Skip to content
Helious Open the terminal →

Methodology

Every number on a Helious page is captured by machine from a primary source, then scored by published rules that were fixed before the print landed.

1. Catching a release

We poll a scheduled US economic release harder as its time gets closer, and every 3 seconds once a print is imminent. The moment a source publishes an actual, the desk puts it up with the forecast and the prior beside it.

The sources are the agencies that publish the statistic: BLS, BEA, Census, the Federal Reserve and Treasury. We read their own releases and the fastest mirrors of them that we trust. Each print page names the one it came from.

MEASURED CAPTURE LATENCY · LAST 7 DAYS
30s
median
216s
90th percentile
31
releases measured

2. Sizing the surprise in σ

A miss only means something next to how much that series usually misses by, so we divide one by the other: z = (actual − forecast) / σ, where σ is the historical standard deviation of that series' own surprises. Each series starts on a seeded σ that we refine as prints accumulate. A |z| under 0.25 is on the screws.

Which way the label falls is not a judgement call. A hotter CPI is bearish Treasuries, a payrolls miss is bullish, and the same per-family rule sets the beat or miss label and its bond or equity direction every time. The z-score on a print page is this number.

3. How an auction gets called

A Treasury auction recap measures the result against that auction's own history. Bid-to-cover goes against its recent average, the stop goes against the when-issued yield, the gap between those two is the tail, and the bidder split shows who actually took the paper. A stop-through on above-average cover reads strong, a tail on weak cover reads weak, and everything in between is in line. The thresholds that split those three do not move story to story.

4. Scores on headlines from the wire

Headlines that come off the wire sit in the live feed with a model score for bond and equity impact. That score is built from keyword evidence, how much weight the source carries and how new the item is, and on a high-conviction item we also check whether the market's first reaction confirms the read. We never present any of it as Helious reporting. Relayed headlines stay out of the search index too: only first-party data pages are indexable.

5. Revisions and corrections

Agencies revise. When one does, the page updates and carries a revised mark. When the mistake is ours we fix it as soon as we find it and the page says so. If you spot something wrong, email [email protected] and we review it the same day.

Watch it capture live →