Document the current behavior and add a changelog

[ESultanik]

Stacked on #24. Last one in the series.

The README still told people to pip3 install, said nothing about which Pythons are supported, and gave no guidance on the one thing that decides whether the tool works at all: whether your keys have enough entropy for your plaintexts.

README

New sections on choosing and testing keys with -t, on --seed being for testing rather than secrecy, on performance and where the time actually goes, on running the tests, and a table of the three ciphertext format versions. Notes that decryption now accepts plain as well as gzipped input, and - for stdin.

Every command in the README was run verbatim against this build, and every number measured on this machine rather than estimated. The ciphertext expansion figure is stated as a range across sizes — 5.5× at 512 B down to 2.7× at 512 KiB — because a single number would have been wrong at both ends: the dictionary header is a fixed cost that only amortizes with size. (My first draft said "around 2.7×", which measurement showed was only true for large inputs.)

CHANGELOG

Leads with the two defects that could lose or expose data, because those are the reason to care about this release rather than the interesting ones:

• --length-checksum corrupted plaintexts in 34 of 120 randomized round-trips, invisibly, because the length header truncated the damage away. Anyone holding ciphertexts from that mode should check them.
• Ciphertexts recorded the output filename in cleartext in the gzip header — for a deniability tool, recoverable metadata in the first 30 bytes.

It also documents the removed and renamed API. Encrypter's hooks changed shape rather than changing meaning under the same names, so an out-of-tree subclass fails loudly instead of silently doing something else.

American English

A second commit converts British spellings (behaviour, colour, labelled, amortised, summarise, honouring, desynchronisation, relabelled) to American across the README, changelog, source comments, docstrings, and affected test names. LICENSE is untouched — it is verbatim GPL text.

Note that commit messages and PR descriptions already merged still contain British spellings; those are in history and not worth rewriting.

Verified

• README usage block, -t example, and stdin example all run verbatim and produce the documented output
• Format-stability fixtures confirm all three versions still decrypt
• Index cost claim (0.25s / 62 MB for 2 × 512 KiB) re-measured at 0.26s / 63 MB
• lenticrypt.lenticrypt gone, from lenticrypt import ... intact, no namespace leaks, ENCRYPTION_VERSION 3 and __version__ 0.4.0 as documented
• 256 passed; ruff and ty clean

🤖 Generated with Claude Code

https://claude.ai/code/session_01367hFob9sd4xpDmVT4uoFy