How many werewolves are there, really?
Correlation analysis between grammar and spelling errors and the lunar cycle. Two scans six years apart, first one with |r| ≤ 0.15, p ≥ 0.24 and second with |r| ≤ 0.27, p ≥ 0.14.
1. Introduction
Do you sometimes watch the news and pause, stare at the ceiling, and think about geopolitics, the economy, war, AI, etc and wonder « what next? » and your brain is like « uhm, werewolf attack »?
Who benefits from a tariff war? From drawn-out conflicts like between Russia and Ukraine? From everyone needing AI to their shoelaces? THEM WEREWOLVES!!! THEY BENEFIT IT’S THEM
My point is, all that is happening in the world right now does help a third party. It feels like being in one of those movie scenes where a group of friends argues which party to go to before they are subsequently being eaten alive by a fleshy monster that has been preying on them all this time.
As if further confirmation were needed, recent silver prices show a suspicious spike. Who’s buying all that silver??? Potentially “bearish” investors - if you know what I mean, like, they might have fur. Everyone knows a werewolf can only be killed with a silver bullet. They’re buying up the silver so we can’t make bullets, obviously.
It gets worse. Repo spot prices lag behind the price of physical silver, which is the only type of silver the werewolf is concerned about.
The werewolves probably never really liked us. They blame you for climate change and, to make things worse, last week, someone stole their favorite chew bone and they think it might have been you.
I SEEK OUT TO OBTAIN CONFIRMATION OF THE THREAT
2. Background
Basically watching Underworld.
Aside from that, the existing research establishes:
- That werewolves get aroused by a full moon for the duration of the exposure.
- Arousal has been shown to increase the number of spelling mistakes in animal studies on humans.
- Humans do NOT get aroused by a full moon.
- There is no research suggesting that machines/bots are aroused by the lunar cycle and such appears to be very unlikely.
This makes a correlation analysis between the current lunar phase and the number of spelling mistakes (as a proxy for arousal) a good test to see if there is anything else using our networks that is neither human nor machine. Like, if you know the method is reliable and you know the expected result ahead, you can test the data.
3. Methods
To achieve final confirmation, we scrape tweets from X.com (formerly known as Twitter), run them through a spellchecker, and then join that dataset with lunar data.
The moon phase will correlate with the number of spelling mistakes made, as the werewolves will get aroused by a full moon and will thus type more sloppily in their aroused, moon-drunken state. The correlation’s strength will reveal their numbers as it indicates their share of the userbase. THIS IS HOW WE WILL GET PROOF.
There were two scans done, six years apart:
December 26, 2019 to February 23, 2020 (“first scan”)
December 14, 2025 to January 13, 2026 (“second scan”)
For each scan, the number of spelling mistakes was counted and subsequent correlation analysis with the lunar phase was performed (see “Results”).
3.1 Scraping
In both scans, I scraped all posts on X/Twitter containing the following hashtags:
The hashtags about Ireland were included as this is one of the locations where werewolves were definitely sighted before according to numerous reports.
The other hashtags were selected for their high posting frequency and geographic diversity, maximizing our chances of capturing werewolf activity across different time zones and social contexts. That werewolves are into food, beer, work, birthdays and challenges is a bit of an assumption but it seems plausible.
For the first scan, I used TwitterScraper [A. Taspinar].
For the second scan, I used the Octoparse Twitter Scraper [Octoparse], as TwitterScraper is no longer maintained (possible werewolf interference, suspicious, will be investigated later)
3.2 Preprocessing
I filtered for English-language posts and cleaned the text by removing URLs and emojis.
3.3 Spelling check
I then analyzed each post for grammar and spelling errors using LanguageTool [LanguageTool; D. Naber].
For the first scan, I used GAMET [S. Crossley et al.], which relies on the Java version of LanguageTool.
For the second scan, I used the Python wrapper for LanguageTool [J. Morris].
I grouped the data by day and calculated the total number of language errors and total number of posts for each day, both by hashtag and across all hashtags.
I used this to calculate the daily error rate: total language errors divided by total posts.
