OPEN RESEARCH · FROM RELIGHT LABS

Signals are everywhere. Knowing which to believe is the hard part.

Attentive.io runs delayed market news through a multi-agent pipeline and shows every step in the open — no black box, no performance claims.

See the pipeline ↓ Follow the lab

A research instrument — not a signal service.

WHAT THIS IS

A research instrument, read openly.

PLAINLY STATED

We show how an AI reads the market's news. We don't tell you what to trade.

Attentive.io is a live, open look at how a multi-agent system reasons about market news. We publish the reasoning, not a recommendation — and we make no claim that any output is tradable. Nothing here is investment advice.

The point isn't the answer. It's that you can see how the machine got there, step by step, including the steps that lead nowhere.

THE PIPELINE

Specialized agents. Every step visible.

Delayed news in, structured reasoning out. Not one black-box model — a chain of agents, each with one job, each leaving its work on the record.

01 · INGEST

SCOUT

Pulls delayed news and market-moving events across feeds, extracts the facts — what happened, who's involved — and filters noise before anything else runs.

02 · CONTEXT

ANALYST

Reads the market and technical context around the event and assembles the structured picture the adjudicator will weigh. Its read is shown, not hidden.

03 · ADJUDICATE

JUDGE

Weighs the inputs and issues an internal confidence — the model's own read on conviction. Labeled plainly for what it is: an internal score, not a signal, not advice.

04 · RECORD

LEDGER

Logs every output and what actually happened next — including the ones that went nowhere. The misses stay on the record, not just the hits.

INGEST → EXTRACT → REASON → ADJUDICATE → RECORD

WHAT AN OUTPUT LOOKS LIKE

The reasoning, on the record.

Each event produces a structured read you can inspect end to end. The live feed updates on a delay. Below is the format — illustrative, not a recommendation.

▢ ILLUSTRATIVE — OUTPUT FORMAT

EVENT · central bank holds rates, guidance turns more hawkish than expected

Scout — extracted: Rate decision unchanged; statement language shifts on inflation persistence; affected sectors flagged. Source and timestamp attached.
Analyst — context: Reads the move against recent positioning and the prevailing volatility regime; notes what would confirm or contradict the read.
Judge — adjudication: Weighs the inputs, surfaces the dissent, and assigns an internal confidence with the full reasoning chain attached.

0.71 INTERNAL MODEL READ
NOT A SIGNAL · NOT ADVICE

CASE STUDY · LIVE

We went looking for an edge — and built the part that tells you if it's real.

We set out to find real alpha. To know whether we actually had it, we built an apparatus that separates a genuine edge from one that only looks like one — a leak, an overfit, beta wearing a costume.

Then we ran it on our own book. The book looked like a winner: net positive, a strong month, a respectable win-rate. The apparatus said the edge wasn't real — most of it was market beta, most of the rest was a handful of trades, and a candidate that looked like real signal was reading information it couldn't have known yet. It was right, and we believed it.

The signal generator is ordinary. The part that tells you the truth is the product — and the reason we can say, plainly, that we make no performance claim here. We hold our own work to the same standard.

Read the full teardown — the method, in detail →

FOLLOW THE LAB

Occasional notes from the bench.

What we're building, what we're finding, and the full case-study teardown when it's ready. No spam, no gate — just leave an address.

Or email info@attentive.io to be added.

Attentive.io is a research instrument. The reasoning shown is informational only, is not a signal or a recommendation, and does not constitute investment advice. No representation is made that any output is tradable or accurate. Markets involve risk of loss.