Everything starts from the release minute. For a report published at 08:30 ET, the reference price is the close of the last one-minute bar before 08:30. Every figure on the site is a distance from that price, in percent, so a reading of 0.50% means the index was half a percent away from where it stood the moment the number came out.
Four marks are read after the release. One minute: the close of the first bar after the print, used on the first-sixty-seconds page. Five minutes: the first direction, used to decide whether a move reversed. Thirty minutes: the window the site is built around, where the swing, the net move and the up-or-down count are taken. Two hours: the check on whether thirty minutes was enough. For an FOMC decision at 14:00 those marks fall at 14:01, 14:05, 14:30 and 16:00, so the two-hour reading lands on the cash close and the thirty-minute reading lands at the exact start of the press conference. For an 08:30 report they fall at 08:31, 08:35, 09:00 and 10:30, so the two-hour reading includes the first hour of the regular session. The pages that use those marks say so, because on those rows the clock is part of the explanation.
One more window is read before the release: the sixty minutes ending at the print, used on the page about the quiet before. It is twice as long as the window after, and the ratio on that page is not adjusted for it.
Swing is the highest price minus the lowest price inside the thirty minutes after the release, in percent of the reference price. It is the headline figure on every release page and the basis of the ranking, and it is deliberately not the net change: a half hour that ran half a percent up and half a percent back down has a swing of a percent and a net move of nothing, and anyone holding a position through it would side with the swing.
Net move is where the price sat at a mark relative to the reference price, with its sign. Distance is the same thing without the sign, used wherever direction would only add noise, which is the two-hour comparison and the first minute. Best and worst are the furthest the price reached above and below the reference inside the window, not where it closed. A reversal is a five-minute net move and a thirty-minute net move with opposite signs, by any margin.
Wherever a page reports a single figure for a release, it is the median across every measured occurrence of that release, never the mean. A mean of thirty half hours is a number the widest three of them decide. The largest single day is reported separately, with its date, so the median has something to sit against.
Shared minutes. When two reports on this site print in the same minute, that occurrence is excluded from both of their medians. The move belongs to both and to neither, and counting it under each one inflates the quieter release. Those dates are still shown on the release pages, marked rather than hidden, and the page on shared minutes reports what they look like counted in.
Small sides. A split by forecast surprise is only shown when each side has at least eight occurrences, and a release only appears in the cross-release tables once it has twelve on record. Below that, a count is an anecdote.
Exact matches. A print that equals its consensus forecast to the published precision belongs to neither the above column nor the below one and is left out of both.
Sessions. A release that landed on a market holiday, or whose price record has a gap inside the window, is not measured. It is better to have one fewer observation than one that is wrong by a session.
Prices are one-minute bars on Nasdaq 100 and S&P 500 index CFDs. They trade around the clock, which is what makes the 08:30 releases measurable at all: the equity market does not open for another hour, and only the futures and the instruments that track them are moving. The CFDs follow the front-month futures closely, but they are not the CME contracts. Every figure is a percentage of the index level, never a contract point, and the basis between the contract and the index is not in these numbers. The site is measured on the Nasdaq 100 by default because the reaction is larger there by a fairly constant margin, which the page comparing the two indexes reports release by release.
Release dates and times are taken from the publishing agencies' own schedules: the Bureau of Labor Statistics, the Census Bureau, the Bureau of Economic Analysis and the Federal Reserve. A calendar rule would put the jobs report on the first Friday and be wrong several times a year, and a date that is wrong by a day measures the wrong minute of the wrong session.
The printed figures are read from two independent sources, the publishing agency's own release text and an economic calendar feed, and where both carry a value for the same release the two are compared at every rebuild. A disagreement is flagged, the agency's figure is the one shown, and the flag is what gets a human to look. That check exists because the site's own parser has been wrong before: it once read the BLS phrase “changed little” as a print of zero, on three payroll reports that were not zero, and the disagreement with the second source is what caught it. The consensus forecasts come from the calendar feed alone and are not cross-checked, which is one reason the forecast splits are shown as counts with their sample sizes rather than as anything firmer.
Volume, so a quiet hour in price may be a busy hour in the book. Slippage, so the first-minute distance is where the index landed and not what a fill would have cost. Options, implied volatility and the futures basis. The size of a forecast surprise, which is folded into above or below. And the direction of the next print, which nothing on this site claims to know. Every table shows how many observations it stands on so you can decide for yourself how much a pattern across a few dozen releases is worth.
The record is rebuilt every Saturday, and again on the afternoon of any day a measured release prints: the thirty-minute window closes, the tick feed publishes the hour, and the site rebuilds with that release added, usually two to three hours after the print. Its sixty-minute and two-hour readings fill in on the next rebuild. Every median, ranking and pattern page is recomputed from the full record each time. The sentences on the pattern pages that quote a figure are filled from the same rebuild, so a number in a paragraph and the number in the table beside it always come from the same run. When a definition changes, the whole record is remeasured under the new one rather than mixing two rules across time.
Releases: Core PCE Price Index Consumer Price Index FOMC Rate Decision GDP JOLTS Job Openings Nonfarm Payrolls Producer Price Index Retail Sales