reverse engineering a viral website’s random number generator

an empirical analysis of a mean-spirited website

Published

March 15, 2026

tldr!

A viral mean-spirited site “Check Her Body Count” claimed to use AI to estimate the number of partners from an Instagram handle. It doesn’t. Independent developers confirmed the site is run entirely client-side; it makes no API calls, doesn’t validate whether a handle exists, and simply generates random numbers.

I had a specific question. If the website does generate random numbers, what probability distribution generates said random numbers? Is it uniform? Bell-shaped? Poisson? Exponential?

I collected 1,000 samples from the site and used maximum likelihood estimation, AIC/BIC model comparison, and goodness-of-fit testing to determine which probability distributions best explain the outputs. All five generated statistics follow discrete uniform distributions, the simplest possible RNG pattern, likely implemented as Math.floor(Math.random() * (max - min + 1)) + min with a different [min, max] range per stat.

1. introduction

In late February 2026, a misogynistic website called checkherbodycount.com claiming to estimate body count from an Instagram handle using AI went viral online. It is, of course, complete fiction, and harmful fiction at that.

Independent developers quickly confirmed it’s entirely a hoax:

  • The site runs 100% client-side in the browser
  • It makes zero requests to Instagram or any external API, and does not validate whether a handle even exists
  • It generates random numbers and caches them in localStorage.

Screenshot of website

Per query, the site generates five fake statistics:

Stat Element ID Example
Body Count #resultNumber 17
Male Followers #statFollowers 2,585
Thirst Comments #statComments 270
Late Night Posts #statLateNight 5
Suspicious Tags #statSuspicious 17

Along with a verdict string (“LOW”, “MEDIUM”, “HIGH”) and a visual risk meter.

However, as a quant, I had a different question. Specifically:

What probability distribution(s) does the site’s random number generator use? Are all five statistics drawn from the same distribution, or do they differ? And can the exact parameters be recovered from the output?

By collecting a sample of the site’s output and applying standard statistical tools, maximum likelihood estimation, information-theoretic model comparison, and formal goodness-of-fit tests, the RNG can be reverse-engineered.

The analysis is organized as follows:

  1. (this) intro
  2. methodology
  3. descriptive & visual exploration
  4. empirical analysis, incl. results of tests
  5. verdict & concluding thoughts

2. methodology

2.1 data collection

I used a headless browser (Puppeteer) to automate 1,000 queries to the site, each with a unique, randomly generated Instagram handle (e.g., alex48291_vibes). Each unique handle returns a sample of all five statistics. Between queries, localStorage is cleared to prevent caching from linking results across handles.

key design decisions:

  • Handles are randomly generated (e.g., alex48291_vibes) to avoid any pattern that could bias a hash-based RNG. If the site seeds its generator from a hash of the input, patterned handles (e.g., test_001, test_002) could introduce spurious structure.
  • Each handle is used exactly once → independent samples
  • All five stats are captured + the verdict string (e.g., “LOW”, “HIGH”) per query

The target sample size of 1,000 was chosen to ensure adequate statistical power for the chi-squared goodness-of-fit test. With the observed ranges, 1,000 samples provide sufficient expected cell counts across bins, though as discussed in Section 4.2, the widest-range variables still present challenges.

2.2 candidate distributions

Seven probability distributions were tested, each representing a plausible RNG implementation the site could use:

Distribution What it would imply Typical JS implementation
Discrete Uniform Every integer in [a, b] equally likely Math.floor(Math.random() * (b-a+1)) + a
Normal Bell-curve centered on some value Box-Muller transform on Math.random()
Poisson Random count with fixed average rate Knuth’s algorithm with cumulative products
Geometric Repeated Bernoulli trials Inverse transform sampling
Exponential Memoryless decay, right-skewed -β * Math.log(Math.random())
Log-Normal Multiplicative random effects Math.exp(μ + σ * normalRandom())
Negative Binomial Overdispersed count data Gamma-Poisson mixture

The first, Discrete Uniform, is the simplest possible pattern and what independent developers suspected from inspecting the client-side code. The remaining six represent progressively more complex alternatives that would require additional code to implement.

