About

Fuzz testing temporal data requires controlled randomization around a known anchor point. A naive approach - uniform random across a calendar year - fails to model real-world clustering where events concentrate near a reference timestamp with predictable decay. This tool applies statistical distributions (uniform, Gaussian via Box-Muller, triangular) to a seed date d₀, producing n output dates within a window of ±R units. Each offset δ is sampled from the chosen distribution, then added at millisecond precision: d_out = d₀ + δ. Incorrect date fuzzing in integration tests masks timezone bugs, off-by-one boundary errors, and leap-second edge cases that surface only in production. The tool caps output at 10,000 samples and computes summary statistics (mean, median, σ) so you can verify the spread before export.

Formulas

The core operation adds a random offset δ (in milliseconds) to the seed date's Unix timestamp:

d_out = d₀ + δ

The total fuzz range R in milliseconds is computed from user inputs:

R = D ⋅ 86400000 + H ⋅ 3600000 + M ⋅ 60000 + S ⋅ 1000

For uniform distribution, where U ∈ [0, 1):

δ = R ⋅ (2U − 1)

For Gaussian distribution, Box-Muller transform with U₁, U₂ ∈ (0, 1):

δ = R3 ⋅ √−2 ln U₁ ⋅ cos(2πU₂)

The σ is set to R3 so that ≈99.7% of samples fall within ±R. Values outside the range are clamped.

For the results summary, standard deviation of the output set:

σ = √1n n∑i=1 (d_i − d)²

Where D = days, H = hours, M = minutes, S = seconds, U = uniform random variate, n = sample count, d = mean of generated timestamps.

Reference Data

Distribution	PDF Shape	Best For	Offset Formula	Clustering
Uniform	Flat	Equal-probability spread	δ = R ⋅ (2U − 1)	None
Gaussian	Bell curve	Natural clustering around center	Box-Muller: δ = σ ⋅ √−2 ln U₁ cos(2πU₂)	Strong center
Triangular	Triangle peak	Moderate center bias	Inverse CDF with mode at center	Moderate center
Edge-weighted	U-shape	Boundary stress testing	Inverted triangular	Strong edges
Common Output Formats
ISO 8601	2024-03-15T14:30:00.000Z
RFC 2822	Fri, 15 Mar 2024 14:30:00 +0000
Unix Timestamp	1710513000
US Locale	03/15/2024 2:30 PM
EU Locale	15.03.2024 14:30
Date Only	2024-03-15
Time Only	14:30:00
YYYY/MM/DD	2024/03/15
DD-Mon-YYYY	15-Mar-2024
Compact	20240315T143000Z
Fuzz Range Reference
1 min	60,000 ms		Micro-jitter for timestamp tests
1 hr	3,600,000 ms		Intra-day scheduling tests
1 day	86,400,000 ms		Daily boundary tests
7 days	604,800,000 ms		Weekly window fuzzing
30 days	2,592,000,000 ms		Monthly report testing
365 days	31,536,000,000 ms		Annual dataset generation

Frequently Asked Questions

The Box-Muller transform can theoretically produce values beyond ±3σ. Since σ = R3, roughly 0.3% of raw samples exceed the range. The tool clamps these outliers to ±R to guarantee all output dates stay within the configured window. This means the tails of the distribution are slightly heavier than a true Gaussian at the boundary points.

Yes. All arithmetic operates on Unix timestamps in milliseconds. The JavaScript Date constructor handles leap years (Feb 29) and month-length variations natively. However, DST transitions are locale-dependent. If you format output in local time, a fuzzed date crossing a DST boundary will show the correct wall-clock offset. UTC output (ISO 8601 with Z suffix) is unaffected by DST entirely.

Triangular distribution has finite support - it never produces values outside the range, and its density drops linearly from the center. Gaussian has infinite theoretical support (clamped here) and drops exponentially. For test data, triangular gives a softer center bias with a hard cutoff, while Gaussian produces tighter central clustering with rare near-boundary samples. Use triangular when you need guaranteed coverage across the range. Use Gaussian when modeling natural event timing (e.g., user login times that cluster around a peak hour).

This tool uses Math.random, which is not seedable in standard JavaScript. For reproducible test suites, copy the generated output and store it as a fixture. Alternatively, export the results and use them as static test data. True seedable PRNG (e.g., xoshiro128) would require a custom implementation; this tool prioritizes cryptographic-quality randomness from the browser engine over reproducibility.

Each sample requires date object creation, distribution sampling, formatting, and DOM rendering. At 10,000 samples, the total computation stays under 50ms on modern hardware. Beyond this, DOM rendering becomes the bottleneck - inserting 100,000 rows into the page causes layout thrashing and scroll jank. If you need larger datasets, generate in batches and use the copy-to-clipboard feature to aggregate results externally.

Edge-weighted (inverted triangular) concentrates samples near ±R boundaries while depopulating the center. This is specifically designed for boundary-value analysis in QA - testing date validators, range checks, and off-by-one errors at the extremes of an acceptance window. Uniform gives equal probability everywhere, which wastes test budget on uninteresting central values when boundary behavior is the concern.