifts.

The reading. Signals on the same contested question cluster into a single synthesis. Each outlet you toggle off changes the mix — and the contestedness score moves with it.

Live demo · toggle outlets

—

—contested

SettledContestedHighly contested

Adjust the toggles to see the reading move.

See the data

Real outlets, real signal counts. Toggling an outlet excludes its signals from the synthesis — the contestedness score recomputes from the remainder.

Outlet	Tier	Signals contributing

Scroll for step 03 → How contested is it?

03 · How contested is it?

A single 0 – 100 score per topic.

A single 0-to-100 number per topic that tells you, at a glance: is this settled, or are smart people still fighting about it?

How we read it. 0 means the question is approximately settled. 100 means experts and popular voices are in open conflict. Drag the slider to see real readings from our database.

Live demo · drag the slider

0 · Settled50 · Contested100 · Highly contested

—

Drag the slider to see a real reading at that contestedness level.

DI · — Signals · — Niche · —

See the data

Three real readings spanning the contestedness spectrum — each one is a published synthesis you can click into.

Topic	Score	Signals	Bucket

Scroll for step 04 → How much evidence is in the room?

04 · How much evidence is in the room?

A 0 – 1 score per topic.

How much of this conversation is grounded in data versus hot takes? A 0-to-1 score we publish next to every reading.

Reading the meter. High means the conversation cites data, studies, or firsthand experience. Low means the topic is dominated by unsourced takes. Read it next to the contestedness score.

Live demo · click a topic

Click a card above to read the synthesis.

See the data

Real readings. Each ratio is the share of substantive signals (cites data, studies, or firsthand experience) over all signals on the topic, weighted by source tier.

Topic	Evidence ratio	Contestedness	Signals

Scroll for step 05 → Cite or die — how we catch sycophants

05 · Cite or die — and how we catch sycophants

Every claim cites a real signal — or it does not ship.

Every claim cites a real source — or it does not ship. Here is one we caught making up a citation, and how the guard flagged it.

Why this matters. A sycophantic reviewer reads correct, fluent, agreeable — and can wave a fabricated citation through. We instrument for that failure mode explicitly. Most aggregators are lazy; we are not.

Real rejection · before publish

Rejected — guard fired before publish

—

See the data

The guard fires whenever (a) the citation does not resolve to a real source, OR (b) the cited stance disagrees with the actual signal mean by more than 0.5 on the −1…+1 axis, OR (c) the verifier returns zero specific disagreements (a sycophantic rubber-stamp).

That's the whole pipeline. See the live readings →