2.3 fitting & testing

For each of the five statistics and each candidate distribution, the analysis proceeds in three stages:

  1. Parameter estimation. Distribution parameters are fit via Maximum Likelihood Estimation (MLE), which finds the parameter values that make the observed data most probable under each candidate.
  2. Model ranking. Candidates are ranked using AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). Both metrics balance goodness-of-fit and penalize complexity: more parameters must meaningfully improve fit to be preferred. Lower values indicate better models. Typically, a model with ΔAIC > 10 has essentially no support.
  3. Goodness-of-fit testing. AIC ranks models relative to each other but does not indicate whether any model fits well in absolute terms. To address this, formal hypothesis tests are applied:
    1. Chi-squared test for discrete distributions -> bins the data and compares observed versus expected counts under the fitted distribution.
    2. Kolmogorov-Smirnov (KS) test for continuous distributions -> compares the empirical CDF to the theoretical CDF. KS tests are approximate here, as the data are discrete.

3. descriptive & visual exploration

3.1 descriptive statistics

Before fitting any models, the shape of each variable was examined using three diagnostic quantities:

  • Skewness ≈ 0 → indicates symmetry, which is consistent with uniform or normal distributions but rules out exponential and geometric distributions (both right-skewed by construction).
  • Kurtosis < 0 (platykurtic) → indicates a distribution flatter than normal. The uniform distribution has a theoretical excess kurtosis of −1.2, a useful benchmark for comparison.
  • Variance/Mean ratio >> 1 → indicates overdispersion relative to a Poisson distribution, which requires the variance to equal the mean. Ratios far above 1 effectively rule out Poisson.
Variable N Min Max Range Mean Median Std Dev Skewness Ex. Kurtosis Var/Mean
Body Count 1000 3 42 39 22.8 23.0 11.5 −0.0193 −1.2270 5.85
Male Followers 1000 423 6200 5777 3360.5 3352.5 1681.1 −0.0190 −1.2330 840.98
Thirst Comments 1000 15 720 705 377.1 383.5 206.1 −0.0376 −1.1998 112.62
Late Night Posts 1000 2 55 53 28.2 28.0 15.6 0.0328 −1.1771 8.59
Suspicious Tags 1000 1 35 34 17.9 18.0 9.8 0.0134 −1.1334 5.43

Across all five variables, skewness hovers near 0 (max |skew| = 0.038) and excess kurtosis clusters near −1.2, closely matching the theoretical discrete uniform benchmark. Combined with high variance-to-mean ratios, this effectively rules out Poisson and strongly suggests a uniform distribution.

Before any formal model fitting, the descriptive statistics alone point strongly toward a uniform distribution. The question for Section 4 is whether this initial impression survives rigorous testing.

3.2 visual overview

A side-by-side look at the histogram and CDF of all five variables.

Four panel chart with histogram, fitted overaly and QQ plot

The histograms confirm the flat, roughly equal-frequency shape expected of uniform distributions. No variable shows the bell-shaped concentration of a normal, the right-skew of an exponential, or the peaked shape of a Poisson.

The empirical CDFs reinforce this reading. A discrete uniform distribution produces a staircase CDF with evenly spaced, equal-height steps, which is exactly what all five panels display. By contrast, a normal distribution would produce the characteristic S-shaped sigmoid, and an exponential would rise steeply then flatten. Neither pattern appears.

Together, the descriptive statistics and visual inspection provide strong preliminary evidence that all five variables follow discrete uniform distributions with different [min, max] ranges. The formal model fitting in Section 4 will test whether this conclusion holds under maximum likelihood estimation and hypothesis testing.

4. empirical analysis

4.1 maximum likelihood estimation (mle) & model ranking

For each of the five statistics, all seven candidate distributions were fit via MLE and ranked by AIC. The results are unambiguous: Discrete Uniform ranks first for all five variables, with margins large enough to rule out alternatives. Shown below is the table for the first variable. The remaining four variables follow the same pattern; full results are in Appendix tables 1.