3.4 Obtaining lunar data
For both scans, I obtained daily lunar data using the suncalc R package [B. Thieurmel and A. Zeileis]. This basically returns data about the moon, such as its phase, fraction that is illuminated, or angle, for a given day.
3.5 Correlation analysis
I calculated Pearson correlation coefficients between the daily error rate and four binary variables representing lunar phases:
1st Quarter
New Moon
3rd Quarter
Full Moon
As well as for three continuous variables:
moon angle
moon fraction
moon phase
3.6 Change between scans
If werewolves recalibrate their countermeasures over time, the strength of a correlation should change between scans even if no single correlation is significant. To look for such an oscillating field, I compared each correlation from the first scan with its counterpart from the second scan using a Fisher z-test for two independent correlations, with n = 60 days in the first scan and n = 31 days in the second. For the binary lunar phases only absolute correlations are available, so this compares strength, not direction.
4. Results
4.1 Results of First Scan
The first scan revealed no statistically significant correlation between error rate and the lunar cycle.
The absolute correlations calculated for each hashtag with the four binary variables - 1st Quarter, New Moon, 3rd Quarter, Full Moon - are not particularly large, and none reach statistical significance (see Table 1).
Similarly, the correlations with the three continuous variables - moon angle, moon fraction and moon phase - were small and not statistically significant (see Table 2).
| Variable | Correlation | p-value |
|---|---|---|
| Moon angle | 0.0745 | 0.5715 |
| Moon fraction | -0.1538 | 0.2408 |
| Moon phase | -0.1003 | 0.4459 |
| Average | -0.0599 | – |
But this does not stop me. Maybe the Moon fraction was showing some promise. A p-value of 0.24 looks reliable to me. The correlation is also negative but this I skip over.
The below charts show the correlation of the above factors again with the individual data points shown in them:
In summary, while this does not yet provide FINAL CONFIRMATION of werewolves, it does also not reject their existing. Some factors, such as the Moon angle, show a promising correlation of 0.07 with a p-value of 0.57.
They are real. They are there. I just know it and I will find proof. Maybe some people say “You’ll never find them!”.
But I’ll never stop looking.
4.2 Results Second Scan
So, six years later, I conducted a second scan.
The first scan’s results were inconclusive after all, and the situation may have changed.
The initial methodology may have been too obvious. I now suspect werewolves saw it coming and deployed countermeasures - specifically, posting deliberately misspelled tweets with negative correlation to the lunar cycle to mask the real signal.
Classic counter-intelligence operation, kindergarden stuff. The null results from the first scan suggest such operations may exist. But now I am onto them and this time, I will not be so easily fooled. Scanning again after six years makes sense, because:
The werewolf in charge of the counter-intelligence operation could have quit in the meantime and no replacement has been found
Maybe we just get lucky the werewolf is currently on holiday or or parental leave
He or she may could have become demotivated over time, has been a “silent quitter” and does not believe in the value of their job anymore because they do not see the scans
They may have run out of funding as they could not show any results from their work and the public grew tired of their taxes being wasted on pointless endevours
They need to index the number of intentionally misspelled tweets during the New Moon to the ratio of their X-using population to the ratio of our X-using-population, further adjusted by any change in usage intensity. This system may not always work perfectly. Maybe the way to find nests is not to look at the correlation but at the change of the strength of the correlation over time. To find a werewolf nest, it might be necessary to look at places where the strength of the link between typos and the lunar phase is oscillating over longer time periods. Such an oscillating field over a narrowly confined geographical area could be a way to find them.
The above process (see Methods section) was thus repeated with as little changes as possible.
However, the second scan yielded results nearly identical to the first: again no statistically significant correlations between error rates and lunar phases (Tables 3 and 4).
Results below:
| Variable | Correlation | p-value |
|---|---|---|
| Moon angle | 0.1955 | 0.2920 |
| Moon fraction | 0.1255 | 0.5013 |
| Moon phase | 0.2723 | 0.1384 |
| Average | 0.1978 | – |
Again, individual data points as charts:
The pattern - or lack thereof - mirrors the first scan almost exactly. This consistency across a six-year gap is suspicious. This is not over yet.
I see sophisticated coordination. This level of preparation indicates the threat is even larger and better organized than initially hypothesized.
