Illustration — not a real scan
Contamination scan · 2026-08-07T09:00:00.000Z · ingot-0.1.5
Read the matches below before concluding anything. Canonical text with one natural phrasing looks identical to leakage in any count, and only the words tell you which this is.
| indexed | your writing, 3 pieces of writing at n=10 |
|---|---|
| corpus | c4-en (26 shards) · 9,264,249 documents · 3,418,685,530 tokens |
| matches kept | 1 |
| dropped as ordinary language | 0 |
| could not be checked | 0 pieces of writing |
| scan time | 3067.0s |
| near-duplicate tier | needs the text of what was indexed; this run carried hashes only |
Every match below is verbatim. Some overlaps are canonical facts with one natural phrasing rather than contamination, and the only way to tell is to read the text.
essay-2 → c4-train.00007 · doc 411,982
…Reprinted without permission: My grandmother kept her recipes on index cards in a tin that had once held shortbread. The handwriting changes across four decades, from a schoolgirl copperplate to something looser and more urgent. What a lovely piece.…
Nothing. Every piece of writing produced at least one 10-gram, so every one of them was genuinely examined.
scanner ingot-0.1.5, index format 3 index your writing · n=10 · stride 1 · 406 grams benchmark hash 89bee225be93a2c85b2597b7da32bada corpus c4-en (26 shards) · 21,330,000,000 bytes · 9,264,249 docs corpus hash fa98d595f8fdea47 corpus sha256 56e0965c421d4d642a18aeba27af3bb6c69f6a56d553bce96ec803261170bf2b generated 2026-08-07T09:00:00.000Z node scripts/build-index.ts reports/writer.jsonl --name "your writing" --item-noun "piece of writing|pieces of writing" --out reports/writer.idx.bin.gz && node src/cli.ts contaminate --index reports/writer.idx.bin.gz --corpus ../corpora/c4-en/ --out report.html
The corpus hash covers the first 64 lines, every thousandth line after that, and the exact line and byte counts. The full SHA-256 above covers every line. Neither stops someone who edits their own corpus to pass a scan.