Check an original word list with an explicit apostrophe policy

Identify the complete original

Each TXT line is a distinct word; a trailing delimiter leaves an empty word. JSON must contain one top-level string array. Duplicates, empty words, whitespace and original escaped lexemes remain. Numbers or booleans are never coerced into words. An initial UTF-8 BOM stays in the original and is excluded from the first word.

["hello","aren’t","aren't","  hello  ","zzz21stzzz","-0","","hello"]

• Select one complete UTF-8 file or pasted word list and separately choose TXT lines or a JSON string array for words and the optional whitelist.
• Select the active apostrophe policy. The whitelist matches original characters and case exactly; an empty string overrides only an exact empty word.

Select a reproducible policy

Native dictionary policy can convert curly apostrophes, so it is not literal equality. Original-apostrophe policy still trims and applies case rules. Settings record all three definitions and the selected policy; correct/incorrect alone does not explain what happened.

Audit each row and retain the whole package

The complete table has 26 columns. Ordinals, source positions and rawToken distinguish duplicates. Transformations and actual candidate attempts come from the same correct call. A whitelisted row has no dictionary call, so null and false mean different things. Complete output and independent readback of this example have been checked; UI and native Worker acceptance remain pending.

Limits, attribution and recovery

Words plus whitelist share 4 MiB UTF-8; words have at most 100,000 rows and the whitelist separately has at most 100,000 entries; each word has at most 128 Unicode codepoints.

The fixed AFF and DIC share 8 MiB; at most 100,000 word checks; loading, expansion, whitelist, lookup and serialization share 100,000,000 observable charged work units.

All 10/11 files plus independent complete copy text share 64 MiB. Held sources, dictionary, rows, serialization, clones, Blobs and UI representations share a conservative 768 MiB preallocation guard.

The absolute 60-second deadline starts at input capture and covers reading, module loading, dictionary initialization, checking, encoding, complete validation, cleanup and first publication; no pre-expanded dictionary bypass.

Regex VM internal steps, native container memory estimates, joint capacity, complete Hunspell equivalence, native Worker and cross-engine acceptance remain pending. The original private word list and environment were not published. This is not grammar certification.

Dictionary includes UKACD, SCOWL, WordNet, VarCon and Ispell notices. Read all licenses in the supplied documentation.

Check again with the same sources and settings

• Changing input or policy immediately invalidates old results; do not mix downloads from different runs.
• The original private word list was not published; this tool cannot infer correctness for that entire corpus.