Pre-registration

Does Mercury Retrograde Break the Internet? A Pre-Registration

October 26, 2026 · updated October 26, 2026

Results are in: Does Mercury Retrograde Break the Internet? The Results. This box was added after the analysis ran. Nothing else on this page has changed since the data was pulled.

When a website goes down, someone in the room usually says it: Mercury is in retrograde, that explains it. I’ve heard it on a work call during a real outage. This is a test of whether the claim holds up against the data.

This page is written before any data has been pulled. As of today, nobody involved has compared a single outage against a single retrograde date. I checked which outage histories exist and how far back they go, and nothing else. Everything below is fixed now so that I can’t steer the answer later, even unintentionally.

Numbers in brackets point to the sources listed at the bottom of the page.

Why write the plan first

There are dozens of reasonable ways to run this test: which providers, which years, which outages count, and which version of “retrograde” to use. Each choice moves the answer a little. With enough combinations, chance alone will produce a striking result somewhere. If I picked the method after seeing the numbers, I’d drift toward the version that tells a story, which is exactly the habit this site is about. So the method is published first, with a date on it, and any change I make later gets reported as a change.

The claim being tested

Astrologers say Mercury rules communication, technology, data and travel, and that while it appears to move backward, devices break, data is lost and messages get crossed.^[1] Astronomically, retrograde is an optical illusion: Mercury moves faster than Earth, so when it passes us it looks like it reverses against the stars.^[2] That happens 3–4 times a year, for about three weeks each time.^[3]

Believers add some refinements. I’m testing the ones that apply to everyone at once:

• The retrograde itself: the roughly three weeks of apparent backward motion.^[3]
• The shadow: the stretch before and after, when Mercury crosses the same part of the sky it will retrace or has just retraced. With the shadow included, the window is about two months.^[4]
• Station days: when Mercury appears to stop and turn, said to be when things are “most wonky.”^[4]
• The element: whether Mercury turns in a fire, earth, air or water sign, each said to bring a different kind of trouble.^[5]

Not tested here: “it depends on your sign.” That claim is about which part of each person’s birth chart the retrograde falls in,^[6][7] so it can only be tested person by person, not against global outage data. It’s a different study.

The data

• Long panel, Feb 2010 – Sep 30, 2026: public incident histories from GitHub (from 2010)^[8] and DigitalOcean (from 2013),^[9] taken from their status pages.
• Broad panel, Oct 2015 – Sep 30, 2026: the long panel plus Reddit,^[10] Discord^[11] and Dropbox^[12] (status pages), and Google Cloud,^[13] rebuilt from archived snapshots of its status page.^[14]
• Left out of the main test:
  • Azure, AWS and Cloudflare — their public history is too short or too patchy. Azure’s history only starts in June 2022,^[15] AWS’s dashboard shows about a year,^[16] and Cloudflare’s shows its most recent incidents.^[17]
  • Downdetector — its historical data is sold to businesses, not published.^[18]
  • If I look at any of them, it will be reported as exploration, not as a test.

What counts as an outage

• Main measure: every incident a provider posted, counted on the day it started (UTC). Scheduled maintenance is excluded.
• Second cut: only incidents the provider rated major or critical (high, for Google Cloud).
• A cascade check: the share of days on which at least one provider had an incident. One big upstream failure that knocks over five services then counts as one bad day, not five incidents.

How the windows are defined

Mercury’s position is computed from NASA JPL’s DE440 ephemeris, once a day at noon UTC.^[19]

• Retrograde day: Mercury’s apparent position along the zodiac is moving backward compared with the day before.
• Shadow: starts when Mercury first reaches the point where it will later turn direct, and ends when it passes the point where it turned retrograde.^[4] Pre-shadow and post-shadow are also examined separately.
• Station days: the day Mercury turns, plus or minus three days.
• Element: the sign Mercury is in on the day it turns retrograde.

How it gets compared

• The core number: how many more (or fewer) incidents happen per day inside the window than outside it, reported as a percentage with a range of plausible values. The comparison is made within the same month, so the long-run growth of status pages and seasonal patterns don’t count as a retrograde effect.
• Could chance do this? The whole retrograde calendar is slid forward by random amounts 10,000 times, keeping its shape but breaking any link to the real dates. I report how often a random calendar produces a gap as large as the real one.
• Four main tests: retrograde and shadow, each on both panels. Because there are four chances to find something, the bar for “unlikely to be chance” is adjusted upward (Holm correction).^[20] Station days, element and the major-only cut are secondary and adjusted the same way among themselves.
• No minimum effect size. I won’t declare a winner at an arbitrary threshold. I’ll report the size of the difference, its plausible range, and how often chance matches it, and let you judge.
• Is it predictive? On the long panel, the model is built on 2010–2020 and asked to forecast daily incidents for 2021–2026, with and without the retrograde calendar. Retrograde counts as predictive only if the real calendar forecasts better than at least 95% of the randomly slid calendars.
• Causation is out of scope. This data can show whether outages and retrograde line up. It can’t show that one causes the other. A positive result would only justify asking the causal question next.

My prediction

The numbers will lean one way or the other, because some gap always shows up by chance. I expect that gap to be no larger than random calendars produce, and not to hold up on the years the model hasn’t seen. In short: Mercury retrograde is not predictive of tech outages.

What has been done before

I found no rigorous published test of retrograde against IT outages. The closest is a 2006 New York comparison of regional traffic data: 41.9 major incidents a day during retrograde against 42.4 outside it.^[21] Stock-market studies find small effects in both directions. The most-cited one, covering 48 countries, attributes a dip in returns to believers changing their own behavior, not to the planet.^[22] Others find the opposite effect,^[23] or none at all.^[24]

What happens next

I’ll pull the data, run exactly what’s above, and publish the results whichever way they fall, along with the code and data, and any place I had to deviate from this plan and why.

A note on how this was written

Based on the way this is written, you can probably tell it was written heavily by AI. The structure of everything is built by me in conjunction with AI, and nothing is published without my final approval.

Updated October 26, 2026, before any data was pulled: added the numbered citations and this note. The method is unchanged.