Datasets:

---

xet-team

/

xet-spec-reference-files

Like 2

Follow

Xet Team 288

Note : this report was AI generated.

The reference xorb file eea25d6ee393ccae385820daed127b96ef0ea034dfb7cf6da3a950ce334b7632.xorb ends with 4 extraneous bytes, so it is not a valid xorb upload body and is rejected by xet-core's own parsing code.

The defect

• Published file size: 14,737,817 bytes (sha256 e72f4e05b9c4b96f1c6b2213d93de00ab943585a2f7c049fd127ba72ec1ff413).
• The valid chunk-records region is the first 14,737,813 bytes; the last 4 bytes are 58 45 54 42 — ASCII "XETB", the first 4 bytes of the 7-byte storage-footer ident XETBLOB (XORB_OBJECT_FORMAT_IDENT in xet-core-structures/src/xorb_object/xorb_object_format.rs).

The spec defines the xorb upload body as a bare concatenation of chunk records with no trailer, and current clients upload exactly that (xet-data's file_upload_session.rs: "XORBs are sent without footer"). xet-core's own validator, reconstruct_xorb_with_footer, rejects this file verbatim: after the last chunk record, 4 bytes remain — too short for an 8-byte chunk header, and not the complete XETBLOB ident that would mark a genuine (tolerated) storage footer — so parsing errors out. Any server that validates uploads with the reference code therefore 400s the published vector as-is, even though it accepts real client traffic.

Evidence the 4 bytes are extraneous

Truncating the file to 14,737,813 bytes yields a body where, independently:

1. It parses as exactly 796 chunks, matching Electric_Vehicle_Population_Data_20250917.csv.chunks.
2. The recomputed xorb Merkle hash equals the filename hash eea25d6e…7632.
3. The total unpacked length is 63,527,244 — the reference CSV's exact size, and decompressing the chunks reproduces the CSV byte-identically.
4. The full-range verification hash matches the published ….xorb.range-hash vector.
5. The shard vectors' CAS-info section lists the same 796 chunk hashes.

Likely origin: xet-core ≤ v1.1.x serialized xorbs with the storage footer; this file looks like a footered serialization whose footer was trimmed during dataset generation, but cut 4 bytes too late.

Suggested fix

Re-publish the file truncated by 4 bytes:

head -c 14737813 eea25d6ee393ccae385820daed127b96ef0ea034dfb7cf6da3a950ce334b7632.xorb

Corrected sha256: b4670b1f13ab42a77490f53ce9167307ffbb9128f82ac82f721f85d2dce0a3a3.

All other files in the dataset (.chunk files, .csv.chunks, the shard variants, .range-hash, the dedup example) check out as internally consistent — this xorb is the only affected file.

Found while building an independent open-source server implementation of the Xet protocol and a conformance harness driven by these vectors.