Body Count — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(3, 42) -3688.879454 7381.758908 7391.574419
2 Normal Normal(μ=22.8, σ=11.5) -3864.309690 7732.619380 7742.434891
3 Neg. Binomial NegBin(n=4.7, p=0.1709) -3892.039173 7788.078347 7797.893857
4 Log-Normal LogNormal(σ=0.67, scale=19.0) -3963.279759 7930.559517 7940.375028
5 Geometric Geometric(p=0.0482) -4008.744424 8019.488848 8024.396603
6 Exponential Exponential(β=22.8) -4125.180341 8252.360681 8257.268437
7 Poisson Poisson(λ=22.8) -5631.539201 11265.078402 11269.986157

🏆 Best: Discrete Uniform — Uniform(3, 42)

A few observable trends of note. First, the ranking order is nearly identical across all five variables: Discrete Uniform first, Normal second, Negative Binomial third, with Poisson consistently last by an enormous margin. This consistency suggests a single underlying RNG pattern applied with different parameters per stat.

Second, the ΔAIC between first and second place ranges from 300.9 to 374.3. Recall from Section 2.3 that a ΔAIC > 10 overwhelmingly suggests the model has no support. These margins exceed that threshold by more than an order of magnitude; the evidence for Discrete Uniform is overwhelming.

Third, the Poisson distribution deserves special mention. For Male Followers (λ = 3360.5) and Thirst Comments (λ = 377.1), the Poisson log-likelihoods are extremley poor (−468,168 and −68,172 respectively). This is expected: the Poisson distribution requires the variance to equal the mean, but these variables have variance-to-mean ratios of 841 and 113. The collected data is overdispersed for Poisson.

4.2 goodness-of-fit (gof) tests

AIC ranks models relative to each other, but doesn’t tell us if any model is actually good in absolute terms. For that I used formal hypothesis tests:

  • Chi-squared test (discrete distributions): bins data and compares observed vs expected counts
  • Kolmogorov-Smirnov test (continuous distributions): compares the empirical CDF to the theoretical CDF

In both tests, the null hypothesis, H0, is that the data follows the candidate distribution; a p-value above α = 0.05 means the distribution cannot be rejected. Shown below is the table for the first variable. The remaining four variables follow the same pattern; full results are in Appendix tables 2.

Body Count — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² 29.680000 8.590930e-01 ✓ Pass
1 Normal KS 0.081390 3.283897e-06 ✗ Reject
2 Poisson χ² 811.023224 2.628034e-157 ✗ Reject
3 Exponential KS 0.189205 8.056775e-32 ✗ Reject
4 Log-Normal KS 0.123700 8.544502e-14 ✗ Reject

The results split into two groups. For Body Count, Late Night Posts, and Suspicious Tags, the chi-squared test passes comfortably for Discrete Uniform (p = 0.86, 0.89, and 0.50 respectively) while every alternative distribution is rejected with p-values below 10⁻⁵. These are clean results; the data is consistent with a uniform distribution and inconsistent with everything else tested.

For Male Followers and Thirst Comments, the chi-squared test for Discrete Uniform returns NaN, a likely methodological issue. The chi-squared test requires binning the data and comparing observed versus expected frequencies; the standard rule of thumb is that each bin should have an expected count of at least 5. Male Followers spans a range of 5,777 unique values and Thirst Comments spans 705. With 1,000 observations spread across thousands of possible values, most bins contain 0 or 1 observations, and the test statistic becomes undefined. This is a known limitation of the chi-squared test for wide-range discrete uniform distributions.

However, several lines of evidence still support Discrete Uniform: - AIC dominance. Discrete Uniform leads by ΔAIC of 367.7 (Male Followers) and 374.3 (Thirst Comments); the largest margins in the entire analysis. - Descriptive statistics. Both variables show near-zero skewness and excess kurtosis of approximately −1.2, matching the theoretical uniform value. - Visual inspection. The histograms display flat, roughly equal-frequency bars across the full range, and the CDFs show even staircase patterns.

An interesting anomaly appears in the Male Followers table: the Poisson distribution passes the chi-squared test (p = 0.41) despite ranking dead last by AIC. This occurs because with λ = 3360.5, the Poisson approximates a normal distribution (by the central limit theorem), and the chi-squared test with wide bins lacks the power to distinguish between them.

This is actually a nice reminder that passing a goodness-of-fit test does not mean a distribution is the correct model; it means the test failed to reject it, which is a weaker claim. AIC, which penalizes the catastrophically poor log-likelihood, correctly identifies Poisson as the worst candidate.

