Celeborn

Practical AI. Real results.

Containment Radar

Paste a passage and see whether it copies, contains, or is contained by anything in a small synthetic corpus. The engine is lexical fingerprinting: word shingles, a MinHash signature, locality-sensitive banding, then exact Jaccard and containment on the candidates. Everything runs in your browser. It is a similarity tool, not a chatbot.

Input

0 words · 0 shingles
Add snippet B for a pairwise compare
0 words · 0 shingles
Samples

Each sample is built from the synthetic corpus to show one kind of result.

Radar

Banding radar Outer ring: one segment per LSH band, lit when the band hash of snippet A matches the selected comparison. Inner ring: one tick per MinHash row, lit when the row values agree.

Comparing A with

Nothing yet

Jaccard (exact)
–
Jaccard (MinHash est.)
–
A inside target
–
Target inside A
–
Band collisions
–
Shared shingles
–
  • Band hash collision
  • MinHash row agrees
  • No match

Corpus hits

Ranked by the stronger of Jaccard and containment.

    Paste text or load a sample to compare against the corpus.

    How it works

    1. Shingle

      Text is lower-cased, punctuation is dropped, and every run of k consecutive words becomes a shingle. Two texts that share wording share shingles.

    2. MinHash

      Each shingle is hashed with 128 different hash functions; the minimum per function forms the signature. The fraction of matching positions estimates Jaccard similarity.

    3. Band

      The signature is cut into 32 bands of 4 rows. Two texts collide in a band when all four rows match. Collisions are the fast, index-friendly way to find candidates; here they light up the radar.

    4. Verify

      Candidates are checked with exact Jaccard (shared ÷ union) and containment (shared ÷ one side). High Jaccard with balanced containment is NEAR; high containment in one direction only is CONTAINS; the rest is CLEAR.