P.S.: Now that I have your attention, replace 'Israel' with 'Islamic Republic of Iran' and 'Gaza' with 'Iran'. The longest internet shutdown in a connected society is happening right now — but crickets from the usual crowd.
The genocidal Israel regime has imposed a brutal digital blackout on Gaza for 44 days, cutting off millions from the world, silencing voices, and piling up human & economic suffering. Where is the outrage from the UN and human rights groups? 😡
The trend shows the problem has been consistent through the years.
However, the top is MSc in Health Sciences (n=78), followed by MSc in Biology (n=70). MSc in HRM is the third (n=46).
The top PhDs are Biochemistry (n=43), Medical Sciences (n=41), and HRM (n=41).
Our Department, Health Research Methodology, has the most theses (88) with the "finalsubmission" error (out of 103 theses with "final" in the PDF name). Most of them are from MSc theses (n=46).
McMaster's naming convention says:
familyname_firstname_middleinitial_finalsubmissionyearmonth_degree
The idea: replace "finalsubmissionyearmonth" with your actual submission date.
What 1,343 students did: copied "finalsubmission" literally.
Doe_John_finalsubmission2025December_PhD.pdf
Only 9.9% of McMaster theses are "final"
I analyzed 17,501 theses in McMaster's repository and found that 1,343 grad students literally typed "finalsubmission" in their filename (out of 1,729 with "final").
We obsess over data sharing, pre-registration, and reproducibility, but we let the evidentiary foundation of our references quietly disappear.
I wrote up the full analysis with all the data and figures. Read it here: choxos.substack.com/p/the-decay-...
The fix? It takes 30 seconds. Before you submit, paste your web references into web.archive.org/save. You get a permanent, timestamped snapshot. Add that archived URL to your reference list. Done.
And here's the part that should concern everyone in research: 64% of dead links have no Internet Archive backup. The cited content is permanently lost from the scientific record.
For citations from 2002–2004, the dead link rate exceeds 78%.
Even when the link still works, 27% of the time the page content has silently changed. The URL loads fine, but the evidence behind the citation is gone.
I checked 17,776 web links cited in 27,000 biomedical research articles.
29% are dead.
The worst offender? The @WHO.int website: 458 broken references. CDC was second with 280. These aren't obscure blogs. They're the most authoritative sources in global health.
I recommend watching this video about the origins of Newroz:
“Three elements dominate Kurdish Newroz: fire, the Kawa vs. Zuhak myth, and Mir Newroz (Prince of Newroz). These rituals trace directly back to the Sumerian New Year festival, Zagmuk (𒍠𒈬).”
youtu.be/t9FUZ-G6zQk
Fyi, it’s not just “Persian”. Many other nations, including the Kurds, celebrate Newroz (which is a festival of fire and not “Haft-Sin”).
@springernature.com successfully sat this one out. No retraction.
Retraction plans abandoned, @bmj.com ?
Are you aware how many children were killed by the Islamic Republic of Iran in a few hours two months ago?
iranhumanrights.org/2026/02/over...
Reference:
- Excess deaths associated with the Iranian COVID-19 epidemic: A province-level analysis
Ghafari, Mahan et al.
International Journal of Infectious Diseases, Volume 107, 101 - 115. doi.org/10.1016/j.ij...
Added Health & Economic Impact. Based on the calculations, 2 million years were lost which can translate to $95.3 billion.
Limitations: Linear scaling assumptions, geographic homogeneity, and subjective priors. This demonstrates QBA methodology applied to crisis estimation, not a definitive count.
Interpreting the 12,000 figure: If reporting sensitivity is 10–25%, the true death toll implied by Iran International's count is 48,000–120,000.
Results: 10,000 Monte Carlo iterations propagate uncertainty through all parameters. At moderate intensity (2x 2019), the median estimate is ~77,000 deaths (95% CrI: ~20,000–300,000). The wide interval reflects genuine uncertainty, not imprecision.
Why these distributions? Triangular is ideal when you have a most likely value but uncertain bounds. Trapezoidal allows a range of equally likely values rather than a single mode, more realistic when we genuinely don't know if sensitivity is 15% or 25%.
- Reporting sensitivity: Trapezoidal distribution (10%–35%)—the flat region (15%–25%) represents equally plausible values, with linear tails capturing extreme scenarios
- Baseline uncertainty: Sampled from truncated normal distribution using Ghafari's 95% CI
- Intensity multiplier: Triangular distribution (min=0.5, mode=2.0, max=10.0) reflecting uncertainty about violence severity relative to 2019
The model:
Estimated Deaths = Baseline × Geographic Factor × Duration Factor × Intensity
This was 4x higher than Reuters' estimate of 1,500, meaning media reporting captured only ~25% of actual deaths even before the current blackout.
The anchor point: Ghafari et al. (2021) analyzed Iranian civil registration data and found 6,040 excess deaths (95% CI: 3,480–8,600) during the November 2019 protests.
Therefore, I built an interactive tool (link in the first comment) that estimates true mortality using Quantitative Bias Analysis (QBA), a systematic method for adjusting epidemiological estimates when data sources have known/unknown biases.
Based on my lived experience in Iran and the news and videos, I thought it's too conservative.