4.3 pairewise correlation check

If the five statistics are generated by independent Math.random() calls, their pairwise correlations should be near zero. A non-trivial correlation would suggest that some stats are derived from others. For example, if one variable were derived from another (e.g., suspicious_tags from body_count), one would expect non-zero correlations.

As shown above, max absolute pairwise correlation is 0.060, which is well within the range expected from sampling noise with N = 1,000 (under independence, the standard error of a sample correlation is approximately 1/√N ≈ 0.032, so |r| < 0.06 is unremarkable).

This suggests that all stats show no evidence of linear dependence and the site runs Math.random() separately for each stat with no cross-dependencies.

A caveat: pairwise Pearson correlation captures only linear relationships. However, given that all five marginal distributions are uniform, a distribution with no natural “shape” to distort, nonlinear dependencies between variables would be unusual.

4.4 verdict classification

The site also assigns each query a text verdict alongside the body count. If the verdict is itself randomly generated, I would expect overlapping body count ranges across verdict categories. If instead it is deterministically derived from the body count, the ranges should be cleanly separated.

Boxplot showing verdict classification

The boxplot above of the four verdict categories of SAFE, MODERATE, HIGH, and EXTREME, occupy non-overlapping body count ranges. No MODERATE result has ever appeared above a body count of 15, and no HIGH result has appeared below 16. The boundaries seem to be consistent with fixed integer cutoffs:

Verdict Inferred Cutoff Rule
SAFE - Low Risk body count ≤ 5
MODERATE - Proceed with Caution 6 ≤ body count ≤ 15
HIGH - Think Twice 16 ≤ body count ≤ 28
EXTREME - Run. body count ≥ 29

The verdict is therefore likely not a separate random variable. It is a deterministic function of the body count, implemented as a simple set of if/else threshold checks.

5. verdict & concluding thoughts

5.1 summary of findings

The following table consolidates the results across all five variables:

Variable Best Fit Parameters AIC ΔAIC to #2 Evidence GOF p-value GOF Result
Body Count Discrete Uniform Uniform(3, 42) 7381.8 350.9 Strong 0.859093 ✓ Pass
Male Followers Discrete Uniform Uniform(423, 6200) 17327.6 367.7 Strong NaN —
Thirst Comments Discrete Uniform Uniform(15, 720) 13123.2 374.3 Strong NaN —
Late Night Posts Discrete Uniform Uniform(2, 55) 7982.0 350.6 Strong 0.887302 ✓ Pass
Suspicious Tags Discrete Uniform Uniform(1, 35) 7114.7 300.9 Strong 0.499790 ✓ Pass

So do all five stats use the same RNG pattern, or does the site use different distributions for different fake stats?

5.2 what this means

The analysis strongly suggests that every variable is best described by a discrete uniform distribution. The AIC margins exceed 300 in all cases, well above the conventional threshold for decisive evidence. Where the chi-squared test is applicable, it confirms the fit with high p-values. The near-zero pairwise correlations further suggest that each statistic has no linear dependence, with no evidence that any variable is derived from another.

In JavaScript, this is implemented as:

Math.floor(Math.random() * (max - min + 1)) + min

Each stat simply uses a different [min, max] range. The strongly negative excess kurtosis, near-zero skewness, dominant AIC scores, and near-zero cross-correlations all converge on the same conclusion.

The site rolls five independent digital dice, one per stat, applies a set of hard-coded cutoffs, and presents the outcome.

5.3 caveats & extensions

Four limitations and potential extensions are worth noting.

  1. Hash-based seeding: If the site seeds its RNG from a hash of the username (rather than Math.random()), the same handle will deterministically produce the same numbers even without caching. This can be tested by querying the same handle across different browsers/devices. The output distribution of a good hash function is still approximately uniform, so our conclusion holds either way.

  2. Correlation structure: The correlation analysis tests only linear association. While nonlinear dependencies are unlikely given the uniform marginals and the site’s minimal codebase, a more rigorous test using mutual information or distance correlation could provide stronger confirmation of independence.

  3. Sample size & test power: The chi-squared test failed for the two widest-range variables due to insufficient expected cell counts. Alternative approaches, such as a simulation-based exact test (Monte Carlo permutation test comparing the observed chi-squared statistic against simulated draws from the fitted discrete uniform), could provide formal GOF confirmation for these variables. For researchers replicating this analysis, a larger sample size would help.

