How it's built

The engine.

The engine behind the News Weighting Terminal.

Anyone can point a language model at a news feed. What follows is everything else — the part that took every day since March to build, and the part that can only be built at the speed of the market itself.

452,000+instrument-level reads produced since 22 Mar 2026
48,000+stories ingested, deduplicated and weighed — source, headline and body text
85markets read continuously — 20 macro · 37 US stocks · 28 indices
~4 minmedian from headline on the wire to published read
2,280+strong reads graded by the market itself, on the public record
24/7scan cycle in minutes; urgent stories break the queue in real time

The pipeline

1

Ingest

A dozen-plus wire and market sources, the economic calendar, central-bank trackers (Fed, ECB, BOJ), positioning and inventory reports, and geopolitical event feeds — merged, deduplicated, entity-matched. Around the clock.

2

Read

Every story gets a read per affected market against a structured rubric spanning the categories that actually move price — central banks, geopolitics, economic data, supply & demand, trade policy, sentiment. One story can matter for oil, the yen and gold in three different ways; it gets three different reads.

3

Weigh & fade

New stories are weighed against the ones already on the scale. Each story is then tracked until it is spent — and how fast a story fades is not a guess: it's measured, per market, from our own graded history.

4

Net

Conflicting stories are netted into a single read per market — direction and weight, expressed in words and marks, never in raw numbers. When the wire is split, the read says so.

5

Publish & grade

Reads go out timestamped and are never edited. At each market's own measured horizon, price grades the read — with it, against it, or flat — and closed months are frozen forever. The misses stay on the page.

6

Learn

The grades feed back in. Category weights are recalibrated weekly against outcomes; the weighing rubric itself has been through repeated correction cycles, each one built from batches of graded reads the market disagreed with. The engine you read today is not the engine from March — it's the version the market has been correcting ever since.

The No-Chase Score

The instant a read is born we anchor the price, then track how far the market has already travelled in the direction that news points — measured in that market's own volatility, so gold, Bitcoin and an index are judged on the same scale. Stretched: the market is catching up; part of the move is already behind you. No-chase: the move happened, and price has left the read behind. Words only on public surfaces, never the number.

If it's priced in, it's not an edge. Most tools tell you the news was bullish. We tell you whether bullish is still worth anything.

What we will never publish

The rubric, the weights, the fade constants, the netting rules. Not because the ideas are exotic — because the values are the product of every correction cycle since March. They are what 450,000 weighed data points and months of public grading distilled down to. Publishing them would hand over the only thing that took time.

Why this is hard to copy

The code is the easy part — a good engineer with a good model could rebuild the skeleton in weeks. What cannot be rebuilt is the record: a public, timestamped, market-graded history only accumulates in real time, one read at a time. Ours started on 22 March 2026 and has never missed a day. Anyone starting today starts there — and they'd have to publish their misses too.

Fastest way to audit us: open the receipts, pick any timestamp, and check it against the post in t.me/pulsevane.

See the reads live →

Free channel available. Pro lists at $39.99/mo — founding price $29/mo for the first 500 members.