I found a real collision: the strings imd-14354348 and imd-48319665 have SHA-256 digests that match in their first 50 bits. I checked it myself with sha256sum and Python's hashlib; the external verifier hasn't seen it yet.

{"algo":"sha256","lambda":25,"inputA":"imd-14354348","inputB":"imd-48319665"}

• The digests: they start 1bf95fda3510a8… and 1bf95fda3510b9…. The first 48 bits are identical. The next hex digits are a (1010) and b (1011), so bits 49–50 are both 10. The digests first differ at bit 51.
• How I found it: I hashed 2²⁶ strings (imd-0 to imd-67108863) and sorted the 50-bit prefixes to find matching pairs. About 2 matching pairs were expected and the run found 3. I checked the other two as well.
• Assumptions the result depends on:
  • "Truncated to 50 bits MSB" means the first 50 bits of the digest in normal order.
  • Inputs without a 0x prefix are hashed as plain text, with no trailing newline. Both strings are ASCII only.

The script and the intermediate files are in test/scratch/ only, so they won't be delivered. I didn't commit anything, and the output files are left untracked as instructed.

Files are in artifacts/ (a copy of collision.json is also at the repo root, since the task didn't say where it should go):

• collision.json
• report.md
• README.md