  4. Likely underestimated max-min ranges: The fitted [min, max] parameters are derived from the observed sample extremes, which may underestimate the true code-level boundaries. For a Discrete Uniform with n possible values, the probability of observing the true maximum in N samples is 1 − ((n−1)/n)^N.

    • For narrow-range variables (Body Count, Late Night Posts, Suspicious Tags), 1,000 samples is likely sufficient, and the probability of missing a boundary value is small.
    • For wide-range variables, however, sampling coverage is sparse. Male Followers spans an estimated 5,778 unique values; with 1,000 samples, the probability of observing the true maximum is only ~16%, and likewise for the minimum. The true range for this variable is almost certainly wider than the observed [423, 6200] — plausible code-level boundaries might be round numbers like [0, 6500] or [500, 6500]. The same concern applies to Thirst Comments, though less severely. A larger sample (10,000+) or direct inspection of the site’s JavaScript source would resolve this.

references

★ Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control, 19(6), 716–723.

★ Britt, T. (2026). Site to ‘check’ women’s ‘body counts’ goes viral — and some men are defending it. Mashable. https://web.archive.org/web/20260304203624/https://mashable.com/article/check-her-body-count-website

★ Burnham, K. P., & Anderson, D. R. (2002). Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach (2nd ed.). Springer-Verlag. ISBN 978-0-387-95364-9.

★ Knuth, D. E. (1997). The Art of Computer Programming, Volume 2: Seminumerical Algorithms (3rd ed.). Addison-Wesley.

★ Massey, F. J. (1951). The Kolmogorov-Smirnov test for goodness of fit. Journal of the American Statistical Association, 46(253), 68–78.

★ Pearson, K. (1900). On the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random sampling. Philosophical Magazine, 50(302), 157–175.

appendix

appendix tables 1: mle

These are the tables for the 4.1 maximum likelihood estimation (mle) & model ranking analysis.

Body Count — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(3, 42) -3688.879454 7381.758908 7391.574419
2 Normal Normal(μ=22.8, σ=11.5) -3864.309690 7732.619380 7742.434891
3 Neg. Binomial NegBin(n=4.7, p=0.1709) -3892.039173 7788.078347 7797.893857
4 Log-Normal LogNormal(σ=0.67, scale=19.0) -3963.279759 7930.559517 7940.375028
5 Geometric Geometric(p=0.0482) -4008.744424 8019.488848 8024.396603
6 Exponential Exponential(β=22.8) -4125.180341 8252.360681 8257.268437
7 Poisson Poisson(λ=22.8) -5631.539201 11265.078402 11269.986157

🏆 Best: Discrete Uniform — Uniform(3, 42)

Male Followers — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(423, 6200) -8661.812881 17327.625762 17337.441273
2 Normal Normal(μ=3360.5, σ=1680.3) -8845.642321 17695.284643 17705.100153
3 Neg. Binomial NegBin(n=4.0, p=0.0012) -8887.450998 17778.901995 17788.717506
4 Log-Normal LogNormal(σ=0.65, scale=2824.4) -8939.994728 17883.989457 17893.804967
5 Geometric Geometric(p=0.0003) -8985.472779 17972.945559 17977.853314
6 Exponential Exponential(β=3360.5) -9119.834934 18241.669868 18246.577623
7 Poisson Poisson(λ=3360.5) -468168.136305 936338.272609 936343.180365

🏆 Best: Discrete Uniform — Uniform(423, 6200)

Suspicious Tags — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(1, 35) -3555.348061 7114.696123 7124.511634
2 Normal Normal(μ=17.9, σ=9.8) -3705.782292 7415.564583 7425.380094
3 Neg. Binomial NegBin(n=4.0, p=0.1842) -3747.963466 7499.926932 7509.742442
4 Geometric Geometric(p=0.0560) -3854.318474 7710.636947 7715.544703
5 Log-Normal LogNormal(σ=0.83, scale=14.0) -3866.093469 7736.186938 7746.002449
6 Exponential Exponential(β=17.9) -3882.843491 7767.686983 7772.594738
7 Poisson Poisson(λ=17.9) -5330.480299 10662.960598 10667.868353