4.3 Change between scans
But “almost exactly” is not “exactly”. Of the 47 comparisons (44 hashtag–phase pairs and 3 continuous variables), three changed significantly at p < 0.05 (see Table 5). All three are New Moon correlations, and one of them is #Dublin, a known werewolf habitat. Across all hashtags, the strength of the New Moon correlation rose by 0.26 on average, and the 1st Quarter by 0.14, while Full Moon and 3rd Quarter fell slightly (−0.06 each).
| Correlation | First scan | Second scan | Change | p-value |
|---|---|---|---|---|
| #beautiful, New Moon | 0.0622 | 0.6102 | +0.5480 | 0.0051 |
| #challenge, New Moon | 0.0555 | 0.5053 | +0.4498 | 0.0300 |
| #Dublin, New Moon | 0.0200 | 0.4778 | +0.4578 | 0.0302 |
| #foodie, 1st Quarter | 0.0525 | 0.3962 | +0.3437 | 0.1122 |
| #birthday, New Moon | 0.0357 | 0.3693 | +0.3336 | 0.1273 |
| Moon angle | 0.0745 | 0.1955 | +0.1210 | 0.5928 |
| Moon fraction | -0.1538 | 0.1255 | +0.2793 | 0.2230 |
| Moon phase | -0.1003 | 0.2723 | +0.3726 | 0.0997 |
This is exactly the countermeasure predicted above: intentionally misspelled tweets during the New Moon, recalibrated since 2020. The field over Dublin is oscillating.
Unfortunately, with 47 comparisons about two significant results are expected by chance alone, and none survives a Bonferroni correction (α = 0.05/47 ≈ 0.001). The continuous variables all moved in the same direction, but none changed significantly either. So it might be a nest. Or it might be noise. A third scan is needed.
5. Discussion
There are some explanations for these results. For example:
Their defense system may be better than assumed, simply exposing no oscillatory pattern that can be detected by our scans
They use spellcheckers i.e. browser plugins to alert them of typographic mistakes
They are trained, from early age on, to just not use the internet during full moon (sounds like a great idea tbh)
They do not use X. Instead, they have their own social networks
The whole thing with werewolves being sensitive to the moon could be a myth, maybe they just aren’t. Instead maybe they just have good eyesight during Full Moon relative to their prey (us) and this is why they show up during that time. Thus, it could be selection bias
All werewolves have better things to do during Full Moon than to use the internet, like they are having an orgy or something
The above analysis could be unintentionally wrong i.e. I could have made a mistake and without my knowledge, report incorrect results
The above analysis could be intentionally wrong i.e. I have an incentive to report incorrect results. I could have held captive by them or I may have been bought with silver. Ultimately, I can not be trusted
I could be a werewolf
I am not a werewolf, not a human and not a machine but a fourth party. I have an incentive for the other parties to fight each other (anyone here playing board games)
The werewolves have infiltrated X, have set up systems in place to detect scanning/scraping and redirect those requests to a subsample of the real data which has been typos-over-time-stratified. It’s a type of shadow-ban where you think you get the real result but you silently don’t
The werewolves have infiltrated the domain name service (DNS), and return a false host for those requests which serves a subsample of the real data which, again, has been typos-over-time-stratified
The werewolves have no defenses to data extraction and do react to the lunar cycle as we think, trigger a spike in misspellings. However, there is also an undiscovered species called counter-werewolves who behave exactly inversely to the lunar cycle and also use the internet, canceling out the werewolves’ pattern
Werewolves have actually evolved to transform at different phases of the lunar cycle, so there are always some transformed werewolves in the herd that can defend the nest, similar to humans having genetically different sleep schedules. This has the added benefit of not exposing a pattern.
But let’s be real, this is all nonsense and I guess that for once everyone can agree that werewolves don’t exist.
HTMEHM WINDOWS hgnjmelklsdncm + ace0huiojpojmklo,¨ o. +uibj iuzb uibh jijkllllllllllllll- ¨ llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllSYSTEM NOTE: this post has been automatically published due to reaching the maximum character limit
Comments
Comments are hosted by Disqus. Showing them loads content from disqus.com, which may set cookies; see the Disqus privacy policy.