Practical guides, how-tos and comparisons to get the most out of our free online tools.
CPL is spend divided by leads, and alone it means almost nothing, because a lead is whatever you decided to call one. Worked here: a channel with twice the CPL producing half the CAC, and the redefinition that moves CPL fivefold without moving CAC at all.
Camille Laurent · 14 min read
Density counted occurrences because retrieval once counted occurrences. TF-IDF, then BM25 with its saturation curve, then embeddings replaced it. Here is the same 800-word page scored three ways, and why the three disagree.
Camille Laurent · 13 min read
Schema.org markup buys eligibility for a rich result, never a guarantee and never a ranking boost. Here is which types still produce something visible in 2026, what each one requires, and the one rule that gets sites penalised.
Camille Laurent · 15 min read
The two directions fail in opposite ways. PNG to JPG silently flattens transparency onto a background colour the converter chooses — we got white from one tool and black from another on the same file. JPG to PNG recovers nothing and cost up to 11.4 times the bytes.
Daniel Okonkwo · 12 min read
A visible mark lives in the pixels, so it survives every copy — and degrades the image for every honest viewer at the same time. Measured: what a corner mark costs, what a tiled mark costs, and how each compares to metadata credit, Content Credentials and invisible watermarking.
Daniel Okonkwo · 14 min read
Five files cover every consumer that exists: the tab, the bookmark, the pinned tile, the Android home screen and the iOS home screen. The twenty-file lists are a dead browser matrix — with measurements showing what a 16-pixel icon can and cannot hold.
Daniel Okonkwo · 13 min read
Unix time counts seconds since 1970 while deliberately ignoring leap seconds, which is why it is not a count of elapsed physical seconds. Here is what that trade buys, what a leap second does to the counter, the exact instant a signed 32-bit counter overflows, and the factor-of-1000 bug that lands you in 1970 or the year 58589.
Daniel Okonkwo · 17 min read
Minification has to preserve meaning, and the interesting part is the whitespace that carries meaning: the descendant combinator, the spaces inside calc(), the gap between two inline elements. Measured here on real files, including how much brotli was going to save you anyway.
Daniel Okonkwo · 16 min read
A checksum catches accidents. A cryptographic hash resists an attacker. A hash-table hash spreads keys. Three different jobs, three different families — and here is a CRC-32 collision constructed by hand in 0.11 seconds to show exactly why you cannot substitute one for another.
Daniel Okonkwo · 15 min read
The WCAG ratio is (L1 + 0.05) ÷ (L2 + 0.05), and L is relative luminance, not brightness. Green carries 71.52% of it and blue 7.22%, which is why pure blue on white passes at 8.59:1 while mid grey fails at 3.95:1. Here is the whole computation, run end to end.
Daniel Okonkwo · 15 min read
Crop factor is a ratio of sensor diagonals. It changes framing, not the lens — proved by computing field of view both ways — and it applies to the f-number as well as the focal length, which is the half most articles leave out.
Daniel Okonkwo · 11 min read
A stop is a factor of two in light. The f-number scale is powers of √2 for a reason you can derive in one line, EV = log2(N²/t) puts a number on it, and the three controls are equivalent in brightness only — never in the picture.
Daniel Okonkwo · 12 min read