🏆 Best: Discrete Uniform — Uniform(1, 35)

Thirst Comments — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(15, 720) -6559.615237 13123.230475 13133.045986
2 Normal Normal(μ=377.1, σ=206.0) -6746.748965 13497.497930 13507.313440
3 Neg. Binomial NegBin(n=3.4, p=0.0089) -6842.071023 13688.142046 13697.957557
4 Geometric Geometric(p=0.0028) -6893.413047 13788.826095 13793.733850
5 Log-Normal LogNormal(σ=0.83, scale=295.4) -6918.746911 13841.493822 13851.309333
6 Exponential Exponential(β=377.1) -6932.619123 13867.238246 13872.146001
7 Poisson Poisson(λ=377.1) -68171.844671 136345.689342 136350.597098

🏆 Best: Discrete Uniform — Uniform(15, 720)

Late Night Posts — Distribution Fit

Rank Distribution Parameters Log-Lik AIC BIC
1 Discrete Uniform Uniform(2, 55) -3988.984047 7981.968093 7991.783604
2 Normal Normal(μ=28.2, σ=15.6) -4164.298698 8332.597397 8342.412908
3 Neg. Binomial NegBin(n=3.7, p=0.1163) -4204.640358 8413.280717 8423.096227
4 Geometric Geometric(p=0.0367) -4285.915268 8573.830535 8578.738291
5 Log-Normal LogNormal(σ=0.80, scale=22.3) -4298.474454 8600.948909 8610.764419
6 Exponential Exponential(β=28.2) -4340.562343 8683.124686 8688.032442
7 Poisson Poisson(λ=28.2) -7311.831685 14625.663369 14630.571125

🏆 Best: Discrete Uniform — Uniform(2, 55)

appendix tables 2: gof tests

These are the tables for the 4.2 goodness-of-fit (gof) analysis.

Body Count — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² 29.680000 8.590930e-01 ✓ Pass
1 Normal KS 0.081390 3.283897e-06 ✗ Reject
2 Poisson χ² 811.023224 2.628034e-157 ✗ Reject
3 Exponential KS 0.189205 8.056775e-32 ✗ Reject
4 Log-Normal KS 0.123700 8.544502e-14 ✗ Reject

Male Followers — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² NaN NaN ✗ Reject
1 Normal KS 0.077837 1.023477e-05 ✗ Reject
2 Poisson χ² 94.482539 4.088680e-01 ✓ Pass
3 Exponential KS 0.187717 2.521115e-31 ✗ Reject
4 Log-Normal KS 0.114558 6.884858e-12 ✗ Reject

Thirst Comments — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² NaN NaN ✗ Reject
1 Normal KS 0.065040 4.029450e-04 ✗ Reject
2 Poisson χ² 150.327876 6.645739e-09 ✗ Reject
3 Exponential KS 0.175139 2.682286e-27 ✗ Reject
4 Log-Normal KS 0.141017 8.205825e-18 ✗ Reject

Late Night Posts — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² 40.904000 8.873015e-01 ✓ Pass
1 Normal KS 0.070845 8.263457e-05 ✗ Reject
2 Poisson χ² 246.226313 2.777597e-39 ✗ Reject
3 Exponential KS 0.174785 3.447390e-27 ✗ Reject
4 Log-Normal KS 0.131097 1.909027e-15 ✗ Reject

Suspicious Tags — Goodness-of-Fit Tests

Distribution Test Statistic p-value Result
0 Discrete Uniform χ² 33.340000 4.997899e-01 ✓ Pass
1 Normal KS 0.076849 1.391566e-05 ✗ Reject
2 Poisson χ² 551.128024 1.006710e-104 ✗ Reject
3 Exponential KS 0.177266 5.854542e-28 ✗ Reject
4 Log-Normal KS 0.143484 1.987062e-18 ✗ Reject

Disclaimer: This essay/blog contains research and analysis that do not reflect my employer. This essay may contain errors, omissions, or outdated information. It is provided “as is,” without warranties of any kind, express or implied. This essay is not investment, legal, security, or policy advice and must not be relied upon for decision-making. You are responsible for independently verifying facts and conclusions. The author does not accept any liability for losses or harms arising from use of this content. No duty to update is assumed.