I found and checked a collision for RIPEMD-160 truncated to its first 48 bits:

{"algo":"ripemd160","lambda":24,"inputA":"idmd-13458001","inputB":"idmd-21777228"}

• Digest of idmd-13458001: 9ff5cb7e7844b4e20774f357c3ce88d721e7832f
• Digest of idmd-21777228: 9ff5cb7e7844b94fb8896a518dd50687714b3a45

The two inputs are different and their first 12 hex digits (48 bits) match. I confirmed both digests separately with the OpenSSL command-line tool. The SIMD verifier itself was not available to me, so that check has not been run.

How it was found: a birthday search hashed the strings idmd-0, idmd-1, and so on, and kept every 48-bit prefix. The first repeat came after about 21.8 million hashes (about 2^24.4), close to the expected 2.1 million × 10. It took about 66 seconds.

One assumption: the inputs are plain text strings with no 0x prefix and no trailing newline. The digests above only hold if the verifier hashes the text exactly as written.

Files (all untracked, nothing committed):

• collision.json at the repo root, with a copy at artifacts/collision.json
• artifacts/report.md, which keeps facts, inferences and open questions separate
• artifacts/README.md, which states the question and its limits

The search script is in test/scratch/, which is excluded from submission. git status shows no other stray files.